about summary refs log tree commit diff
path: root/compat
diff options
context:
space:
mode:
Diffstat (limited to 'compat')
-rw-r--r--compat/access.c31
-rw-r--r--compat/apple-common-crypto.h96
-rw-r--r--compat/basename.c71
-rw-r--r--compat/bswap.h217
-rw-r--r--compat/fileno.c7
-rw-r--r--compat/fopen.c37
-rw-r--r--compat/gmtime.c29
-rw-r--r--compat/hstrerror.c21
-rw-r--r--compat/inet_ntop.c185
-rw-r--r--compat/inet_pton.c215
-rw-r--r--compat/memmem.c32
-rw-r--r--compat/mingw.c2543
-rw-r--r--compat/mingw.h583
-rw-r--r--compat/mkdir.c24
-rw-r--r--compat/mkdtemp.c8
-rw-r--r--compat/mmap.c40
-rw-r--r--compat/msvc.c6
-rw-r--r--compat/msvc.h33
-rw-r--r--compat/nedmalloc/License.txt23
-rw-r--r--compat/nedmalloc/Readme.txt136
-rw-r--r--compat/nedmalloc/malloc.c.h5761
-rw-r--r--compat/nedmalloc/nedmalloc.c953
-rw-r--r--compat/nedmalloc/nedmalloc.h180
-rw-r--r--compat/obstack.c413
-rw-r--r--compat/obstack.h509
-rw-r--r--compat/poll/poll.c633
-rw-r--r--compat/poll/poll.h67
-rw-r--r--compat/pread.c18
-rw-r--r--compat/precompose_utf8.c184
-rw-r--r--compat/precompose_utf8.h46
-rw-r--r--compat/qsort.c62
-rw-r--r--compat/qsort_s.c69
-rw-r--r--compat/regex/regcomp.c3891
-rw-r--r--compat/regex/regex.c87
-rw-r--r--compat/regex/regex.h581
-rw-r--r--compat/regex/regex_internal.c1743
-rw-r--r--compat/regex/regex_internal.h809
-rw-r--r--compat/regex/regexec.c4368
-rw-r--r--compat/setenv.c40
-rw-r--r--compat/sha1-chunked.c19
-rw-r--r--compat/sha1-chunked.h2
-rw-r--r--compat/snprintf.c69
-rw-r--r--compat/stat.c48
-rw-r--r--compat/strcasestr.c22
-rw-r--r--compat/strdup.c11
-rw-r--r--compat/strlcpy.c13
-rw-r--r--compat/strtoimax.c10
-rw-r--r--compat/strtoumax.c10
-rw-r--r--compat/terminal.c147
-rw-r--r--compat/terminal.h6
-rw-r--r--compat/unsetenv.c27
-rw-r--r--compat/vcbuild/.gitignore3
-rw-r--r--compat/vcbuild/README112
-rw-r--r--compat/vcbuild/find_vs_env.bat168
-rw-r--r--compat/vcbuild/include/sys/param.h1
-rw-r--r--compat/vcbuild/include/sys/time.h1
-rw-r--r--compat/vcbuild/include/sys/utime.h34
-rw-r--r--compat/vcbuild/include/unistd.h103
-rw-r--r--compat/vcbuild/include/utime.h1
-rwxr-xr-xcompat/vcbuild/scripts/clink.pl85
-rwxr-xr-xcompat/vcbuild/scripts/lib.pl26
-rw-r--r--compat/vcbuild/vcpkg_copy_dlls.bat39
-rw-r--r--compat/vcbuild/vcpkg_install.bat80
-rw-r--r--compat/win32.h41
-rw-r--r--compat/win32/alloca.h1
-rw-r--r--compat/win32/dirent.c92
-rw-r--r--compat/win32/dirent.h20
-rw-r--r--compat/win32/git.manifest25
-rw-r--r--compat/win32/lazyload.h57
-rw-r--r--compat/win32/path-utils.c28
-rw-r--r--compat/win32/path-utils.h20
-rw-r--r--compat/win32/pthread.c58
-rw-r--r--compat/win32/pthread.h100
-rw-r--r--compat/win32/syslog.c80
-rw-r--r--compat/win32/syslog.h20
-rw-r--r--compat/win32/trace2_win32_process_info.c191
-rw-r--r--compat/win32mmap.c46
-rw-r--r--compat/winansi.c671
78 files changed, 27238 insertions, 0 deletions
diff --git a/compat/access.c b/compat/access.c
new file mode 100644
index 000000000000..19fda3e87764
--- /dev/null
+++ b/compat/access.c
@@ -0,0 +1,31 @@
+#define COMPAT_CODE_ACCESS
+#include "../git-compat-util.h"
+
+/* Do the same thing access(2) does, but use the effective uid,
+ * and don't make the mistake of telling root that any file is
+ * executable.  This version uses stat(2).
+ */
+int git_access(const char *path, int mode)
+{
+	struct stat st;
+
+	/* do not interfere a normal user */
+	if (geteuid())
+		return access(path, mode);
+
+	if (stat(path, &st) < 0)
+		return -1;
+
+	/* Root can read or write any file. */
+	if (!(mode & X_OK))
+		return 0;
+
+	/* Root can execute any file that has any one of the execute
+	 * bits set.
+	 */
+	if (st.st_mode & (S_IXUSR | S_IXGRP | S_IXOTH))
+		return 0;
+
+	errno = EACCES;
+	return -1;
+}
diff --git a/compat/apple-common-crypto.h b/compat/apple-common-crypto.h
new file mode 100644
index 000000000000..11727f3e1ed7
--- /dev/null
+++ b/compat/apple-common-crypto.h
@@ -0,0 +1,96 @@
+/* suppress inclusion of conflicting openssl functions */
+#define OPENSSL_NO_MD5
+#define HEADER_HMAC_H
+#define HEADER_SHA_H
+#include <CommonCrypto/CommonHMAC.h>
+#define EVP_md5(...) kCCHmacAlgMD5
+/* CCHmac doesn't take md_len and the return type is void */
+#define HMAC git_CC_HMAC
+static inline unsigned char *git_CC_HMAC(CCHmacAlgorithm alg,
+		const void *key, int key_len,
+		const unsigned char *data, size_t data_len,
+		unsigned char *md, unsigned int *md_len)
+{
+	CCHmac(alg, key, key_len, data, data_len, md);
+	return md;
+}
+
+#if __MAC_OS_X_VERSION_MIN_REQUIRED >= 1070
+#define APPLE_LION_OR_NEWER
+#include <Security/Security.h>
+/* Apple's TYPE_BOOL conflicts with config.c */
+#undef TYPE_BOOL
+#endif
+
+#ifndef SHA1_MAX_BLOCK_SIZE
+#error Using Apple Common Crypto library requires setting SHA1_MAX_BLOCK_SIZE
+#endif
+
+#ifdef APPLE_LION_OR_NEWER
+#define git_CC_error_check(pattern, err) \
+	do { \
+		if (err) { \
+			die(pattern, (long)CFErrorGetCode(err)); \
+		} \
+	} while(0)
+
+#define EVP_EncodeBlock git_CC_EVP_EncodeBlock
+static inline int git_CC_EVP_EncodeBlock(unsigned char *out,
+		const unsigned char *in, int inlen)
+{
+	CFErrorRef err;
+	SecTransformRef encoder;
+	CFDataRef input, output;
+	CFIndex length;
+
+	encoder = SecEncodeTransformCreate(kSecBase64Encoding, &err);
+	git_CC_error_check("SecEncodeTransformCreate failed: %ld", err);
+
+	input = CFDataCreate(kCFAllocatorDefault, in, inlen);
+	SecTransformSetAttribute(encoder, kSecTransformInputAttributeName,
+			input, &err);
+	git_CC_error_check("SecTransformSetAttribute failed: %ld", err);
+
+	output = SecTransformExecute(encoder, &err);
+	git_CC_error_check("SecTransformExecute failed: %ld", err);
+
+	length = CFDataGetLength(output);
+	CFDataGetBytes(output, CFRangeMake(0, length), out);
+
+	CFRelease(output);
+	CFRelease(input);
+	CFRelease(encoder);
+
+	return (int)strlen((const char *)out);
+}
+
+#define EVP_DecodeBlock git_CC_EVP_DecodeBlock
+static int inline git_CC_EVP_DecodeBlock(unsigned char *out,
+		const unsigned char *in, int inlen)
+{
+	CFErrorRef err;
+	SecTransformRef decoder;
+	CFDataRef input, output;
+	CFIndex length;
+
+	decoder = SecDecodeTransformCreate(kSecBase64Encoding, &err);
+	git_CC_error_check("SecEncodeTransformCreate failed: %ld", err);
+
+	input = CFDataCreate(kCFAllocatorDefault, in, inlen);
+	SecTransformSetAttribute(decoder, kSecTransformInputAttributeName,
+			input, &err);
+	git_CC_error_check("SecTransformSetAttribute failed: %ld", err);
+
+	output = SecTransformExecute(decoder, &err);
+	git_CC_error_check("SecTransformExecute failed: %ld", err);
+
+	length = CFDataGetLength(output);
+	CFDataGetBytes(output, CFRangeMake(0, length), out);
+
+	CFRelease(output);
+	CFRelease(input);
+	CFRelease(decoder);
+
+	return (int)strlen((const char *)out);
+}
+#endif /* APPLE_LION_OR_NEWER */
diff --git a/compat/basename.c b/compat/basename.c
new file mode 100644
index 000000000000..96bd9533b448
--- /dev/null
+++ b/compat/basename.c
@@ -0,0 +1,71 @@
+#include "../git-compat-util.h"
+#include "../strbuf.h"
+
+/* Adapted from libiberty's basename.c.  */
+char *gitbasename (char *path)
+{
+	const char *base;
+
+	if (path)
+		skip_dos_drive_prefix(&path);
+
+	if (!path || !*path)
+		return ".";
+
+	for (base = path; *path; path++) {
+		if (!is_dir_sep(*path))
+			continue;
+		do {
+			path++;
+		} while (is_dir_sep(*path));
+		if (*path)
+			base = path;
+		else
+			while (--path != base && is_dir_sep(*path))
+				*path = '\0';
+	}
+	return (char *)base;
+}
+
+char *gitdirname(char *path)
+{
+	static struct strbuf buf = STRBUF_INIT;
+	char *p = path, *slash = NULL, c;
+	int dos_drive_prefix;
+
+	if (!p)
+		return ".";
+
+	if ((dos_drive_prefix = skip_dos_drive_prefix(&p)) && !*p)
+		goto dot;
+
+	/*
+	 * POSIX.1-2001 says dirname("/") should return "/", and dirname("//")
+	 * should return "//", but dirname("///") should return "/" again.
+	 */
+	if (is_dir_sep(*p)) {
+		if (!p[1] || (is_dir_sep(p[1]) && !p[2]))
+			return path;
+		slash = ++p;
+	}
+	while ((c = *(p++)))
+		if (is_dir_sep(c)) {
+			char *tentative = p - 1;
+
+			/* POSIX.1-2001 says to ignore trailing slashes */
+			while (is_dir_sep(*p))
+				p++;
+			if (*p)
+				slash = tentative;
+		}
+
+	if (slash) {
+		*slash = '\0';
+		return path;
+	}
+
+dot:
+	strbuf_reset(&buf);
+	strbuf_addf(&buf, "%.*s.", dos_drive_prefix, path);
+	return buf.buf;
+}
diff --git a/compat/bswap.h b/compat/bswap.h
new file mode 100644
index 000000000000..e4e25735ceb3
--- /dev/null
+++ b/compat/bswap.h
@@ -0,0 +1,217 @@
+#ifndef COMPAT_BSWAP_H
+#define COMPAT_BSWAP_H
+
+/*
+ * Let's make sure we always have a sane definition for ntohl()/htonl().
+ * Some libraries define those as a function call, just to perform byte
+ * shifting, bringing significant overhead to what should be a simple
+ * operation.
+ */
+
+/*
+ * Default version that the compiler ought to optimize properly with
+ * constant values.
+ */
+static inline uint32_t default_swab32(uint32_t val)
+{
+	return (((val & 0xff000000) >> 24) |
+		((val & 0x00ff0000) >>  8) |
+		((val & 0x0000ff00) <<  8) |
+		((val & 0x000000ff) << 24));
+}
+
+static inline uint64_t default_bswap64(uint64_t val)
+{
+	return (((val & (uint64_t)0x00000000000000ffULL) << 56) |
+		((val & (uint64_t)0x000000000000ff00ULL) << 40) |
+		((val & (uint64_t)0x0000000000ff0000ULL) << 24) |
+		((val & (uint64_t)0x00000000ff000000ULL) <<  8) |
+		((val & (uint64_t)0x000000ff00000000ULL) >>  8) |
+		((val & (uint64_t)0x0000ff0000000000ULL) >> 24) |
+		((val & (uint64_t)0x00ff000000000000ULL) >> 40) |
+		((val & (uint64_t)0xff00000000000000ULL) >> 56));
+}
+
+#undef bswap32
+#undef bswap64
+
+#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
+
+#define bswap32 git_bswap32
+static inline uint32_t git_bswap32(uint32_t x)
+{
+	uint32_t result;
+	if (__builtin_constant_p(x))
+		result = default_swab32(x);
+	else
+		__asm__("bswap %0" : "=r" (result) : "0" (x));
+	return result;
+}
+
+#define bswap64 git_bswap64
+#if defined(__x86_64__)
+static inline uint64_t git_bswap64(uint64_t x)
+{
+	uint64_t result;
+	if (__builtin_constant_p(x))
+		result = default_bswap64(x);
+	else
+		__asm__("bswap %q0" : "=r" (result) : "0" (x));
+	return result;
+}
+#else
+static inline uint64_t git_bswap64(uint64_t x)
+{
+	union { uint64_t i64; uint32_t i32[2]; } tmp, result;
+	if (__builtin_constant_p(x))
+		result.i64 = default_bswap64(x);
+	else {
+		tmp.i64 = x;
+		result.i32[0] = git_bswap32(tmp.i32[1]);
+		result.i32[1] = git_bswap32(tmp.i32[0]);
+	}
+	return result.i64;
+}
+#endif
+
+#elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
+
+#include <stdlib.h>
+
+#define bswap32(x) _byteswap_ulong(x)
+#define bswap64(x) _byteswap_uint64(x)
+
+#endif
+
+#if defined(bswap32)
+
+#undef ntohl
+#undef htonl
+#define ntohl(x) bswap32(x)
+#define htonl(x) bswap32(x)
+
+#endif
+
+#if defined(bswap64)
+
+#undef ntohll
+#undef htonll
+#define ntohll(x) bswap64(x)
+#define htonll(x) bswap64(x)
+
+#else
+
+#undef ntohll
+#undef htonll
+
+#if defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && defined(__BIG_ENDIAN)
+
+# define GIT_BYTE_ORDER __BYTE_ORDER
+# define GIT_LITTLE_ENDIAN __LITTLE_ENDIAN
+# define GIT_BIG_ENDIAN __BIG_ENDIAN
+
+#elif defined(BYTE_ORDER) && defined(LITTLE_ENDIAN) && defined(BIG_ENDIAN)
+
+# define GIT_BYTE_ORDER BYTE_ORDER
+# define GIT_LITTLE_ENDIAN LITTLE_ENDIAN
+# define GIT_BIG_ENDIAN BIG_ENDIAN
+
+#else
+
+# define GIT_BIG_ENDIAN 4321
+# define GIT_LITTLE_ENDIAN 1234
+
+# if defined(_BIG_ENDIAN) && !defined(_LITTLE_ENDIAN)
+#  define GIT_BYTE_ORDER GIT_BIG_ENDIAN
+# elif defined(_LITTLE_ENDIAN) && !defined(_BIG_ENDIAN)
+#  define GIT_BYTE_ORDER GIT_LITTLE_ENDIAN
+# elif defined(__THW_BIG_ENDIAN__) && !defined(__THW_LITTLE_ENDIAN__)
+#  define GIT_BYTE_ORDER GIT_BIG_ENDIAN
+# elif defined(__THW_LITTLE_ENDIAN__) && !defined(__THW_BIG_ENDIAN__)
+#  define GIT_BYTE_ORDER GIT_LITTLE_ENDIAN
+# else
+#  error "Cannot determine endianness"
+# endif
+
+#endif
+
+#if GIT_BYTE_ORDER == GIT_BIG_ENDIAN
+# define ntohll(n) (n)
+# define htonll(n) (n)
+#else
+# define ntohll(n) default_bswap64(n)
+# define htonll(n) default_bswap64(n)
+#endif
+
+#endif
+
+/*
+ * Performance might be improved if the CPU architecture is OK with
+ * unaligned 32-bit loads and a fast ntohl() is available.
+ * Otherwise fall back to byte loads and shifts which is portable,
+ * and is faster on architectures with memory alignment issues.
+ */
+
+#if !defined(NO_UNALIGNED_LOADS) && ( \
+    defined(__i386__) || defined(__x86_64__) || \
+    defined(_M_IX86) || defined(_M_X64) || \
+    defined(__ppc__) || defined(__ppc64__) || \
+    defined(__powerpc__) || defined(__powerpc64__) || \
+    defined(__s390__) || defined(__s390x__))
+
+#define get_be16(p)	ntohs(*(unsigned short *)(p))
+#define get_be32(p)	ntohl(*(unsigned int *)(p))
+#define get_be64(p)	ntohll(*(uint64_t *)(p))
+#define put_be32(p, v)	do { *(unsigned int *)(p) = htonl(v); } while (0)
+#define put_be64(p, v)	do { *(uint64_t *)(p) = htonll(v); } while (0)
+
+#else
+
+static inline uint16_t get_be16(const void *ptr)
+{
+	const unsigned char *p = ptr;
+	return	(uint16_t)p[0] << 8 |
+		(uint16_t)p[1] << 0;
+}
+
+static inline uint32_t get_be32(const void *ptr)
+{
+	const unsigned char *p = ptr;
+	return	(uint32_t)p[0] << 24 |
+		(uint32_t)p[1] << 16 |
+		(uint32_t)p[2] <<  8 |
+		(uint32_t)p[3] <<  0;
+}
+
+static inline uint64_t get_be64(const void *ptr)
+{
+	const unsigned char *p = ptr;
+	return	(uint64_t)get_be32(&p[0]) << 32 |
+		(uint64_t)get_be32(&p[4]) <<  0;
+}
+
+static inline void put_be32(void *ptr, uint32_t value)
+{
+	unsigned char *p = ptr;
+	p[0] = value >> 24;
+	p[1] = value >> 16;
+	p[2] = value >>  8;
+	p[3] = value >>  0;
+}
+
+static inline void put_be64(void *ptr, uint64_t value)
+{
+	unsigned char *p = ptr;
+	p[0] = value >> 56;
+	p[1] = value >> 48;
+	p[2] = value >> 40;
+	p[3] = value >> 32;
+	p[4] = value >> 24;
+	p[5] = value >> 16;
+	p[6] = value >>  8;
+	p[7] = value >>  0;
+}
+
+#endif
+
+#endif /* COMPAT_BSWAP_H */
diff --git a/compat/fileno.c b/compat/fileno.c
new file mode 100644
index 000000000000..8e80ef335d84
--- /dev/null
+++ b/compat/fileno.c
@@ -0,0 +1,7 @@
+#define COMPAT_CODE_FILENO
+#include "../git-compat-util.h"
+
+int git_fileno(FILE *stream)
+{
+	return fileno(stream);
+}
diff --git a/compat/fopen.c b/compat/fopen.c
new file mode 100644
index 000000000000..107b3e8182fd
--- /dev/null
+++ b/compat/fopen.c
@@ -0,0 +1,37 @@
+/*
+ *  The order of the following two lines is important.
+ *
+ *  SUPPRESS_FOPEN_REDEFINITION is defined before including git-compat-util.h
+ *  to avoid the redefinition of fopen within git-compat-util.h. This is
+ *  necessary since fopen is a macro on some platforms which may be set
+ *  based on compiler options. For example, on AIX fopen is set to fopen64
+ *  when _LARGE_FILES is defined. The previous technique of merely undefining
+ *  fopen after including git-compat-util.h is inadequate in this case.
+ */
+#define SUPPRESS_FOPEN_REDEFINITION
+#include "../git-compat-util.h"
+
+FILE *git_fopen(const char *path, const char *mode)
+{
+	FILE *fp;
+	struct stat st;
+
+	if (mode[0] == 'w' || mode[0] == 'a')
+		return fopen(path, mode);
+
+	if (!(fp = fopen(path, mode)))
+		return NULL;
+
+	if (fstat(fileno(fp), &st)) {
+		fclose(fp);
+		return NULL;
+	}
+
+	if (S_ISDIR(st.st_mode)) {
+		fclose(fp);
+		errno = EISDIR;
+		return NULL;
+	}
+
+	return fp;
+}
diff --git a/compat/gmtime.c b/compat/gmtime.c
new file mode 100644
index 000000000000..e8362dd2b965
--- /dev/null
+++ b/compat/gmtime.c
@@ -0,0 +1,29 @@
+#include "../git-compat-util.h"
+#undef gmtime
+#undef gmtime_r
+
+struct tm *git_gmtime(const time_t *timep)
+{
+	static struct tm result;
+	return git_gmtime_r(timep, &result);
+}
+
+struct tm *git_gmtime_r(const time_t *timep, struct tm *result)
+{
+	struct tm *ret;
+
+	memset(result, 0, sizeof(*result));
+	ret = gmtime_r(timep, result);
+
+	/*
+	 * Rather than NULL, FreeBSD gmtime simply leaves the "struct tm"
+	 * untouched when it encounters overflow. Since "mday" cannot otherwise
+	 * be zero, we can test this very quickly.
+	 */
+	if (ret && !ret->tm_mday) {
+		ret = NULL;
+		errno = EOVERFLOW;
+	}
+
+	return ret;
+}
diff --git a/compat/hstrerror.c b/compat/hstrerror.c
new file mode 100644
index 000000000000..b85a2fa9561f
--- /dev/null
+++ b/compat/hstrerror.c
@@ -0,0 +1,21 @@
+#include <string.h>
+#include <stdio.h>
+#include <netdb.h>
+
+const char *githstrerror(int err)
+{
+	static char buffer[48];
+	switch (err)
+	{
+	case HOST_NOT_FOUND:
+		return "Authoritative answer: host not found";
+	case NO_DATA:
+		return "Valid name, no data record of requested type";
+	case NO_RECOVERY:
+		return "Non recoverable errors, FORMERR, REFUSED, NOTIMP";
+	case TRY_AGAIN:
+		return "Non-authoritative \"host not found\", or SERVERFAIL";
+	}
+	snprintf(buffer, sizeof(buffer), "Name resolution error %d", err);
+	return buffer;
+}
diff --git a/compat/inet_ntop.c b/compat/inet_ntop.c
new file mode 100644
index 000000000000..68307262be04
--- /dev/null
+++ b/compat/inet_ntop.c
@@ -0,0 +1,185 @@
+/*
+ * Copyright (c) 1996-1999 by Internet Software Consortium.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM DISCLAIMS
+ * ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL INTERNET SOFTWARE
+ * CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
+ * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
+ * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
+ * ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
+ * SOFTWARE.
+ */
+
+#include "../git-compat-util.h"
+
+#ifndef NS_INADDRSZ
+#define NS_INADDRSZ	4
+#endif
+#ifndef NS_IN6ADDRSZ
+#define NS_IN6ADDRSZ	16
+#endif
+#ifndef NS_INT16SZ
+#define NS_INT16SZ	2
+#endif
+
+/*
+ * WARNING: Don't even consider trying to compile this on a system where
+ * sizeof(int) < 4.  sizeof(int) > 4 is fine; all the world's not a VAX.
+ */
+
+/* const char *
+ * inet_ntop4(src, dst, size)
+ *	format an IPv4 address
+ * return:
+ *	`dst' (as a const)
+ * notes:
+ *	(1) uses no statics
+ *	(2) takes a u_char* not an in_addr as input
+ * author:
+ *	Paul Vixie, 1996.
+ */
+static const char *
+inet_ntop4(const u_char *src, char *dst, size_t size)
+{
+	static const char fmt[] = "%u.%u.%u.%u";
+	char tmp[sizeof "255.255.255.255"];
+	int nprinted;
+
+	nprinted = snprintf(tmp, sizeof(tmp), fmt, src[0], src[1], src[2], src[3]);
+	if (nprinted < 0)
+		return (NULL);	/* we assume "errno" was set by "snprintf()" */
+	if ((size_t)nprinted >= size) {
+		errno = ENOSPC;
+		return (NULL);
+	}
+	strlcpy(dst, tmp, size);
+	return (dst);
+}
+
+#ifndef NO_IPV6
+/* const char *
+ * inet_ntop6(src, dst, size)
+ *	convert IPv6 binary address into presentation (printable) format
+ * author:
+ *	Paul Vixie, 1996.
+ */
+static const char *
+inet_ntop6(const u_char *src, char *dst, size_t size)
+{
+	/*
+	 * Note that int32_t and int16_t need only be "at least" large enough
+	 * to contain a value of the specified size.  On some systems, like
+	 * Crays, there is no such thing as an integer variable with 16 bits.
+	 * Keep this in mind if you think this function should have been coded
+	 * to use pointer overlays.  All the world's not a VAX.
+	 */
+	char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"], *tp;
+	struct { int base, len; } best, cur;
+	unsigned int words[NS_IN6ADDRSZ / NS_INT16SZ];
+	int i;
+
+	/*
+	 * Preprocess:
+	 *	Copy the input (bytewise) array into a wordwise array.
+	 *	Find the longest run of 0x00's in src[] for :: shorthanding.
+	 */
+	memset(words, '\0', sizeof words);
+	for (i = 0; i < NS_IN6ADDRSZ; i++)
+		words[i / 2] |= (src[i] << ((1 - (i % 2)) << 3));
+	best.base = -1;
+	best.len = 0;
+	cur.base = -1;
+	cur.len = 0;
+	for (i = 0; i < (NS_IN6ADDRSZ / NS_INT16SZ); i++) {
+		if (words[i] == 0) {
+			if (cur.base == -1)
+				cur.base = i, cur.len = 1;
+			else
+				cur.len++;
+		} else {
+			if (cur.base != -1) {
+				if (best.base == -1 || cur.len > best.len)
+					best = cur;
+				cur.base = -1;
+			}
+		}
+	}
+	if (cur.base != -1) {
+		if (best.base == -1 || cur.len > best.len)
+			best = cur;
+	}
+	if (best.base != -1 && best.len < 2)
+		best.base = -1;
+
+	/*
+	 * Format the result.
+	 */
+	tp = tmp;
+	for (i = 0; i < (NS_IN6ADDRSZ / NS_INT16SZ); i++) {
+		/* Are we inside the best run of 0x00's? */
+		if (best.base != -1 && i >= best.base &&
+		    i < (best.base + best.len)) {
+			if (i == best.base)
+				*tp++ = ':';
+			continue;
+		}
+		/* Are we following an initial run of 0x00s or any real hex? */
+		if (i != 0)
+			*tp++ = ':';
+		/* Is this address an encapsulated IPv4? */
+		if (i == 6 && best.base == 0 &&
+		    (best.len == 6 || (best.len == 5 && words[5] == 0xffff))) {
+			if (!inet_ntop4(src+12, tp, sizeof tmp - (tp - tmp)))
+				return (NULL);
+			tp += strlen(tp);
+			break;
+		}
+		tp += snprintf(tp, sizeof tmp - (tp - tmp), "%x", words[i]);
+	}
+	/* Was it a trailing run of 0x00's? */
+	if (best.base != -1 && (best.base + best.len) ==
+	    (NS_IN6ADDRSZ / NS_INT16SZ))
+		*tp++ = ':';
+	*tp++ = '\0';
+
+	/*
+	 * Check for overflow, copy, and we're done.
+	 */
+	if ((size_t)(tp - tmp) > size) {
+		errno = ENOSPC;
+		return (NULL);
+	}
+	strlcpy(dst, tmp, size);
+	return (dst);
+}
+#endif
+
+/* char *
+ * inet_ntop(af, src, dst, size)
+ *	convert a network format address to presentation format.
+ * return:
+ *	pointer to presentation format address (`dst'), or NULL (see errno).
+ * author:
+ *	Paul Vixie, 1996.
+ */
+const char *
+inet_ntop(int af, const void *src, char *dst, size_t size)
+{
+	switch (af) {
+	case AF_INET:
+		return (inet_ntop4(src, dst, size));
+#ifndef NO_IPV6
+	case AF_INET6:
+		return (inet_ntop6(src, dst, size));
+#endif
+	default:
+		errno = EAFNOSUPPORT;
+		return (NULL);
+	}
+	/* NOTREACHED */
+}
diff --git a/compat/inet_pton.c b/compat/inet_pton.c
new file mode 100644
index 000000000000..2b9a0a4e22fa
--- /dev/null
+++ b/compat/inet_pton.c
@@ -0,0 +1,215 @@
+/*
+ * Copyright (C) 1996-2001  Internet Software Consortium.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM
+ * DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL
+ * INTERNET SOFTWARE CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING
+ * FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT,
+ * NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION
+ * WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "../git-compat-util.h"
+
+#ifndef NS_INT16SZ
+#define NS_INT16SZ       2
+#endif
+
+#ifndef NS_INADDRSZ
+#define NS_INADDRSZ      4
+#endif
+
+#ifndef NS_IN6ADDRSZ
+#define NS_IN6ADDRSZ    16
+#endif
+
+/*
+ * WARNING: Don't even consider trying to compile this on a system where
+ * sizeof(int) < 4.  sizeof(int) > 4 is fine; all the world's not a VAX.
+ */
+
+static int inet_pton4(const char *src, unsigned char *dst);
+#ifndef NO_IPV6
+static int inet_pton6(const char *src, unsigned char *dst);
+#endif
+
+/* int
+ * inet_pton4(src, dst)
+ *      like inet_aton() but without all the hexadecimal and shorthand.
+ * return:
+ *      1 if `src' is a valid dotted quad, else 0.
+ * notice:
+ *      does not touch `dst' unless it's returning 1.
+ * author:
+ *      Paul Vixie, 1996.
+ */
+static int
+inet_pton4(const char *src, unsigned char *dst)
+{
+        static const char digits[] = "0123456789";
+        int saw_digit, octets, ch;
+        unsigned char tmp[NS_INADDRSZ], *tp;
+
+        saw_digit = 0;
+        octets = 0;
+        *(tp = tmp) = 0;
+        while ((ch = *src++) != '\0') {
+                const char *pch;
+
+                if ((pch = strchr(digits, ch)) != NULL) {
+                        unsigned int new = *tp * 10 + (pch - digits);
+
+                        if (new > 255)
+                                return (0);
+                        *tp = new;
+                        if (! saw_digit) {
+                                if (++octets > 4)
+                                        return (0);
+                                saw_digit = 1;
+                        }
+                } else if (ch == '.' && saw_digit) {
+                        if (octets == 4)
+                                return (0);
+                        *++tp = 0;
+                        saw_digit = 0;
+                } else
+                        return (0);
+        }
+        if (octets < 4)
+                return (0);
+        memcpy(dst, tmp, NS_INADDRSZ);
+        return (1);
+}
+
+/* int
+ * inet_pton6(src, dst)
+ *      convert presentation level address to network order binary form.
+ * return:
+ *      1 if `src' is a valid [RFC1884 2.2] address, else 0.
+ * notice:
+ *      (1) does not touch `dst' unless it's returning 1.
+ *      (2) :: in a full address is silently ignored.
+ * credit:
+ *      inspired by Mark Andrews.
+ * author:
+ *      Paul Vixie, 1996.
+ */
+
+#ifndef NO_IPV6
+static int
+inet_pton6(const char *src, unsigned char *dst)
+{
+        static const char xdigits_l[] = "0123456789abcdef",
+                          xdigits_u[] = "0123456789ABCDEF";
+        unsigned char tmp[NS_IN6ADDRSZ], *tp, *endp, *colonp;
+        const char *xdigits, *curtok;
+        int ch, saw_xdigit;
+        unsigned int val;
+
+        memset((tp = tmp), '\0', NS_IN6ADDRSZ);
+        endp = tp + NS_IN6ADDRSZ;
+        colonp = NULL;
+        /* Leading :: requires some special handling. */
+        if (*src == ':')
+                if (*++src != ':')
+                        return (0);
+        curtok = src;
+        saw_xdigit = 0;
+        val = 0;
+        while ((ch = *src++) != '\0') {
+                const char *pch;
+
+                if ((pch = strchr((xdigits = xdigits_l), ch)) == NULL)
+                        pch = strchr((xdigits = xdigits_u), ch);
+                if (pch != NULL) {
+                        val <<= 4;
+                        val |= (pch - xdigits);
+                        if (val > 0xffff)
+                                return (0);
+                        saw_xdigit = 1;
+                        continue;
+                }
+                if (ch == ':') {
+                        curtok = src;
+                        if (!saw_xdigit) {
+                                if (colonp)
+                                        return (0);
+                                colonp = tp;
+                                continue;
+                        }
+                        if (tp + NS_INT16SZ > endp)
+                                return (0);
+                        *tp++ = (unsigned char) (val >> 8) & 0xff;
+                        *tp++ = (unsigned char) val & 0xff;
+                        saw_xdigit = 0;
+                        val = 0;
+                        continue;
+                }
+                if (ch == '.' && ((tp + NS_INADDRSZ) <= endp) &&
+                    inet_pton4(curtok, tp) > 0) {
+                        tp += NS_INADDRSZ;
+                        saw_xdigit = 0;
+                        break;  /* '\0' was seen by inet_pton4(). */
+                }
+                return (0);
+        }
+        if (saw_xdigit) {
+                if (tp + NS_INT16SZ > endp)
+                        return (0);
+                *tp++ = (unsigned char) (val >> 8) & 0xff;
+                *tp++ = (unsigned char) val & 0xff;
+        }
+        if (colonp != NULL) {
+                /*
+                 * Since some memmove()'s erroneously fail to handle
+                 * overlapping regions, we'll do the shift by hand.
+                 */
+                const int n = tp - colonp;
+                int i;
+
+                for (i = 1; i <= n; i++) {
+                        endp[- i] = colonp[n - i];
+                        colonp[n - i] = 0;
+                }
+                tp = endp;
+        }
+        if (tp != endp)
+                return (0);
+        memcpy(dst, tmp, NS_IN6ADDRSZ);
+        return (1);
+}
+#endif
+
+/* int
+ * isc_net_pton(af, src, dst)
+ *      convert from presentation format (which usually means ASCII printable)
+ *      to network format (which is usually some kind of binary format).
+ * return:
+ *      1 if the address was valid for the specified address family
+ *      0 if the address wasn't valid (`dst' is untouched in this case)
+ *      -1 if some other error occurred (`dst' is untouched in this case, too)
+ * author:
+ *      Paul Vixie, 1996.
+ */
+int
+inet_pton(int af, const char *src, void *dst)
+{
+        switch (af) {
+        case AF_INET:
+                return (inet_pton4(src, dst));
+#ifndef NO_IPV6
+        case AF_INET6:
+                return (inet_pton6(src, dst));
+#endif
+        default:
+                errno = EAFNOSUPPORT;
+                return (-1);
+        }
+        /* NOTREACHED */
+}
diff --git a/compat/memmem.c b/compat/memmem.c
new file mode 100644
index 000000000000..56bcb4277f47
--- /dev/null
+++ b/compat/memmem.c
@@ -0,0 +1,32 @@
+#include "../git-compat-util.h"
+
+void *gitmemmem(const void *haystack, size_t haystack_len,
+                const void *needle, size_t needle_len)
+{
+	const char *begin = haystack;
+	const char *last_possible = begin + haystack_len - needle_len;
+	const char *tail = needle;
+	char point;
+
+	/*
+	 * The first occurrence of the empty string is deemed to occur at
+	 * the beginning of the string.
+	 */
+	if (needle_len == 0)
+		return (void *)begin;
+
+	/*
+	 * Sanity check, otherwise the loop might search through the whole
+	 * memory.
+	 */
+	if (haystack_len < needle_len)
+		return NULL;
+
+	point = *tail++;
+	for (; begin <= last_possible; begin++) {
+		if (*begin == point && !memcmp(begin + 1, tail, needle_len - 1))
+			return (void *)begin;
+	}
+
+	return NULL;
+}
diff --git a/compat/mingw.c b/compat/mingw.c
new file mode 100644
index 000000000000..738f0a826a51
--- /dev/null
+++ b/compat/mingw.c
@@ -0,0 +1,2543 @@
+#include "../git-compat-util.h"
+#include "win32.h"
+#include <conio.h>
+#include <wchar.h>
+#include "../strbuf.h"
+#include "../run-command.h"
+#include "../cache.h"
+#include "win32/lazyload.h"
+#include "../config.h"
+#include "dir.h"
+
+#define HCAST(type, handle) ((type)(intptr_t)handle)
+
+static const int delay[] = { 0, 1, 10, 20, 40 };
+
+int err_win_to_posix(DWORD winerr)
+{
+	int error = ENOSYS;
+	switch(winerr) {
+	case ERROR_ACCESS_DENIED: error = EACCES; break;
+	case ERROR_ACCOUNT_DISABLED: error = EACCES; break;
+	case ERROR_ACCOUNT_RESTRICTION: error = EACCES; break;
+	case ERROR_ALREADY_ASSIGNED: error = EBUSY; break;
+	case ERROR_ALREADY_EXISTS: error = EEXIST; break;
+	case ERROR_ARITHMETIC_OVERFLOW: error = ERANGE; break;
+	case ERROR_BAD_COMMAND: error = EIO; break;
+	case ERROR_BAD_DEVICE: error = ENODEV; break;
+	case ERROR_BAD_DRIVER_LEVEL: error = ENXIO; break;
+	case ERROR_BAD_EXE_FORMAT: error = ENOEXEC; break;
+	case ERROR_BAD_FORMAT: error = ENOEXEC; break;
+	case ERROR_BAD_LENGTH: error = EINVAL; break;
+	case ERROR_BAD_PATHNAME: error = ENOENT; break;
+	case ERROR_BAD_PIPE: error = EPIPE; break;
+	case ERROR_BAD_UNIT: error = ENODEV; break;
+	case ERROR_BAD_USERNAME: error = EINVAL; break;
+	case ERROR_BROKEN_PIPE: error = EPIPE; break;
+	case ERROR_BUFFER_OVERFLOW: error = ENAMETOOLONG; break;
+	case ERROR_BUSY: error = EBUSY; break;
+	case ERROR_BUSY_DRIVE: error = EBUSY; break;
+	case ERROR_CALL_NOT_IMPLEMENTED: error = ENOSYS; break;
+	case ERROR_CANNOT_MAKE: error = EACCES; break;
+	case ERROR_CANTOPEN: error = EIO; break;
+	case ERROR_CANTREAD: error = EIO; break;
+	case ERROR_CANTWRITE: error = EIO; break;
+	case ERROR_CRC: error = EIO; break;
+	case ERROR_CURRENT_DIRECTORY: error = EACCES; break;
+	case ERROR_DEVICE_IN_USE: error = EBUSY; break;
+	case ERROR_DEV_NOT_EXIST: error = ENODEV; break;
+	case ERROR_DIRECTORY: error = EINVAL; break;
+	case ERROR_DIR_NOT_EMPTY: error = ENOTEMPTY; break;
+	case ERROR_DISK_CHANGE: error = EIO; break;
+	case ERROR_DISK_FULL: error = ENOSPC; break;
+	case ERROR_DRIVE_LOCKED: error = EBUSY; break;
+	case ERROR_ENVVAR_NOT_FOUND: error = EINVAL; break;
+	case ERROR_EXE_MARKED_INVALID: error = ENOEXEC; break;
+	case ERROR_FILENAME_EXCED_RANGE: error = ENAMETOOLONG; break;
+	case ERROR_FILE_EXISTS: error = EEXIST; break;
+	case ERROR_FILE_INVALID: error = ENODEV; break;
+	case ERROR_FILE_NOT_FOUND: error = ENOENT; break;
+	case ERROR_GEN_FAILURE: error = EIO; break;
+	case ERROR_HANDLE_DISK_FULL: error = ENOSPC; break;
+	case ERROR_INSUFFICIENT_BUFFER: error = ENOMEM; break;
+	case ERROR_INVALID_ACCESS: error = EACCES; break;
+	case ERROR_INVALID_ADDRESS: error = EFAULT; break;
+	case ERROR_INVALID_BLOCK: error = EFAULT; break;
+	case ERROR_INVALID_DATA: error = EINVAL; break;
+	case ERROR_INVALID_DRIVE: error = ENODEV; break;
+	case ERROR_INVALID_EXE_SIGNATURE: error = ENOEXEC; break;
+	case ERROR_INVALID_FLAGS: error = EINVAL; break;
+	case ERROR_INVALID_FUNCTION: error = ENOSYS; break;
+	case ERROR_INVALID_HANDLE: error = EBADF; break;
+	case ERROR_INVALID_LOGON_HOURS: error = EACCES; break;
+	case ERROR_INVALID_NAME: error = EINVAL; break;
+	case ERROR_INVALID_OWNER: error = EINVAL; break;
+	case ERROR_INVALID_PARAMETER: error = EINVAL; break;
+	case ERROR_INVALID_PASSWORD: error = EPERM; break;
+	case ERROR_INVALID_PRIMARY_GROUP: error = EINVAL; break;
+	case ERROR_INVALID_SIGNAL_NUMBER: error = EINVAL; break;
+	case ERROR_INVALID_TARGET_HANDLE: error = EIO; break;
+	case ERROR_INVALID_WORKSTATION: error = EACCES; break;
+	case ERROR_IO_DEVICE: error = EIO; break;
+	case ERROR_IO_INCOMPLETE: error = EINTR; break;
+	case ERROR_LOCKED: error = EBUSY; break;
+	case ERROR_LOCK_VIOLATION: error = EACCES; break;
+	case ERROR_LOGON_FAILURE: error = EACCES; break;
+	case ERROR_MAPPED_ALIGNMENT: error = EINVAL; break;
+	case ERROR_META_EXPANSION_TOO_LONG: error = E2BIG; break;
+	case ERROR_MORE_DATA: error = EPIPE; break;
+	case ERROR_NEGATIVE_SEEK: error = ESPIPE; break;
+	case ERROR_NOACCESS: error = EFAULT; break;
+	case ERROR_NONE_MAPPED: error = EINVAL; break;
+	case ERROR_NOT_ENOUGH_MEMORY: error = ENOMEM; break;
+	case ERROR_NOT_READY: error = EAGAIN; break;
+	case ERROR_NOT_SAME_DEVICE: error = EXDEV; break;
+	case ERROR_NO_DATA: error = EPIPE; break;
+	case ERROR_NO_MORE_SEARCH_HANDLES: error = EIO; break;
+	case ERROR_NO_PROC_SLOTS: error = EAGAIN; break;
+	case ERROR_NO_SUCH_PRIVILEGE: error = EACCES; break;
+	case ERROR_OPEN_FAILED: error = EIO; break;
+	case ERROR_OPEN_FILES: error = EBUSY; break;
+	case ERROR_OPERATION_ABORTED: error = EINTR; break;
+	case ERROR_OUTOFMEMORY: error = ENOMEM; break;
+	case ERROR_PASSWORD_EXPIRED: error = EACCES; break;
+	case ERROR_PATH_BUSY: error = EBUSY; break;
+	case ERROR_PATH_NOT_FOUND: error = ENOENT; break;
+	case ERROR_PIPE_BUSY: error = EBUSY; break;
+	case ERROR_PIPE_CONNECTED: error = EPIPE; break;
+	case ERROR_PIPE_LISTENING: error = EPIPE; break;
+	case ERROR_PIPE_NOT_CONNECTED: error = EPIPE; break;
+	case ERROR_PRIVILEGE_NOT_HELD: error = EACCES; break;
+	case ERROR_READ_FAULT: error = EIO; break;
+	case ERROR_SEEK: error = EIO; break;
+	case ERROR_SEEK_ON_DEVICE: error = ESPIPE; break;
+	case ERROR_SHARING_BUFFER_EXCEEDED: error = ENFILE; break;
+	case ERROR_SHARING_VIOLATION: error = EACCES; break;
+	case ERROR_STACK_OVERFLOW: error = ENOMEM; break;
+	case ERROR_SWAPERROR: error = ENOENT; break;
+	case ERROR_TOO_MANY_MODULES: error = EMFILE; break;
+	case ERROR_TOO_MANY_OPEN_FILES: error = EMFILE; break;
+	case ERROR_UNRECOGNIZED_MEDIA: error = ENXIO; break;
+	case ERROR_UNRECOGNIZED_VOLUME: error = ENODEV; break;
+	case ERROR_WAIT_NO_CHILDREN: error = ECHILD; break;
+	case ERROR_WRITE_FAULT: error = EIO; break;
+	case ERROR_WRITE_PROTECT: error = EROFS; break;
+	}
+	return error;
+}
+
+static inline int is_file_in_use_error(DWORD errcode)
+{
+	switch (errcode) {
+	case ERROR_SHARING_VIOLATION:
+	case ERROR_ACCESS_DENIED:
+		return 1;
+	}
+
+	return 0;
+}
+
+static int read_yes_no_answer(void)
+{
+	char answer[1024];
+
+	if (fgets(answer, sizeof(answer), stdin)) {
+		size_t answer_len = strlen(answer);
+		int got_full_line = 0, c;
+
+		/* remove the newline */
+		if (answer_len >= 2 && answer[answer_len-2] == '\r') {
+			answer[answer_len-2] = '\0';
+			got_full_line = 1;
+		} else if (answer_len >= 1 && answer[answer_len-1] == '\n') {
+			answer[answer_len-1] = '\0';
+			got_full_line = 1;
+		}
+		/* flush the buffer in case we did not get the full line */
+		if (!got_full_line)
+			while ((c = getchar()) != EOF && c != '\n')
+				;
+	} else
+		/* we could not read, return the
+		 * default answer which is no */
+		return 0;
+
+	if (tolower(answer[0]) == 'y' && !answer[1])
+		return 1;
+	if (!strncasecmp(answer, "yes", sizeof(answer)))
+		return 1;
+	if (tolower(answer[0]) == 'n' && !answer[1])
+		return 0;
+	if (!strncasecmp(answer, "no", sizeof(answer)))
+		return 0;
+
+	/* did not find an answer we understand */
+	return -1;
+}
+
+static int ask_yes_no_if_possible(const char *format, ...)
+{
+	char question[4096];
+	const char *retry_hook[] = { NULL, NULL, NULL };
+	va_list args;
+
+	va_start(args, format);
+	vsnprintf(question, sizeof(question), format, args);
+	va_end(args);
+
+	if ((retry_hook[0] = mingw_getenv("GIT_ASK_YESNO"))) {
+		retry_hook[1] = question;
+		return !run_command_v_opt(retry_hook, 0);
+	}
+
+	if (!isatty(_fileno(stdin)) || !isatty(_fileno(stderr)))
+		return 0;
+
+	while (1) {
+		int answer;
+		fprintf(stderr, "%s (y/n) ", question);
+
+		if ((answer = read_yes_no_answer()) >= 0)
+			return answer;
+
+		fprintf(stderr, "Sorry, I did not understand your answer. "
+				"Please type 'y' or 'n'\n");
+	}
+}
+
+/* Windows only */
+enum hide_dotfiles_type {
+	HIDE_DOTFILES_FALSE = 0,
+	HIDE_DOTFILES_TRUE,
+	HIDE_DOTFILES_DOTGITONLY
+};
+
+static enum hide_dotfiles_type hide_dotfiles = HIDE_DOTFILES_DOTGITONLY;
+static char *unset_environment_variables;
+
+int mingw_core_config(const char *var, const char *value, void *cb)
+{
+	if (!strcmp(var, "core.hidedotfiles")) {
+		if (value && !strcasecmp(value, "dotgitonly"))
+			hide_dotfiles = HIDE_DOTFILES_DOTGITONLY;
+		else
+			hide_dotfiles = git_config_bool(var, value);
+		return 0;
+	}
+
+	if (!strcmp(var, "core.unsetenvvars")) {
+		free(unset_environment_variables);
+		unset_environment_variables = xstrdup(value);
+		return 0;
+	}
+
+	return 0;
+}
+
+/* Normalizes NT paths as returned by some low-level APIs. */
+static wchar_t *normalize_ntpath(wchar_t *wbuf)
+{
+	int i;
+	/* fix absolute path prefixes */
+	if (wbuf[0] == '\\') {
+		/* strip NT namespace prefixes */
+		if (!wcsncmp(wbuf, L"\\??\\", 4) ||
+		    !wcsncmp(wbuf, L"\\\\?\\", 4))
+			wbuf += 4;
+		else if (!wcsnicmp(wbuf, L"\\DosDevices\\", 12))
+			wbuf += 12;
+		/* replace remaining '...UNC\' with '\\' */
+		if (!wcsnicmp(wbuf, L"UNC\\", 4)) {
+			wbuf += 2;
+			*wbuf = '\\';
+		}
+	}
+	/* convert backslashes to slashes */
+	for (i = 0; wbuf[i]; i++)
+		if (wbuf[i] == '\\')
+			wbuf[i] = '/';
+	return wbuf;
+}
+
+int mingw_unlink(const char *pathname)
+{
+	int ret, tries = 0;
+	wchar_t wpathname[MAX_PATH];
+	if (xutftowcs_path(wpathname, pathname) < 0)
+		return -1;
+
+	/* read-only files cannot be removed */
+	_wchmod(wpathname, 0666);
+	while ((ret = _wunlink(wpathname)) == -1 && tries < ARRAY_SIZE(delay)) {
+		if (!is_file_in_use_error(GetLastError()))
+			break;
+		/*
+		 * We assume that some other process had the source or
+		 * destination file open at the wrong moment and retry.
+		 * In order to give the other process a higher chance to
+		 * complete its operation, we give up our time slice now.
+		 * If we have to retry again, we do sleep a bit.
+		 */
+		Sleep(delay[tries]);
+		tries++;
+	}
+	while (ret == -1 && is_file_in_use_error(GetLastError()) &&
+	       ask_yes_no_if_possible("Unlink of file '%s' failed. "
+			"Should I try again?", pathname))
+	       ret = _wunlink(wpathname);
+	return ret;
+}
+
+static int is_dir_empty(const wchar_t *wpath)
+{
+	WIN32_FIND_DATAW findbuf;
+	HANDLE handle;
+	wchar_t wbuf[MAX_PATH + 2];
+	wcscpy(wbuf, wpath);
+	wcscat(wbuf, L"\\*");
+	handle = FindFirstFileW(wbuf, &findbuf);
+	if (handle == INVALID_HANDLE_VALUE)
+		return GetLastError() == ERROR_NO_MORE_FILES;
+
+	while (!wcscmp(findbuf.cFileName, L".") ||
+			!wcscmp(findbuf.cFileName, L".."))
+		if (!FindNextFileW(handle, &findbuf)) {
+			DWORD err = GetLastError();
+			FindClose(handle);
+			return err == ERROR_NO_MORE_FILES;
+		}
+	FindClose(handle);
+	return 0;
+}
+
+int mingw_rmdir(const char *pathname)
+{
+	int ret, tries = 0;
+	wchar_t wpathname[MAX_PATH];
+	if (xutftowcs_path(wpathname, pathname) < 0)
+		return -1;
+
+	while ((ret = _wrmdir(wpathname)) == -1 && tries < ARRAY_SIZE(delay)) {
+		if (!is_file_in_use_error(GetLastError()))
+			errno = err_win_to_posix(GetLastError());
+		if (errno != EACCES)
+			break;
+		if (!is_dir_empty(wpathname)) {
+			errno = ENOTEMPTY;
+			break;
+		}
+		/*
+		 * We assume that some other process had the source or
+		 * destination file open at the wrong moment and retry.
+		 * In order to give the other process a higher chance to
+		 * complete its operation, we give up our time slice now.
+		 * If we have to retry again, we do sleep a bit.
+		 */
+		Sleep(delay[tries]);
+		tries++;
+	}
+	while (ret == -1 && errno == EACCES && is_file_in_use_error(GetLastError()) &&
+	       ask_yes_no_if_possible("Deletion of directory '%s' failed. "
+			"Should I try again?", pathname))
+	       ret = _wrmdir(wpathname);
+	return ret;
+}
+
+static inline int needs_hiding(const char *path)
+{
+	const char *basename;
+
+	if (hide_dotfiles == HIDE_DOTFILES_FALSE)
+		return 0;
+
+	/* We cannot use basename(), as it would remove trailing slashes */
+	win32_skip_dos_drive_prefix((char **)&path);
+	if (!*path)
+		return 0;
+
+	for (basename = path; *path; path++)
+		if (is_dir_sep(*path)) {
+			do {
+				path++;
+			} while (is_dir_sep(*path));
+			/* ignore trailing slashes */
+			if (*path)
+				basename = path;
+		}
+
+	if (hide_dotfiles == HIDE_DOTFILES_TRUE)
+		return *basename == '.';
+
+	assert(hide_dotfiles == HIDE_DOTFILES_DOTGITONLY);
+	return !strncasecmp(".git", basename, 4) &&
+		(!basename[4] || is_dir_sep(basename[4]));
+}
+
+static int set_hidden_flag(const wchar_t *path, int set)
+{
+	DWORD original = GetFileAttributesW(path), modified;
+	if (set)
+		modified = original | FILE_ATTRIBUTE_HIDDEN;
+	else
+		modified = original & ~FILE_ATTRIBUTE_HIDDEN;
+	if (original == modified || SetFileAttributesW(path, modified))
+		return 0;
+	errno = err_win_to_posix(GetLastError());
+	return -1;
+}
+
+int mingw_mkdir(const char *path, int mode)
+{
+	int ret;
+	wchar_t wpath[MAX_PATH];
+	if (xutftowcs_path(wpath, path) < 0)
+		return -1;
+	ret = _wmkdir(wpath);
+	if (!ret && needs_hiding(path))
+		return set_hidden_flag(wpath, 1);
+	return ret;
+}
+
+/*
+ * Calling CreateFile() using FILE_APPEND_DATA and without FILE_WRITE_DATA
+ * is documented in [1] as opening a writable file handle in append mode.
+ * (It is believed that) this is atomic since it is maintained by the
+ * kernel unlike the O_APPEND flag which is racily maintained by the CRT.
+ *
+ * [1] https://docs.microsoft.com/en-us/windows/desktop/fileio/file-access-rights-constants
+ *
+ * This trick does not appear to work for named pipes.  Instead it creates
+ * a named pipe client handle that cannot be written to.  Callers should
+ * just use the regular _wopen() for them.  (And since client handle gets
+ * bound to a unique server handle, it isn't really an issue.)
+ */
+static int mingw_open_append(wchar_t const *wfilename, int oflags, ...)
+{
+	HANDLE handle;
+	int fd;
+	DWORD create = (oflags & O_CREAT) ? OPEN_ALWAYS : OPEN_EXISTING;
+
+	/* only these flags are supported */
+	if ((oflags & ~O_CREAT) != (O_WRONLY | O_APPEND))
+		return errno = ENOSYS, -1;
+
+	/*
+	 * FILE_SHARE_WRITE is required to permit child processes
+	 * to append to the file.
+	 */
+	handle = CreateFileW(wfilename, FILE_APPEND_DATA,
+			FILE_SHARE_WRITE | FILE_SHARE_READ,
+			NULL, create, FILE_ATTRIBUTE_NORMAL, NULL);
+	if (handle == INVALID_HANDLE_VALUE)
+		return errno = err_win_to_posix(GetLastError()), -1;
+
+	/*
+	 * No O_APPEND here, because the CRT uses it only to reset the
+	 * file pointer to EOF before each write(); but that is not
+	 * necessary (and may lead to races) for a file created with
+	 * FILE_APPEND_DATA.
+	 */
+	fd = _open_osfhandle((intptr_t)handle, O_BINARY);
+	if (fd < 0)
+		CloseHandle(handle);
+	return fd;
+}
+
+/*
+ * Does the pathname map to the local named pipe filesystem?
+ * That is, does it have a "//./pipe/" prefix?
+ */
+static int is_local_named_pipe_path(const char *filename)
+{
+	return (is_dir_sep(filename[0]) &&
+		is_dir_sep(filename[1]) &&
+		filename[2] == '.'  &&
+		is_dir_sep(filename[3]) &&
+		!strncasecmp(filename+4, "pipe", 4) &&
+		is_dir_sep(filename[8]) &&
+		filename[9]);
+}
+
+int mingw_open (const char *filename, int oflags, ...)
+{
+	typedef int (*open_fn_t)(wchar_t const *wfilename, int oflags, ...);
+	va_list args;
+	unsigned mode;
+	int fd;
+	wchar_t wfilename[MAX_PATH];
+	open_fn_t open_fn;
+
+	va_start(args, oflags);
+	mode = va_arg(args, int);
+	va_end(args);
+
+	if (filename && !strcmp(filename, "/dev/null"))
+		filename = "nul";
+
+	if ((oflags & O_APPEND) && !is_local_named_pipe_path(filename))
+		open_fn = mingw_open_append;
+	else
+		open_fn = _wopen;
+
+	if (xutftowcs_path(wfilename, filename) < 0)
+		return -1;
+	fd = open_fn(wfilename, oflags, mode);
+
+	if (fd < 0 && (oflags & O_ACCMODE) != O_RDONLY && errno == EACCES) {
+		DWORD attrs = GetFileAttributesW(wfilename);
+		if (attrs != INVALID_FILE_ATTRIBUTES && (attrs & FILE_ATTRIBUTE_DIRECTORY))
+			errno = EISDIR;
+	}
+	if ((oflags & O_CREAT) && needs_hiding(filename)) {
+		/*
+		 * Internally, _wopen() uses the CreateFile() API which errors
+		 * out with an ERROR_ACCESS_DENIED if CREATE_ALWAYS was
+		 * specified and an already existing file's attributes do not
+		 * match *exactly*. As there is no mode or flag we can set that
+		 * would correspond to FILE_ATTRIBUTE_HIDDEN, let's just try
+		 * again *without* the O_CREAT flag (that corresponds to the
+		 * CREATE_ALWAYS flag of CreateFile()).
+		 */
+		if (fd < 0 && errno == EACCES)
+			fd = open_fn(wfilename, oflags & ~O_CREAT, mode);
+		if (fd >= 0 && set_hidden_flag(wfilename, 1))
+			warning("could not mark '%s' as hidden.", filename);
+	}
+	return fd;
+}
+
+static BOOL WINAPI ctrl_ignore(DWORD type)
+{
+	return TRUE;
+}
+
+#undef fgetc
+int mingw_fgetc(FILE *stream)
+{
+	int ch;
+	if (!isatty(_fileno(stream)))
+		return fgetc(stream);
+
+	SetConsoleCtrlHandler(ctrl_ignore, TRUE);
+	while (1) {
+		ch = fgetc(stream);
+		if (ch != EOF || GetLastError() != ERROR_OPERATION_ABORTED)
+			break;
+
+		/* Ctrl+C was pressed, simulate SIGINT and retry */
+		mingw_raise(SIGINT);
+	}
+	SetConsoleCtrlHandler(ctrl_ignore, FALSE);
+	return ch;
+}
+
+#undef fopen
+FILE *mingw_fopen (const char *filename, const char *otype)
+{
+	int hide = needs_hiding(filename);
+	FILE *file;
+	wchar_t wfilename[MAX_PATH], wotype[4];
+	if (filename && !strcmp(filename, "/dev/null"))
+		filename = "nul";
+	if (xutftowcs_path(wfilename, filename) < 0 ||
+		xutftowcs(wotype, otype, ARRAY_SIZE(wotype)) < 0)
+		return NULL;
+	if (hide && !access(filename, F_OK) && set_hidden_flag(wfilename, 0)) {
+		error("could not unhide %s", filename);
+		return NULL;
+	}
+	file = _wfopen(wfilename, wotype);
+	if (!file && GetLastError() == ERROR_INVALID_NAME)
+		errno = ENOENT;
+	if (file && hide && set_hidden_flag(wfilename, 1))
+		warning("could not mark '%s' as hidden.", filename);
+	return file;
+}
+
+FILE *mingw_freopen (const char *filename, const char *otype, FILE *stream)
+{
+	int hide = needs_hiding(filename);
+	FILE *file;
+	wchar_t wfilename[MAX_PATH], wotype[4];
+	if (filename && !strcmp(filename, "/dev/null"))
+		filename = "nul";
+	if (xutftowcs_path(wfilename, filename) < 0 ||
+		xutftowcs(wotype, otype, ARRAY_SIZE(wotype)) < 0)
+		return NULL;
+	if (hide && !access(filename, F_OK) && set_hidden_flag(wfilename, 0)) {
+		error("could not unhide %s", filename);
+		return NULL;
+	}
+	file = _wfreopen(wfilename, wotype, stream);
+	if (file && hide && set_hidden_flag(wfilename, 1))
+		warning("could not mark '%s' as hidden.", filename);
+	return file;
+}
+
+#undef fflush
+int mingw_fflush(FILE *stream)
+{
+	int ret = fflush(stream);
+
+	/*
+	 * write() is used behind the scenes of stdio output functions.
+	 * Since git code does not check for errors after each stdio write
+	 * operation, it can happen that write() is called by a later
+	 * stdio function even if an earlier write() call failed. In the
+	 * case of a pipe whose readable end was closed, only the first
+	 * call to write() reports EPIPE on Windows. Subsequent write()
+	 * calls report EINVAL. It is impossible to notice whether this
+	 * fflush invocation triggered such a case, therefore, we have to
+	 * catch all EINVAL errors whole-sale.
+	 */
+	if (ret && errno == EINVAL)
+		errno = EPIPE;
+
+	return ret;
+}
+
+#undef write
+ssize_t mingw_write(int fd, const void *buf, size_t len)
+{
+	ssize_t result = write(fd, buf, len);
+
+	if (result < 0 && errno == EINVAL && buf) {
+		/* check if fd is a pipe */
+		HANDLE h = (HANDLE) _get_osfhandle(fd);
+		if (GetFileType(h) == FILE_TYPE_PIPE)
+			errno = EPIPE;
+		else
+			errno = EINVAL;
+	}
+
+	return result;
+}
+
+int mingw_access(const char *filename, int mode)
+{
+	wchar_t wfilename[MAX_PATH];
+	if (xutftowcs_path(wfilename, filename) < 0)
+		return -1;
+	/* X_OK is not supported by the MSVCRT version */
+	return _waccess(wfilename, mode & ~X_OK);
+}
+
+int mingw_chdir(const char *dirname)
+{
+	wchar_t wdirname[MAX_PATH];
+	if (xutftowcs_path(wdirname, dirname) < 0)
+		return -1;
+	return _wchdir(wdirname);
+}
+
+int mingw_chmod(const char *filename, int mode)
+{
+	wchar_t wfilename[MAX_PATH];
+	if (xutftowcs_path(wfilename, filename) < 0)
+		return -1;
+	return _wchmod(wfilename, mode);
+}
+
+/*
+ * The unit of FILETIME is 100-nanoseconds since January 1, 1601, UTC.
+ * Returns the 100-nanoseconds ("hekto nanoseconds") since the epoch.
+ */
+static inline long long filetime_to_hnsec(const FILETIME *ft)
+{
+	long long winTime = ((long long)ft->dwHighDateTime << 32) + ft->dwLowDateTime;
+	/* Windows to Unix Epoch conversion */
+	return winTime - 116444736000000000LL;
+}
+
+static inline void filetime_to_timespec(const FILETIME *ft, struct timespec *ts)
+{
+	long long hnsec = filetime_to_hnsec(ft);
+	ts->tv_sec = (time_t)(hnsec / 10000000);
+	ts->tv_nsec = (hnsec % 10000000) * 100;
+}
+
+/**
+ * Verifies that safe_create_leading_directories() would succeed.
+ */
+static int has_valid_directory_prefix(wchar_t *wfilename)
+{
+	int n = wcslen(wfilename);
+
+	while (n > 0) {
+		wchar_t c = wfilename[--n];
+		DWORD attributes;
+
+		if (!is_dir_sep(c))
+			continue;
+
+		wfilename[n] = L'\0';
+		attributes = GetFileAttributesW(wfilename);
+		wfilename[n] = c;
+		if (attributes == FILE_ATTRIBUTE_DIRECTORY ||
+				attributes == FILE_ATTRIBUTE_DEVICE)
+			return 1;
+		if (attributes == INVALID_FILE_ATTRIBUTES)
+			switch (GetLastError()) {
+			case ERROR_PATH_NOT_FOUND:
+				continue;
+			case ERROR_FILE_NOT_FOUND:
+				/* This implies parent directory exists. */
+				return 1;
+			}
+		return 0;
+	}
+	return 1;
+}
+
+/* We keep the do_lstat code in a separate function to avoid recursion.
+ * When a path ends with a slash, the stat will fail with ENOENT. In
+ * this case, we strip the trailing slashes and stat again.
+ *
+ * If follow is true then act like stat() and report on the link
+ * target. Otherwise report on the link itself.
+ */
+static int do_lstat(int follow, const char *file_name, struct stat *buf)
+{
+	WIN32_FILE_ATTRIBUTE_DATA fdata;
+	wchar_t wfilename[MAX_PATH];
+	if (xutftowcs_path(wfilename, file_name) < 0)
+		return -1;
+
+	if (GetFileAttributesExW(wfilename, GetFileExInfoStandard, &fdata)) {
+		buf->st_ino = 0;
+		buf->st_gid = 0;
+		buf->st_uid = 0;
+		buf->st_nlink = 1;
+		buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes);
+		buf->st_size = fdata.nFileSizeLow |
+			(((off_t)fdata.nFileSizeHigh)<<32);
+		buf->st_dev = buf->st_rdev = 0; /* not used by Git */
+		filetime_to_timespec(&(fdata.ftLastAccessTime), &(buf->st_atim));
+		filetime_to_timespec(&(fdata.ftLastWriteTime), &(buf->st_mtim));
+		filetime_to_timespec(&(fdata.ftCreationTime), &(buf->st_ctim));
+		if (fdata.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) {
+			WIN32_FIND_DATAW findbuf;
+			HANDLE handle = FindFirstFileW(wfilename, &findbuf);
+			if (handle != INVALID_HANDLE_VALUE) {
+				if ((findbuf.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
+						(findbuf.dwReserved0 == IO_REPARSE_TAG_SYMLINK)) {
+					if (follow) {
+						char buffer[MAXIMUM_REPARSE_DATA_BUFFER_SIZE];
+						buf->st_size = readlink(file_name, buffer, MAXIMUM_REPARSE_DATA_BUFFER_SIZE);
+					} else {
+						buf->st_mode = S_IFLNK;
+					}
+					buf->st_mode |= S_IREAD;
+					if (!(findbuf.dwFileAttributes & FILE_ATTRIBUTE_READONLY))
+						buf->st_mode |= S_IWRITE;
+				}
+				FindClose(handle);
+			}
+		}
+		return 0;
+	}
+	switch (GetLastError()) {
+	case ERROR_ACCESS_DENIED:
+	case ERROR_SHARING_VIOLATION:
+	case ERROR_LOCK_VIOLATION:
+	case ERROR_SHARING_BUFFER_EXCEEDED:
+		errno = EACCES;
+		break;
+	case ERROR_BUFFER_OVERFLOW:
+		errno = ENAMETOOLONG;
+		break;
+	case ERROR_NOT_ENOUGH_MEMORY:
+		errno = ENOMEM;
+		break;
+	case ERROR_PATH_NOT_FOUND:
+		if (!has_valid_directory_prefix(wfilename)) {
+			errno = ENOTDIR;
+			break;
+		}
+		/* fallthru */
+	default:
+		errno = ENOENT;
+		break;
+	}
+	return -1;
+}
+
+/* We provide our own lstat/fstat functions, since the provided
+ * lstat/fstat functions are so slow. These stat functions are
+ * tailored for Git's usage (read: fast), and are not meant to be
+ * complete. Note that Git stat()s are redirected to mingw_lstat()
+ * too, since Windows doesn't really handle symlinks that well.
+ */
+static int do_stat_internal(int follow, const char *file_name, struct stat *buf)
+{
+	int namelen;
+	char alt_name[PATH_MAX];
+
+	if (!do_lstat(follow, file_name, buf))
+		return 0;
+
+	/* if file_name ended in a '/', Windows returned ENOENT;
+	 * try again without trailing slashes
+	 */
+	if (errno != ENOENT)
+		return -1;
+
+	namelen = strlen(file_name);
+	if (namelen && file_name[namelen-1] != '/')
+		return -1;
+	while (namelen && file_name[namelen-1] == '/')
+		--namelen;
+	if (!namelen || namelen >= PATH_MAX)
+		return -1;
+
+	memcpy(alt_name, file_name, namelen);
+	alt_name[namelen] = 0;
+	return do_lstat(follow, alt_name, buf);
+}
+
+static int get_file_info_by_handle(HANDLE hnd, struct stat *buf)
+{
+	BY_HANDLE_FILE_INFORMATION fdata;
+
+	if (!GetFileInformationByHandle(hnd, &fdata)) {
+		errno = err_win_to_posix(GetLastError());
+		return -1;
+	}
+
+	buf->st_ino = 0;
+	buf->st_gid = 0;
+	buf->st_uid = 0;
+	buf->st_nlink = 1;
+	buf->st_mode = file_attr_to_st_mode(fdata.dwFileAttributes);
+	buf->st_size = fdata.nFileSizeLow |
+		(((off_t)fdata.nFileSizeHigh)<<32);
+	buf->st_dev = buf->st_rdev = 0; /* not used by Git */
+	filetime_to_timespec(&(fdata.ftLastAccessTime), &(buf->st_atim));
+	filetime_to_timespec(&(fdata.ftLastWriteTime), &(buf->st_mtim));
+	filetime_to_timespec(&(fdata.ftCreationTime), &(buf->st_ctim));
+	return 0;
+}
+
+int mingw_lstat(const char *file_name, struct stat *buf)
+{
+	return do_stat_internal(0, file_name, buf);
+}
+int mingw_stat(const char *file_name, struct stat *buf)
+{
+	return do_stat_internal(1, file_name, buf);
+}
+
+int mingw_fstat(int fd, struct stat *buf)
+{
+	HANDLE fh = (HANDLE)_get_osfhandle(fd);
+	DWORD avail, type = GetFileType(fh) & ~FILE_TYPE_REMOTE;
+
+	switch (type) {
+	case FILE_TYPE_DISK:
+		return get_file_info_by_handle(fh, buf);
+
+	case FILE_TYPE_CHAR:
+	case FILE_TYPE_PIPE:
+		/* initialize stat fields */
+		memset(buf, 0, sizeof(*buf));
+		buf->st_nlink = 1;
+
+		if (type == FILE_TYPE_CHAR) {
+			buf->st_mode = _S_IFCHR;
+		} else {
+			buf->st_mode = _S_IFIFO;
+			if (PeekNamedPipe(fh, NULL, 0, NULL, &avail, NULL))
+				buf->st_size = avail;
+		}
+		return 0;
+
+	default:
+		errno = EBADF;
+		return -1;
+	}
+}
+
+static inline void time_t_to_filetime(time_t t, FILETIME *ft)
+{
+	long long winTime = t * 10000000LL + 116444736000000000LL;
+	ft->dwLowDateTime = winTime;
+	ft->dwHighDateTime = winTime >> 32;
+}
+
+int mingw_utime (const char *file_name, const struct utimbuf *times)
+{
+	FILETIME mft, aft;
+	int fh, rc;
+	DWORD attrs;
+	wchar_t wfilename[MAX_PATH];
+	if (xutftowcs_path(wfilename, file_name) < 0)
+		return -1;
+
+	/* must have write permission */
+	attrs = GetFileAttributesW(wfilename);
+	if (attrs != INVALID_FILE_ATTRIBUTES &&
+	    (attrs & FILE_ATTRIBUTE_READONLY)) {
+		/* ignore errors here; open() will report them */
+		SetFileAttributesW(wfilename, attrs & ~FILE_ATTRIBUTE_READONLY);
+	}
+
+	if ((fh = _wopen(wfilename, O_RDWR | O_BINARY)) < 0) {
+		rc = -1;
+		goto revert_attrs;
+	}
+
+	if (times) {
+		time_t_to_filetime(times->modtime, &mft);
+		time_t_to_filetime(times->actime, &aft);
+	} else {
+		GetSystemTimeAsFileTime(&mft);
+		aft = mft;
+	}
+	if (!SetFileTime((HANDLE)_get_osfhandle(fh), NULL, &aft, &mft)) {
+		errno = EINVAL;
+		rc = -1;
+	} else
+		rc = 0;
+	close(fh);
+
+revert_attrs:
+	if (attrs != INVALID_FILE_ATTRIBUTES &&
+	    (attrs & FILE_ATTRIBUTE_READONLY)) {
+		/* ignore errors again */
+		SetFileAttributesW(wfilename, attrs);
+	}
+	return rc;
+}
+
+#undef strftime
+size_t mingw_strftime(char *s, size_t max,
+		      const char *format, const struct tm *tm)
+{
+	size_t ret = strftime(s, max, format, tm);
+
+	if (!ret && errno == EINVAL)
+		die("invalid strftime format: '%s'", format);
+	return ret;
+}
+
+unsigned int sleep (unsigned int seconds)
+{
+	Sleep(seconds*1000);
+	return 0;
+}
+
+char *mingw_mktemp(char *template)
+{
+	wchar_t wtemplate[MAX_PATH];
+	if (xutftowcs_path(wtemplate, template) < 0)
+		return NULL;
+	if (!_wmktemp(wtemplate))
+		return NULL;
+	if (xwcstoutf(template, wtemplate, strlen(template) + 1) < 0)
+		return NULL;
+	return template;
+}
+
+int mkstemp(char *template)
+{
+	char *filename = mktemp(template);
+	if (filename == NULL)
+		return -1;
+	return open(filename, O_RDWR | O_CREAT, 0600);
+}
+
+int gettimeofday(struct timeval *tv, void *tz)
+{
+	FILETIME ft;
+	long long hnsec;
+
+	GetSystemTimeAsFileTime(&ft);
+	hnsec = filetime_to_hnsec(&ft);
+	tv->tv_sec = hnsec / 10000000;
+	tv->tv_usec = (hnsec % 10000000) / 10;
+	return 0;
+}
+
+int pipe(int filedes[2])
+{
+	HANDLE h[2];
+
+	/* this creates non-inheritable handles */
+	if (!CreatePipe(&h[0], &h[1], NULL, 8192)) {
+		errno = err_win_to_posix(GetLastError());
+		return -1;
+	}
+	filedes[0] = _open_osfhandle(HCAST(int, h[0]), O_NOINHERIT);
+	if (filedes[0] < 0) {
+		CloseHandle(h[0]);
+		CloseHandle(h[1]);
+		return -1;
+	}
+	filedes[1] = _open_osfhandle(HCAST(int, h[1]), O_NOINHERIT);
+	if (filedes[1] < 0) {
+		close(filedes[0]);
+		CloseHandle(h[1]);
+		return -1;
+	}
+	return 0;
+}
+
+struct tm *gmtime_r(const time_t *timep, struct tm *result)
+{
+	/* gmtime() in MSVCRT.DLL is thread-safe, but not reentrant */
+	memcpy(result, gmtime(timep), sizeof(struct tm));
+	return result;
+}
+
+struct tm *localtime_r(const time_t *timep, struct tm *result)
+{
+	/* localtime() in MSVCRT.DLL is thread-safe, but not reentrant */
+	memcpy(result, localtime(timep), sizeof(struct tm));
+	return result;
+}
+
+char *mingw_getcwd(char *pointer, int len)
+{
+	wchar_t cwd[MAX_PATH], wpointer[MAX_PATH];
+	DWORD ret = GetCurrentDirectoryW(ARRAY_SIZE(cwd), cwd);
+
+	if (!ret || ret >= ARRAY_SIZE(cwd)) {
+		errno = ret ? ENAMETOOLONG : err_win_to_posix(GetLastError());
+		return NULL;
+	}
+	ret = GetLongPathNameW(cwd, wpointer, ARRAY_SIZE(wpointer));
+	if (!ret && GetLastError() == ERROR_ACCESS_DENIED) {
+		HANDLE hnd = CreateFileW(cwd, 0,
+			FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL,
+			OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
+		if (hnd == INVALID_HANDLE_VALUE)
+			return NULL;
+		ret = GetFinalPathNameByHandleW(hnd, wpointer, ARRAY_SIZE(wpointer), 0);
+		CloseHandle(hnd);
+		if (!ret || ret >= ARRAY_SIZE(wpointer))
+			return NULL;
+		if (xwcstoutf(pointer, normalize_ntpath(wpointer), len) < 0)
+			return NULL;
+		return pointer;
+	}
+	if (!ret || ret >= ARRAY_SIZE(wpointer))
+		return NULL;
+	if (xwcstoutf(pointer, wpointer, len) < 0)
+		return NULL;
+	convert_slashes(pointer);
+	return pointer;
+}
+
+/*
+ * See "Parsing C++ Command-Line Arguments" at Microsoft's Docs:
+ * https://docs.microsoft.com/en-us/cpp/cpp/parsing-cpp-command-line-arguments
+ */
+static const char *quote_arg_msvc(const char *arg)
+{
+	/* count chars to quote */
+	int len = 0, n = 0;
+	int force_quotes = 0;
+	char *q, *d;
+	const char *p = arg;
+	if (!*p) force_quotes = 1;
+	while (*p) {
+		if (isspace(*p) || *p == '*' || *p == '?' || *p == '{' || *p == '\'')
+			force_quotes = 1;
+		else if (*p == '"')
+			n++;
+		else if (*p == '\\') {
+			int count = 0;
+			while (*p == '\\') {
+				count++;
+				p++;
+				len++;
+			}
+			if (*p == '"')
+				n += count*2 + 1;
+			continue;
+		}
+		len++;
+		p++;
+	}
+	if (!force_quotes && n == 0)
+		return arg;
+
+	/* insert \ where necessary */
+	d = q = xmalloc(st_add3(len, n, 3));
+	*d++ = '"';
+	while (*arg) {
+		if (*arg == '"')
+			*d++ = '\\';
+		else if (*arg == '\\') {
+			int count = 0;
+			while (*arg == '\\') {
+				count++;
+				*d++ = *arg++;
+			}
+			if (*arg == '"') {
+				while (count-- > 0)
+					*d++ = '\\';
+				*d++ = '\\';
+			}
+		}
+		*d++ = *arg++;
+	}
+	*d++ = '"';
+	*d++ = 0;
+	return q;
+}
+
+#include "quote.h"
+
+static const char *quote_arg_msys2(const char *arg)
+{
+	struct strbuf buf = STRBUF_INIT;
+	const char *p2 = arg, *p;
+
+	for (p = arg; *p; p++) {
+		int ws = isspace(*p);
+		if (!ws && *p != '\\' && *p != '"' && *p != '{')
+			continue;
+		if (!buf.len)
+			strbuf_addch(&buf, '"');
+		if (p != p2)
+			strbuf_add(&buf, p2, p - p2);
+		if (!ws && *p != '{')
+			strbuf_addch(&buf, '\\');
+		p2 = p;
+	}
+
+	if (p == arg)
+		strbuf_addch(&buf, '"');
+	else if (!buf.len)
+		return arg;
+	else
+		strbuf_add(&buf, p2, p - p2),
+
+	strbuf_addch(&buf, '"');
+	return strbuf_detach(&buf, 0);
+}
+
+static const char *parse_interpreter(const char *cmd)
+{
+	static char buf[100];
+	char *p, *opt;
+	int n, fd;
+
+	/* don't even try a .exe */
+	n = strlen(cmd);
+	if (n >= 4 && !strcasecmp(cmd+n-4, ".exe"))
+		return NULL;
+
+	fd = open(cmd, O_RDONLY);
+	if (fd < 0)
+		return NULL;
+	n = read(fd, buf, sizeof(buf)-1);
+	close(fd);
+	if (n < 4)	/* at least '#!/x' and not error */
+		return NULL;
+
+	if (buf[0] != '#' || buf[1] != '!')
+		return NULL;
+	buf[n] = '\0';
+	p = buf + strcspn(buf, "\r\n");
+	if (!*p)
+		return NULL;
+
+	*p = '\0';
+	if (!(p = strrchr(buf+2, '/')) && !(p = strrchr(buf+2, '\\')))
+		return NULL;
+	/* strip options */
+	if ((opt = strchr(p+1, ' ')))
+		*opt = '\0';
+	return p+1;
+}
+
+/*
+ * exe_only means that we only want to detect .exe files, but not scripts
+ * (which do not have an extension)
+ */
+static char *lookup_prog(const char *dir, int dirlen, const char *cmd,
+			 int isexe, int exe_only)
+{
+	char path[MAX_PATH];
+	snprintf(path, sizeof(path), "%.*s\\%s.exe", dirlen, dir, cmd);
+
+	if (!isexe && access(path, F_OK) == 0)
+		return xstrdup(path);
+	path[strlen(path)-4] = '\0';
+	if ((!exe_only || isexe) && access(path, F_OK) == 0)
+		if (!(GetFileAttributes(path) & FILE_ATTRIBUTE_DIRECTORY))
+			return xstrdup(path);
+	return NULL;
+}
+
+/*
+ * Determines the absolute path of cmd using the split path in path.
+ * If cmd contains a slash or backslash, no lookup is performed.
+ */
+static char *path_lookup(const char *cmd, int exe_only)
+{
+	const char *path;
+	char *prog = NULL;
+	int len = strlen(cmd);
+	int isexe = len >= 4 && !strcasecmp(cmd+len-4, ".exe");
+
+	if (strchr(cmd, '/') || strchr(cmd, '\\'))
+		return xstrdup(cmd);
+
+	path = mingw_getenv("PATH");
+	if (!path)
+		return NULL;
+
+	while (!prog) {
+		const char *sep = strchrnul(path, ';');
+		int dirlen = sep - path;
+		if (dirlen)
+			prog = lookup_prog(path, dirlen, cmd, isexe, exe_only);
+		if (!*sep)
+			break;
+		path = sep + 1;
+	}
+
+	return prog;
+}
+
+static const wchar_t *wcschrnul(const wchar_t *s, wchar_t c)
+{
+	while (*s && *s != c)
+		s++;
+	return s;
+}
+
+/* Compare only keys */
+static int wenvcmp(const void *a, const void *b)
+{
+	wchar_t *p = *(wchar_t **)a, *q = *(wchar_t **)b;
+	size_t p_len, q_len;
+
+	/* Find the keys */
+	p_len = wcschrnul(p, L'=') - p;
+	q_len = wcschrnul(q, L'=') - q;
+
+	/* If the length differs, include the shorter key's NUL */
+	if (p_len < q_len)
+		p_len++;
+	else if (p_len > q_len)
+		p_len = q_len + 1;
+
+	return _wcsnicmp(p, q, p_len);
+}
+
+/* We need a stable sort to convert the environment between UTF-16 <-> UTF-8 */
+#ifndef INTERNAL_QSORT
+#include "qsort.c"
+#endif
+
+/*
+ * Build an environment block combining the inherited environment
+ * merged with the given list of settings.
+ *
+ * Values of the form "KEY=VALUE" in deltaenv override inherited values.
+ * Values of the form "KEY" in deltaenv delete inherited values.
+ *
+ * Multiple entries in deltaenv for the same key are explicitly allowed.
+ *
+ * We return a contiguous block of UNICODE strings with a final trailing
+ * zero word.
+ */
+static wchar_t *make_environment_block(char **deltaenv)
+{
+	wchar_t *wenv = GetEnvironmentStringsW(), *wdeltaenv, *result, *p;
+	size_t wlen, s, delta_size, size;
+
+	wchar_t **array = NULL;
+	size_t alloc = 0, nr = 0, i;
+
+	size = 1; /* for extra NUL at the end */
+
+	/* If there is no deltaenv to apply, simply return a copy. */
+	if (!deltaenv || !*deltaenv) {
+		for (p = wenv; p && *p; ) {
+			size_t s = wcslen(p) + 1;
+			size += s;
+			p += s;
+		}
+
+		ALLOC_ARRAY(result, size);
+		memcpy(result, wenv, size * sizeof(*wenv));
+		FreeEnvironmentStringsW(wenv);
+		return result;
+	}
+
+	/*
+	 * If there is a deltaenv, let's accumulate all keys into `array`,
+	 * sort them using the stable git_qsort() and then copy, skipping
+	 * duplicate keys
+	 */
+	for (p = wenv; p && *p; ) {
+		ALLOC_GROW(array, nr + 1, alloc);
+		s = wcslen(p) + 1;
+		array[nr++] = p;
+		p += s;
+		size += s;
+	}
+
+	/* (over-)assess size needed for wchar version of deltaenv */
+	for (delta_size = 0, i = 0; deltaenv[i]; i++)
+		delta_size += strlen(deltaenv[i]) * 2 + 1;
+	ALLOC_ARRAY(wdeltaenv, delta_size);
+
+	/* convert the deltaenv, appending to array */
+	for (i = 0, p = wdeltaenv; deltaenv[i]; i++) {
+		ALLOC_GROW(array, nr + 1, alloc);
+		wlen = xutftowcs(p, deltaenv[i], wdeltaenv + delta_size - p);
+		array[nr++] = p;
+		p += wlen + 1;
+	}
+
+	git_qsort(array, nr, sizeof(*array), wenvcmp);
+	ALLOC_ARRAY(result, size + delta_size);
+
+	for (p = result, i = 0; i < nr; i++) {
+		/* Skip any duplicate keys; last one wins */
+		while (i + 1 < nr && !wenvcmp(array + i, array + i + 1))
+		       i++;
+
+		/* Skip "to delete" entry */
+		if (!wcschr(array[i], L'='))
+			continue;
+
+		size = wcslen(array[i]) + 1;
+		memcpy(p, array[i], size * sizeof(*p));
+		p += size;
+	}
+	*p = L'\0';
+
+	free(array);
+	free(wdeltaenv);
+	FreeEnvironmentStringsW(wenv);
+	return result;
+}
+
+static void do_unset_environment_variables(void)
+{
+	static int done;
+	char *p = unset_environment_variables;
+
+	if (done || !p)
+		return;
+	done = 1;
+
+	for (;;) {
+		char *comma = strchr(p, ',');
+
+		if (comma)
+			*comma = '\0';
+		unsetenv(p);
+		if (!comma)
+			break;
+		p = comma + 1;
+	}
+}
+
+struct pinfo_t {
+	struct pinfo_t *next;
+	pid_t pid;
+	HANDLE proc;
+};
+static struct pinfo_t *pinfo = NULL;
+CRITICAL_SECTION pinfo_cs;
+
+/* Used to match and chomp off path components */
+static inline int match_last_path_component(const char *path, size_t *len,
+					    const char *component)
+{
+	size_t component_len = strlen(component);
+	if (*len < component_len + 1 ||
+	    !is_dir_sep(path[*len - component_len - 1]) ||
+	    fspathncmp(path + *len - component_len, component, component_len))
+		return 0;
+	*len -= component_len + 1;
+	/* chomp off repeated dir separators */
+	while (*len > 0 && is_dir_sep(path[*len - 1]))
+		(*len)--;
+	return 1;
+}
+
+static int is_msys2_sh(const char *cmd)
+{
+	if (cmd && !strcmp(cmd, "sh")) {
+		static int ret = -1;
+		char *p;
+
+		if (ret >= 0)
+			return ret;
+
+		p = path_lookup(cmd, 0);
+		if (!p)
+			ret = 0;
+		else {
+			size_t len = strlen(p);
+
+			ret = match_last_path_component(p, &len, "sh.exe") &&
+				match_last_path_component(p, &len, "bin") &&
+				match_last_path_component(p, &len, "usr");
+			free(p);
+		}
+		return ret;
+	}
+	return 0;
+}
+
+static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **deltaenv,
+			      const char *dir,
+			      int prepend_cmd, int fhin, int fhout, int fherr)
+{
+	STARTUPINFOW si;
+	PROCESS_INFORMATION pi;
+	struct strbuf args;
+	wchar_t wcmd[MAX_PATH], wdir[MAX_PATH], *wargs, *wenvblk = NULL;
+	unsigned flags = CREATE_UNICODE_ENVIRONMENT;
+	BOOL ret;
+	HANDLE cons;
+	const char *(*quote_arg)(const char *arg) =
+		is_msys2_sh(*argv) ? quote_arg_msys2 : quote_arg_msvc;
+
+	do_unset_environment_variables();
+
+	/* Determine whether or not we are associated to a console */
+	cons = CreateFileW(L"CONOUT$", GENERIC_WRITE,
+			FILE_SHARE_WRITE, NULL, OPEN_EXISTING,
+			FILE_ATTRIBUTE_NORMAL, NULL);
+	if (cons == INVALID_HANDLE_VALUE) {
+		/* There is no console associated with this process.
+		 * Since the child is a console process, Windows
+		 * would normally create a console window. But
+		 * since we'll be redirecting std streams, we do
+		 * not need the console.
+		 * It is necessary to use DETACHED_PROCESS
+		 * instead of CREATE_NO_WINDOW to make ssh
+		 * recognize that it has no console.
+		 */
+		flags |= DETACHED_PROCESS;
+	} else {
+		/* There is already a console. If we specified
+		 * DETACHED_PROCESS here, too, Windows would
+		 * disassociate the child from the console.
+		 * The same is true for CREATE_NO_WINDOW.
+		 * Go figure!
+		 */
+		CloseHandle(cons);
+	}
+	memset(&si, 0, sizeof(si));
+	si.cb = sizeof(si);
+	si.dwFlags = STARTF_USESTDHANDLES;
+	si.hStdInput = winansi_get_osfhandle(fhin);
+	si.hStdOutput = winansi_get_osfhandle(fhout);
+	si.hStdError = winansi_get_osfhandle(fherr);
+
+	if (*argv && !strcmp(cmd, *argv))
+		wcmd[0] = L'\0';
+	else if (xutftowcs_path(wcmd, cmd) < 0)
+		return -1;
+	if (dir && xutftowcs_path(wdir, dir) < 0)
+		return -1;
+
+	/* concatenate argv, quoting args as we go */
+	strbuf_init(&args, 0);
+	if (prepend_cmd) {
+		char *quoted = (char *)quote_arg(cmd);
+		strbuf_addstr(&args, quoted);
+		if (quoted != cmd)
+			free(quoted);
+	}
+	for (; *argv; argv++) {
+		char *quoted = (char *)quote_arg(*argv);
+		if (*args.buf)
+			strbuf_addch(&args, ' ');
+		strbuf_addstr(&args, quoted);
+		if (quoted != *argv)
+			free(quoted);
+	}
+
+	ALLOC_ARRAY(wargs, st_add(st_mult(2, args.len), 1));
+	xutftowcs(wargs, args.buf, 2 * args.len + 1);
+	strbuf_release(&args);
+
+	wenvblk = make_environment_block(deltaenv);
+
+	memset(&pi, 0, sizeof(pi));
+	ret = CreateProcessW(*wcmd ? wcmd : NULL, wargs, NULL, NULL, TRUE,
+		flags, wenvblk, dir ? wdir : NULL, &si, &pi);
+
+	free(wenvblk);
+	free(wargs);
+
+	if (!ret) {
+		errno = ENOENT;
+		return -1;
+	}
+	CloseHandle(pi.hThread);
+
+	/*
+	 * The process ID is the human-readable identifier of the process
+	 * that we want to present in log and error messages. The handle
+	 * is not useful for this purpose. But we cannot close it, either,
+	 * because it is not possible to turn a process ID into a process
+	 * handle after the process terminated.
+	 * Keep the handle in a list for waitpid.
+	 */
+	EnterCriticalSection(&pinfo_cs);
+	{
+		struct pinfo_t *info = xmalloc(sizeof(struct pinfo_t));
+		info->pid = pi.dwProcessId;
+		info->proc = pi.hProcess;
+		info->next = pinfo;
+		pinfo = info;
+	}
+	LeaveCriticalSection(&pinfo_cs);
+
+	return (pid_t)pi.dwProcessId;
+}
+
+static pid_t mingw_spawnv(const char *cmd, const char **argv, int prepend_cmd)
+{
+	return mingw_spawnve_fd(cmd, argv, NULL, NULL, prepend_cmd, 0, 1, 2);
+}
+
+pid_t mingw_spawnvpe(const char *cmd, const char **argv, char **deltaenv,
+		     const char *dir,
+		     int fhin, int fhout, int fherr)
+{
+	pid_t pid;
+	char *prog = path_lookup(cmd, 0);
+
+	if (!prog) {
+		errno = ENOENT;
+		pid = -1;
+	}
+	else {
+		const char *interpr = parse_interpreter(prog);
+
+		if (interpr) {
+			const char *argv0 = argv[0];
+			char *iprog = path_lookup(interpr, 1);
+			argv[0] = prog;
+			if (!iprog) {
+				errno = ENOENT;
+				pid = -1;
+			}
+			else {
+				pid = mingw_spawnve_fd(iprog, argv, deltaenv, dir, 1,
+						       fhin, fhout, fherr);
+				free(iprog);
+			}
+			argv[0] = argv0;
+		}
+		else
+			pid = mingw_spawnve_fd(prog, argv, deltaenv, dir, 0,
+					       fhin, fhout, fherr);
+		free(prog);
+	}
+	return pid;
+}
+
+static int try_shell_exec(const char *cmd, char *const *argv)
+{
+	const char *interpr = parse_interpreter(cmd);
+	char *prog;
+	int pid = 0;
+
+	if (!interpr)
+		return 0;
+	prog = path_lookup(interpr, 1);
+	if (prog) {
+		int exec_id;
+		int argc = 0;
+#ifndef _MSC_VER
+		const
+#endif
+		char **argv2;
+		while (argv[argc]) argc++;
+		ALLOC_ARRAY(argv2, argc + 1);
+		argv2[0] = (char *)cmd;	/* full path to the script file */
+		memcpy(&argv2[1], &argv[1], sizeof(*argv) * argc);
+		exec_id = trace2_exec(prog, argv2);
+		pid = mingw_spawnv(prog, argv2, 1);
+		if (pid >= 0) {
+			int status;
+			if (waitpid(pid, &status, 0) < 0)
+				status = 255;
+			trace2_exec_result(exec_id, status);
+			exit(status);
+		}
+		trace2_exec_result(exec_id, -1);
+		pid = 1;	/* indicate that we tried but failed */
+		free(prog);
+		free(argv2);
+	}
+	return pid;
+}
+
+int mingw_execv(const char *cmd, char *const *argv)
+{
+	/* check if git_command is a shell script */
+	if (!try_shell_exec(cmd, argv)) {
+		int pid, status;
+		int exec_id;
+
+		exec_id = trace2_exec(cmd, (const char **)argv);
+		pid = mingw_spawnv(cmd, (const char **)argv, 0);
+		if (pid < 0) {
+			trace2_exec_result(exec_id, -1);
+			return -1;
+		}
+		if (waitpid(pid, &status, 0) < 0)
+			status = 255;
+		trace2_exec_result(exec_id, status);
+		exit(status);
+	}
+	return -1;
+}
+
+int mingw_execvp(const char *cmd, char *const *argv)
+{
+	char *prog = path_lookup(cmd, 0);
+
+	if (prog) {
+		mingw_execv(prog, argv);
+		free(prog);
+	} else
+		errno = ENOENT;
+
+	return -1;
+}
+
+int mingw_kill(pid_t pid, int sig)
+{
+	if (pid > 0 && sig == SIGTERM) {
+		HANDLE h = OpenProcess(PROCESS_TERMINATE, FALSE, pid);
+
+		if (TerminateProcess(h, -1)) {
+			CloseHandle(h);
+			return 0;
+		}
+
+		errno = err_win_to_posix(GetLastError());
+		CloseHandle(h);
+		return -1;
+	} else if (pid > 0 && sig == 0) {
+		HANDLE h = OpenProcess(PROCESS_QUERY_INFORMATION, FALSE, pid);
+		if (h) {
+			CloseHandle(h);
+			return 0;
+		}
+	}
+
+	errno = EINVAL;
+	return -1;
+}
+
+/*
+ * UTF-8 versions of getenv(), putenv() and unsetenv().
+ * Internally, they use the CRT's stock UNICODE routines
+ * to avoid data loss.
+ */
+char *mingw_getenv(const char *name)
+{
+#define GETENV_MAX_RETAIN 64
+	static char *values[GETENV_MAX_RETAIN];
+	static int value_counter;
+	int len_key, len_value;
+	wchar_t *w_key;
+	char *value;
+	wchar_t w_value[32768];
+
+	if (!name || !*name)
+		return NULL;
+
+	len_key = strlen(name) + 1;
+	/* We cannot use xcalloc() here because that uses getenv() itself */
+	w_key = calloc(len_key, sizeof(wchar_t));
+	if (!w_key)
+		die("Out of memory, (tried to allocate %u wchar_t's)", len_key);
+	xutftowcs(w_key, name, len_key);
+	len_value = GetEnvironmentVariableW(w_key, w_value, ARRAY_SIZE(w_value));
+	if (!len_value && GetLastError() == ERROR_ENVVAR_NOT_FOUND) {
+		free(w_key);
+		return NULL;
+	}
+	free(w_key);
+
+	len_value = len_value * 3 + 1;
+	/* We cannot use xcalloc() here because that uses getenv() itself */
+	value = calloc(len_value, sizeof(char));
+	if (!value)
+		die("Out of memory, (tried to allocate %u bytes)", len_value);
+	xwcstoutf(value, w_value, len_value);
+
+	/*
+	 * We return `value` which is an allocated value and the caller is NOT
+	 * expecting to have to free it, so we keep a round-robin array,
+	 * invalidating the buffer after GETENV_MAX_RETAIN getenv() calls.
+	 */
+	free(values[value_counter]);
+	values[value_counter++] = value;
+	if (value_counter >= ARRAY_SIZE(values))
+		value_counter = 0;
+
+	return value;
+}
+
+int mingw_putenv(const char *namevalue)
+{
+	int size;
+	wchar_t *wide, *equal;
+	BOOL result;
+
+	if (!namevalue || !*namevalue)
+		return 0;
+
+	size = strlen(namevalue) * 2 + 1;
+	wide = calloc(size, sizeof(wchar_t));
+	if (!wide)
+		die("Out of memory, (tried to allocate %u wchar_t's)", size);
+	xutftowcs(wide, namevalue, size);
+	equal = wcschr(wide, L'=');
+	if (!equal)
+		result = SetEnvironmentVariableW(wide, NULL);
+	else {
+		*equal = L'\0';
+		result = SetEnvironmentVariableW(wide, equal + 1);
+	}
+	free(wide);
+
+	if (!result)
+		errno = err_win_to_posix(GetLastError());
+
+	return result ? 0 : -1;
+}
+
+static void ensure_socket_initialization(void)
+{
+	WSADATA wsa;
+	static int initialized = 0;
+
+	if (initialized)
+		return;
+
+	if (WSAStartup(MAKEWORD(2,2), &wsa))
+		die("unable to initialize winsock subsystem, error %d",
+			WSAGetLastError());
+
+	atexit((void(*)(void)) WSACleanup);
+	initialized = 1;
+}
+
+#undef gethostname
+int mingw_gethostname(char *name, int namelen)
+{
+    ensure_socket_initialization();
+    return gethostname(name, namelen);
+}
+
+#undef gethostbyname
+struct hostent *mingw_gethostbyname(const char *host)
+{
+	ensure_socket_initialization();
+	return gethostbyname(host);
+}
+
+#undef getaddrinfo
+int mingw_getaddrinfo(const char *node, const char *service,
+		      const struct addrinfo *hints, struct addrinfo **res)
+{
+	ensure_socket_initialization();
+	return getaddrinfo(node, service, hints, res);
+}
+
+int mingw_socket(int domain, int type, int protocol)
+{
+	int sockfd;
+	SOCKET s;
+
+	ensure_socket_initialization();
+	s = WSASocket(domain, type, protocol, NULL, 0, 0);
+	if (s == INVALID_SOCKET) {
+		/*
+		 * WSAGetLastError() values are regular BSD error codes
+		 * biased by WSABASEERR.
+		 * However, strerror() does not know about networking
+		 * specific errors, which are values beginning at 38 or so.
+		 * Therefore, we choose to leave the biased error code
+		 * in errno so that _if_ someone looks up the code somewhere,
+		 * then it is at least the number that are usually listed.
+		 */
+		errno = WSAGetLastError();
+		return -1;
+	}
+	/* convert into a file descriptor */
+	if ((sockfd = _open_osfhandle(s, O_RDWR|O_BINARY)) < 0) {
+		closesocket(s);
+		return error("unable to make a socket file descriptor: %s",
+			strerror(errno));
+	}
+	return sockfd;
+}
+
+#undef connect
+int mingw_connect(int sockfd, struct sockaddr *sa, size_t sz)
+{
+	SOCKET s = (SOCKET)_get_osfhandle(sockfd);
+	return connect(s, sa, sz);
+}
+
+#undef bind
+int mingw_bind(int sockfd, struct sockaddr *sa, size_t sz)
+{
+	SOCKET s = (SOCKET)_get_osfhandle(sockfd);
+	return bind(s, sa, sz);
+}
+
+#undef setsockopt
+int mingw_setsockopt(int sockfd, int lvl, int optname, void *optval, int optlen)
+{
+	SOCKET s = (SOCKET)_get_osfhandle(sockfd);
+	return setsockopt(s, lvl, optname, (const char*)optval, optlen);
+}
+
+#undef shutdown
+int mingw_shutdown(int sockfd, int how)
+{
+	SOCKET s = (SOCKET)_get_osfhandle(sockfd);
+	return shutdown(s, how);
+}
+
+#undef listen
+int mingw_listen(int sockfd, int backlog)
+{
+	SOCKET s = (SOCKET)_get_osfhandle(sockfd);
+	return listen(s, backlog);
+}
+
+#undef accept
+int mingw_accept(int sockfd1, struct sockaddr *sa, socklen_t *sz)
+{
+	int sockfd2;
+
+	SOCKET s1 = (SOCKET)_get_osfhandle(sockfd1);
+	SOCKET s2 = accept(s1, sa, sz);
+
+	/* convert into a file descriptor */
+	if ((sockfd2 = _open_osfhandle(s2, O_RDWR|O_BINARY)) < 0) {
+		int err = errno;
+		closesocket(s2);
+		return error("unable to make a socket file descriptor: %s",
+			strerror(err));
+	}
+	return sockfd2;
+}
+
+#undef rename
+int mingw_rename(const char *pold, const char *pnew)
+{
+	DWORD attrs, gle;
+	int tries = 0;
+	wchar_t wpold[MAX_PATH], wpnew[MAX_PATH];
+	if (xutftowcs_path(wpold, pold) < 0 || xutftowcs_path(wpnew, pnew) < 0)
+		return -1;
+
+	/*
+	 * Try native rename() first to get errno right.
+	 * It is based on MoveFile(), which cannot overwrite existing files.
+	 */
+	if (!_wrename(wpold, wpnew))
+		return 0;
+	if (errno != EEXIST)
+		return -1;
+repeat:
+	if (MoveFileExW(wpold, wpnew, MOVEFILE_REPLACE_EXISTING))
+		return 0;
+	/* TODO: translate more errors */
+	gle = GetLastError();
+	if (gle == ERROR_ACCESS_DENIED &&
+	    (attrs = GetFileAttributesW(wpnew)) != INVALID_FILE_ATTRIBUTES) {
+		if (attrs & FILE_ATTRIBUTE_DIRECTORY) {
+			DWORD attrsold = GetFileAttributesW(wpold);
+			if (attrsold == INVALID_FILE_ATTRIBUTES ||
+			    !(attrsold & FILE_ATTRIBUTE_DIRECTORY))
+				errno = EISDIR;
+			else if (!_wrmdir(wpnew))
+				goto repeat;
+			return -1;
+		}
+		if ((attrs & FILE_ATTRIBUTE_READONLY) &&
+		    SetFileAttributesW(wpnew, attrs & ~FILE_ATTRIBUTE_READONLY)) {
+			if (MoveFileExW(wpold, wpnew, MOVEFILE_REPLACE_EXISTING))
+				return 0;
+			gle = GetLastError();
+			/* revert file attributes on failure */
+			SetFileAttributesW(wpnew, attrs);
+		}
+	}
+	if (tries < ARRAY_SIZE(delay) && gle == ERROR_ACCESS_DENIED) {
+		/*
+		 * We assume that some other process had the source or
+		 * destination file open at the wrong moment and retry.
+		 * In order to give the other process a higher chance to
+		 * complete its operation, we give up our time slice now.
+		 * If we have to retry again, we do sleep a bit.
+		 */
+		Sleep(delay[tries]);
+		tries++;
+		goto repeat;
+	}
+	if (gle == ERROR_ACCESS_DENIED &&
+	       ask_yes_no_if_possible("Rename from '%s' to '%s' failed. "
+		       "Should I try again?", pold, pnew))
+		goto repeat;
+
+	errno = EACCES;
+	return -1;
+}
+
+/*
+ * Note that this doesn't return the actual pagesize, but
+ * the allocation granularity. If future Windows specific git code
+ * needs the real getpagesize function, we need to find another solution.
+ */
+int mingw_getpagesize(void)
+{
+	SYSTEM_INFO si;
+	GetSystemInfo(&si);
+	return si.dwAllocationGranularity;
+}
+
+/* See https://msdn.microsoft.com/en-us/library/windows/desktop/ms724435.aspx */
+enum EXTENDED_NAME_FORMAT {
+	NameDisplay = 3,
+	NameUserPrincipal = 8
+};
+
+static char *get_extended_user_info(enum EXTENDED_NAME_FORMAT type)
+{
+	DECLARE_PROC_ADDR(secur32.dll, BOOL, GetUserNameExW,
+		enum EXTENDED_NAME_FORMAT, LPCWSTR, PULONG);
+	static wchar_t wbuffer[1024];
+	DWORD len;
+
+	if (!INIT_PROC_ADDR(GetUserNameExW))
+		return NULL;
+
+	len = ARRAY_SIZE(wbuffer);
+	if (GetUserNameExW(type, wbuffer, &len)) {
+		char *converted = xmalloc((len *= 3));
+		if (xwcstoutf(converted, wbuffer, len) >= 0)
+			return converted;
+		free(converted);
+	}
+
+	return NULL;
+}
+
+char *mingw_query_user_email(void)
+{
+	return get_extended_user_info(NameUserPrincipal);
+}
+
+struct passwd *getpwuid(int uid)
+{
+	static unsigned initialized;
+	static char user_name[100];
+	static struct passwd *p;
+	wchar_t buf[100];
+	DWORD len;
+
+	if (initialized)
+		return p;
+
+	len = ARRAY_SIZE(buf);
+	if (!GetUserNameW(buf, &len)) {
+		initialized = 1;
+		return NULL;
+	}
+
+	if (xwcstoutf(user_name, buf, sizeof(user_name)) < 0) {
+		initialized = 1;
+		return NULL;
+	}
+
+	p = xmalloc(sizeof(*p));
+	p->pw_name = user_name;
+	p->pw_gecos = get_extended_user_info(NameDisplay);
+	if (!p->pw_gecos)
+		p->pw_gecos = "unknown";
+	p->pw_dir = NULL;
+
+	initialized = 1;
+	return p;
+}
+
+static HANDLE timer_event;
+static HANDLE timer_thread;
+static int timer_interval;
+static int one_shot;
+static sig_handler_t timer_fn = SIG_DFL, sigint_fn = SIG_DFL;
+
+/* The timer works like this:
+ * The thread, ticktack(), is a trivial routine that most of the time
+ * only waits to receive the signal to terminate. The main thread tells
+ * the thread to terminate by setting the timer_event to the signalled
+ * state.
+ * But ticktack() interrupts the wait state after the timer's interval
+ * length to call the signal handler.
+ */
+
+static unsigned __stdcall ticktack(void *dummy)
+{
+	while (WaitForSingleObject(timer_event, timer_interval) == WAIT_TIMEOUT) {
+		mingw_raise(SIGALRM);
+		if (one_shot)
+			break;
+	}
+	return 0;
+}
+
+static int start_timer_thread(void)
+{
+	timer_event = CreateEvent(NULL, FALSE, FALSE, NULL);
+	if (timer_event) {
+		timer_thread = (HANDLE) _beginthreadex(NULL, 0, ticktack, NULL, 0, NULL);
+		if (!timer_thread )
+			return errno = ENOMEM,
+				error("cannot start timer thread");
+	} else
+		return errno = ENOMEM,
+			error("cannot allocate resources for timer");
+	return 0;
+}
+
+static void stop_timer_thread(void)
+{
+	if (timer_event)
+		SetEvent(timer_event);	/* tell thread to terminate */
+	if (timer_thread) {
+		int rc = WaitForSingleObject(timer_thread, 10000);
+		if (rc == WAIT_TIMEOUT)
+			error("timer thread did not terminate timely");
+		else if (rc != WAIT_OBJECT_0)
+			error("waiting for timer thread failed: %lu",
+			      GetLastError());
+		CloseHandle(timer_thread);
+	}
+	if (timer_event)
+		CloseHandle(timer_event);
+	timer_event = NULL;
+	timer_thread = NULL;
+}
+
+static inline int is_timeval_eq(const struct timeval *i1, const struct timeval *i2)
+{
+	return i1->tv_sec == i2->tv_sec && i1->tv_usec == i2->tv_usec;
+}
+
+int setitimer(int type, struct itimerval *in, struct itimerval *out)
+{
+	static const struct timeval zero;
+	static int atexit_done;
+
+	if (out != NULL)
+		return errno = EINVAL,
+			error("setitimer param 3 != NULL not implemented");
+	if (!is_timeval_eq(&in->it_interval, &zero) &&
+	    !is_timeval_eq(&in->it_interval, &in->it_value))
+		return errno = EINVAL,
+			error("setitimer: it_interval must be zero or eq it_value");
+
+	if (timer_thread)
+		stop_timer_thread();
+
+	if (is_timeval_eq(&in->it_value, &zero) &&
+	    is_timeval_eq(&in->it_interval, &zero))
+		return 0;
+
+	timer_interval = in->it_value.tv_sec * 1000 + in->it_value.tv_usec / 1000;
+	one_shot = is_timeval_eq(&in->it_interval, &zero);
+	if (!atexit_done) {
+		atexit(stop_timer_thread);
+		atexit_done = 1;
+	}
+	return start_timer_thread();
+}
+
+int sigaction(int sig, struct sigaction *in, struct sigaction *out)
+{
+	if (sig != SIGALRM)
+		return errno = EINVAL,
+			error("sigaction only implemented for SIGALRM");
+	if (out != NULL)
+		return errno = EINVAL,
+			error("sigaction: param 3 != NULL not implemented");
+
+	timer_fn = in->sa_handler;
+	return 0;
+}
+
+#undef signal
+sig_handler_t mingw_signal(int sig, sig_handler_t handler)
+{
+	sig_handler_t old;
+
+	switch (sig) {
+	case SIGALRM:
+		old = timer_fn;
+		timer_fn = handler;
+		break;
+
+	case SIGINT:
+		old = sigint_fn;
+		sigint_fn = handler;
+		break;
+
+	default:
+		return signal(sig, handler);
+	}
+
+	return old;
+}
+
+#undef raise
+int mingw_raise(int sig)
+{
+	switch (sig) {
+	case SIGALRM:
+		if (timer_fn == SIG_DFL) {
+			if (isatty(STDERR_FILENO))
+				fputs("Alarm clock\n", stderr);
+			exit(128 + SIGALRM);
+		} else if (timer_fn != SIG_IGN)
+			timer_fn(SIGALRM);
+		return 0;
+
+	case SIGINT:
+		if (sigint_fn == SIG_DFL)
+			exit(128 + SIGINT);
+		else if (sigint_fn != SIG_IGN)
+			sigint_fn(SIGINT);
+		return 0;
+
+#if defined(_MSC_VER)
+	case SIGILL:
+	case SIGFPE:
+	case SIGSEGV:
+	case SIGTERM:
+	case SIGBREAK:
+	case SIGABRT:
+	case SIGABRT_COMPAT:
+		/*
+		 * The <signal.h> header in the MS C Runtime defines 8 signals
+		 * as being supported on the platform. Anything else causes an
+		 * "Invalid signal or error" (which in DEBUG builds causes the
+		 * Abort/Retry/Ignore dialog). We by-pass the CRT for things we
+		 * already know will fail.
+		 */
+		return raise(sig);
+	default:
+		errno = EINVAL;
+		return -1;
+
+#else
+
+	default:
+		return raise(sig);
+
+#endif
+
+	}
+}
+
+int link(const char *oldpath, const char *newpath)
+{
+	wchar_t woldpath[MAX_PATH], wnewpath[MAX_PATH];
+	if (xutftowcs_path(woldpath, oldpath) < 0 ||
+		xutftowcs_path(wnewpath, newpath) < 0)
+		return -1;
+
+	if (!CreateHardLinkW(wnewpath, woldpath, NULL)) {
+		errno = err_win_to_posix(GetLastError());
+		return -1;
+	}
+	return 0;
+}
+
+pid_t waitpid(pid_t pid, int *status, int options)
+{
+	HANDLE h = OpenProcess(SYNCHRONIZE | PROCESS_QUERY_INFORMATION,
+	    FALSE, pid);
+	if (!h) {
+		errno = ECHILD;
+		return -1;
+	}
+
+	if (pid > 0 && options & WNOHANG) {
+		if (WAIT_OBJECT_0 != WaitForSingleObject(h, 0)) {
+			CloseHandle(h);
+			return 0;
+		}
+		options &= ~WNOHANG;
+	}
+
+	if (options == 0) {
+		struct pinfo_t **ppinfo;
+		if (WaitForSingleObject(h, INFINITE) != WAIT_OBJECT_0) {
+			CloseHandle(h);
+			return 0;
+		}
+
+		if (status)
+			GetExitCodeProcess(h, (LPDWORD)status);
+
+		EnterCriticalSection(&pinfo_cs);
+
+		ppinfo = &pinfo;
+		while (*ppinfo) {
+			struct pinfo_t *info = *ppinfo;
+			if (info->pid == pid) {
+				CloseHandle(info->proc);
+				*ppinfo = info->next;
+				free(info);
+				break;
+			}
+			ppinfo = &info->next;
+		}
+
+		LeaveCriticalSection(&pinfo_cs);
+
+		CloseHandle(h);
+		return pid;
+	}
+	CloseHandle(h);
+
+	errno = EINVAL;
+	return -1;
+}
+
+int xutftowcsn(wchar_t *wcs, const char *utfs, size_t wcslen, int utflen)
+{
+	int upos = 0, wpos = 0;
+	const unsigned char *utf = (const unsigned char*) utfs;
+	if (!utf || !wcs || wcslen < 1) {
+		errno = EINVAL;
+		return -1;
+	}
+	/* reserve space for \0 */
+	wcslen--;
+	if (utflen < 0)
+		utflen = INT_MAX;
+
+	while (upos < utflen) {
+		int c = utf[upos++] & 0xff;
+		if (utflen == INT_MAX && c == 0)
+			break;
+
+		if (wpos >= wcslen) {
+			wcs[wpos] = 0;
+			errno = ERANGE;
+			return -1;
+		}
+
+		if (c < 0x80) {
+			/* ASCII */
+			wcs[wpos++] = c;
+		} else if (c >= 0xc2 && c < 0xe0 && upos < utflen &&
+				(utf[upos] & 0xc0) == 0x80) {
+			/* 2-byte utf-8 */
+			c = ((c & 0x1f) << 6);
+			c |= (utf[upos++] & 0x3f);
+			wcs[wpos++] = c;
+		} else if (c >= 0xe0 && c < 0xf0 && upos + 1 < utflen &&
+				!(c == 0xe0 && utf[upos] < 0xa0) && /* over-long encoding */
+				(utf[upos] & 0xc0) == 0x80 &&
+				(utf[upos + 1] & 0xc0) == 0x80) {
+			/* 3-byte utf-8 */
+			c = ((c & 0x0f) << 12);
+			c |= ((utf[upos++] & 0x3f) << 6);
+			c |= (utf[upos++] & 0x3f);
+			wcs[wpos++] = c;
+		} else if (c >= 0xf0 && c < 0xf5 && upos + 2 < utflen &&
+				wpos + 1 < wcslen &&
+				!(c == 0xf0 && utf[upos] < 0x90) && /* over-long encoding */
+				!(c == 0xf4 && utf[upos] >= 0x90) && /* > \u10ffff */
+				(utf[upos] & 0xc0) == 0x80 &&
+				(utf[upos + 1] & 0xc0) == 0x80 &&
+				(utf[upos + 2] & 0xc0) == 0x80) {
+			/* 4-byte utf-8: convert to \ud8xx \udcxx surrogate pair */
+			c = ((c & 0x07) << 18);
+			c |= ((utf[upos++] & 0x3f) << 12);
+			c |= ((utf[upos++] & 0x3f) << 6);
+			c |= (utf[upos++] & 0x3f);
+			c -= 0x10000;
+			wcs[wpos++] = 0xd800 | (c >> 10);
+			wcs[wpos++] = 0xdc00 | (c & 0x3ff);
+		} else if (c >= 0xa0) {
+			/* invalid utf-8 byte, printable unicode char: convert 1:1 */
+			wcs[wpos++] = c;
+		} else {
+			/* invalid utf-8 byte, non-printable unicode: convert to hex */
+			static const char *hex = "0123456789abcdef";
+			wcs[wpos++] = hex[c >> 4];
+			if (wpos < wcslen)
+				wcs[wpos++] = hex[c & 0x0f];
+		}
+	}
+	wcs[wpos] = 0;
+	return wpos;
+}
+
+int xwcstoutf(char *utf, const wchar_t *wcs, size_t utflen)
+{
+	if (!wcs || !utf || utflen < 1) {
+		errno = EINVAL;
+		return -1;
+	}
+	utflen = WideCharToMultiByte(CP_UTF8, 0, wcs, -1, utf, utflen, NULL, NULL);
+	if (utflen)
+		return utflen - 1;
+	errno = ERANGE;
+	return -1;
+}
+
+static void setup_windows_environment(void)
+{
+	char *tmp = getenv("TMPDIR");
+
+	/* on Windows it is TMP and TEMP */
+	if (!tmp) {
+		if (!(tmp = getenv("TMP")))
+			tmp = getenv("TEMP");
+		if (tmp) {
+			setenv("TMPDIR", tmp, 1);
+			tmp = getenv("TMPDIR");
+		}
+	}
+
+	if (tmp) {
+		/*
+		 * Convert all dir separators to forward slashes,
+		 * to help shell commands called from the Git
+		 * executable (by not mistaking the dir separators
+		 * for escape characters).
+		 */
+		convert_slashes(tmp);
+	}
+
+	/* simulate TERM to enable auto-color (see color.c) */
+	if (!getenv("TERM"))
+		setenv("TERM", "cygwin", 1);
+
+	/* calculate HOME if not set */
+	if (!getenv("HOME")) {
+		/*
+		 * try $HOMEDRIVE$HOMEPATH - the home share may be a network
+		 * location, thus also check if the path exists (i.e. is not
+		 * disconnected)
+		 */
+		if ((tmp = getenv("HOMEDRIVE"))) {
+			struct strbuf buf = STRBUF_INIT;
+			strbuf_addstr(&buf, tmp);
+			if ((tmp = getenv("HOMEPATH"))) {
+				strbuf_addstr(&buf, tmp);
+				if (is_directory(buf.buf))
+					setenv("HOME", buf.buf, 1);
+				else
+					tmp = NULL; /* use $USERPROFILE */
+			}
+			strbuf_release(&buf);
+		}
+		/* use $USERPROFILE if the home share is not available */
+		if (!tmp && (tmp = getenv("USERPROFILE")))
+			setenv("HOME", tmp, 1);
+	}
+}
+
+#if !defined(_MSC_VER)
+/*
+ * Disable MSVCRT command line wildcard expansion (__getmainargs called from
+ * mingw startup code, see init.c in mingw runtime).
+ */
+int _CRT_glob = 0;
+#endif
+
+static NORETURN void die_startup(void)
+{
+	fputs("fatal: not enough memory for initialization", stderr);
+	exit(128);
+}
+
+static void *malloc_startup(size_t size)
+{
+	void *result = malloc(size);
+	if (!result)
+		die_startup();
+	return result;
+}
+
+static char *wcstoutfdup_startup(char *buffer, const wchar_t *wcs, size_t len)
+{
+	len = xwcstoutf(buffer, wcs, len) + 1;
+	return memcpy(malloc_startup(len), buffer, len);
+}
+
+static void maybe_redirect_std_handle(const wchar_t *key, DWORD std_id, int fd,
+				      DWORD desired_access, DWORD flags)
+{
+	DWORD create_flag = fd ? OPEN_ALWAYS : OPEN_EXISTING;
+	wchar_t buf[MAX_PATH];
+	DWORD max = ARRAY_SIZE(buf);
+	HANDLE handle;
+	DWORD ret = GetEnvironmentVariableW(key, buf, max);
+
+	if (!ret || ret >= max)
+		return;
+
+	/* make sure this does not leak into child processes */
+	SetEnvironmentVariableW(key, NULL);
+	if (!wcscmp(buf, L"off")) {
+		close(fd);
+		handle = GetStdHandle(std_id);
+		if (handle != INVALID_HANDLE_VALUE)
+			CloseHandle(handle);
+		return;
+	}
+	if (std_id == STD_ERROR_HANDLE && !wcscmp(buf, L"2>&1")) {
+		handle = GetStdHandle(STD_OUTPUT_HANDLE);
+		if (handle == INVALID_HANDLE_VALUE) {
+			close(fd);
+			handle = GetStdHandle(std_id);
+			if (handle != INVALID_HANDLE_VALUE)
+				CloseHandle(handle);
+		} else {
+			int new_fd = _open_osfhandle((intptr_t)handle, O_BINARY);
+			SetStdHandle(std_id, handle);
+			dup2(new_fd, fd);
+			/* do *not* close the new_fd: that would close stdout */
+		}
+		return;
+	}
+	handle = CreateFileW(buf, desired_access, 0, NULL, create_flag,
+			     flags, NULL);
+	if (handle != INVALID_HANDLE_VALUE) {
+		int new_fd = _open_osfhandle((intptr_t)handle, O_BINARY);
+		SetStdHandle(std_id, handle);
+		dup2(new_fd, fd);
+		close(new_fd);
+	}
+}
+
+static void maybe_redirect_std_handles(void)
+{
+	maybe_redirect_std_handle(L"GIT_REDIRECT_STDIN", STD_INPUT_HANDLE, 0,
+				  GENERIC_READ, FILE_ATTRIBUTE_NORMAL);
+	maybe_redirect_std_handle(L"GIT_REDIRECT_STDOUT", STD_OUTPUT_HANDLE, 1,
+				  GENERIC_WRITE, FILE_ATTRIBUTE_NORMAL);
+	maybe_redirect_std_handle(L"GIT_REDIRECT_STDERR", STD_ERROR_HANDLE, 2,
+				  GENERIC_WRITE, FILE_FLAG_NO_BUFFERING);
+}
+
+#ifdef _MSC_VER
+#ifdef _DEBUG
+#include <crtdbg.h>
+#endif
+#endif
+
+/*
+ * We implement wmain() and compile with -municode, which would
+ * normally ignore main(), but we call the latter from the former
+ * so that we can handle non-ASCII command-line parameters
+ * appropriately.
+ *
+ * To be more compatible with the core git code, we convert
+ * argv into UTF8 and pass them directly to main().
+ */
+int wmain(int argc, const wchar_t **wargv)
+{
+	int i, maxlen, exit_status;
+	char *buffer, **save;
+	const char **argv;
+
+	trace2_initialize_clock();
+
+#ifdef _MSC_VER
+#ifdef _DEBUG
+	_CrtSetReportMode(_CRT_ASSERT, _CRTDBG_MODE_DEBUG);
+#endif
+
+#ifdef USE_MSVC_CRTDBG
+	_CrtSetDbgFlag(_CRTDBG_ALLOC_MEM_DF | _CRTDBG_LEAK_CHECK_DF);
+#endif
+#endif
+
+	maybe_redirect_std_handles();
+
+	/* determine size of argv and environ conversion buffer */
+	maxlen = wcslen(wargv[0]);
+	for (i = 1; i < argc; i++)
+		maxlen = max(maxlen, wcslen(wargv[i]));
+
+	/* allocate buffer (wchar_t encodes to max 3 UTF-8 bytes) */
+	maxlen = 3 * maxlen + 1;
+	buffer = malloc_startup(maxlen);
+
+	/*
+	 * Create a UTF-8 version of w_argv. Also create a "save" copy
+	 * to remember all the string pointers because parse_options()
+	 * will remove claimed items from the argv that we pass down.
+	 */
+	ALLOC_ARRAY(argv, argc + 1);
+	ALLOC_ARRAY(save, argc + 1);
+	for (i = 0; i < argc; i++)
+		argv[i] = save[i] = wcstoutfdup_startup(buffer, wargv[i], maxlen);
+	argv[i] = save[i] = NULL;
+	free(buffer);
+
+	/* fix Windows specific environment settings */
+	setup_windows_environment();
+
+	unset_environment_variables = xstrdup("PERL5LIB");
+
+	/* initialize critical section for waitpid pinfo_t list */
+	InitializeCriticalSection(&pinfo_cs);
+
+	/* set up default file mode and file modes for stdin/out/err */
+	_fmode = _O_BINARY;
+	_setmode(_fileno(stdin), _O_BINARY);
+	_setmode(_fileno(stdout), _O_BINARY);
+	_setmode(_fileno(stderr), _O_BINARY);
+
+	/* initialize Unicode console */
+	winansi_init();
+
+	/* invoke the real main() using our utf8 version of argv. */
+	exit_status = main(argc, argv);
+
+	for (i = 0; i < argc; i++)
+		free(save[i]);
+	free(save);
+	free(argv);
+
+	return exit_status;
+}
+
+int uname(struct utsname *buf)
+{
+	unsigned v = (unsigned)GetVersion();
+	memset(buf, 0, sizeof(*buf));
+	xsnprintf(buf->sysname, sizeof(buf->sysname), "Windows");
+	xsnprintf(buf->release, sizeof(buf->release),
+		 "%u.%u", v & 0xff, (v >> 8) & 0xff);
+	/* assuming NT variants only.. */
+	xsnprintf(buf->version, sizeof(buf->version),
+		  "%u", (v >> 16) & 0x7fff);
+	return 0;
+}
diff --git a/compat/mingw.h b/compat/mingw.h
new file mode 100644
index 000000000000..a03e40e6e2a6
--- /dev/null
+++ b/compat/mingw.h
@@ -0,0 +1,583 @@
+#ifdef __MINGW64_VERSION_MAJOR
+#include <stdint.h>
+#include <wchar.h>
+typedef _sigset_t sigset_t;
+#endif
+#include <winsock2.h>
+#include <ws2tcpip.h>
+
+/* MinGW-w64 reports to have flockfile, but it does not actually have it. */
+#ifdef __MINGW64_VERSION_MAJOR
+#undef _POSIX_THREAD_SAFE_FUNCTIONS
+#endif
+
+extern int mingw_core_config(const char *var, const char *value, void *cb);
+#define platform_core_config mingw_core_config
+
+/*
+ * things that are not available in header files
+ */
+
+typedef int uid_t;
+typedef int socklen_t;
+#ifndef __MINGW64_VERSION_MAJOR
+typedef int pid_t;
+#define hstrerror strerror
+#endif
+
+#define S_IFLNK    0120000 /* Symbolic link */
+#define S_ISLNK(x) (((x) & S_IFMT) == S_IFLNK)
+#define S_ISSOCK(x) 0
+
+#ifndef S_IRWXG
+#define S_IRGRP 0
+#define S_IWGRP 0
+#define S_IXGRP 0
+#define S_IRWXG (S_IRGRP | S_IWGRP | S_IXGRP)
+#endif
+#ifndef S_IRWXO
+#define S_IROTH 0
+#define S_IWOTH 0
+#define S_IXOTH 0
+#define S_IRWXO (S_IROTH | S_IWOTH | S_IXOTH)
+#endif
+
+#define S_ISUID 0004000
+#define S_ISGID 0002000
+#define S_ISVTX 0001000
+
+#define WIFEXITED(x) 1
+#define WIFSIGNALED(x) 0
+#define WEXITSTATUS(x) ((x) & 0xff)
+#define WTERMSIG(x) SIGTERM
+
+#ifndef EWOULDBLOCK
+#define EWOULDBLOCK EAGAIN
+#endif
+#ifndef ELOOP
+#define ELOOP EMLINK
+#endif
+#define SHUT_WR SD_SEND
+
+#define SIGHUP 1
+#define SIGQUIT 3
+#define SIGKILL 9
+#define SIGPIPE 13
+#define SIGALRM 14
+#define SIGCHLD 17
+
+#define F_GETFD 1
+#define F_SETFD 2
+#define FD_CLOEXEC 0x1
+
+#if !defined O_CLOEXEC && defined O_NOINHERIT
+#define O_CLOEXEC	O_NOINHERIT
+#endif
+
+#ifndef EAFNOSUPPORT
+#define EAFNOSUPPORT WSAEAFNOSUPPORT
+#endif
+#ifndef ECONNABORTED
+#define ECONNABORTED WSAECONNABORTED
+#endif
+#ifndef ENOTSOCK
+#define ENOTSOCK WSAENOTSOCK
+#endif
+
+struct passwd {
+	char *pw_name;
+	char *pw_gecos;
+	char *pw_dir;
+};
+
+typedef void (__cdecl *sig_handler_t)(int);
+struct sigaction {
+	sig_handler_t sa_handler;
+	unsigned sa_flags;
+};
+#define SA_RESTART 0
+
+struct itimerval {
+	struct timeval it_value, it_interval;
+};
+#define ITIMER_REAL 0
+
+struct utsname {
+	char sysname[16];
+	char nodename[1];
+	char release[16];
+	char version[16];
+	char machine[1];
+};
+
+/*
+ * sanitize preprocessor namespace polluted by Windows headers defining
+ * macros which collide with git local versions
+ */
+#undef HELP_COMMAND /* from winuser.h */
+
+/*
+ * trivial stubs
+ */
+
+static inline int readlink(const char *path, char *buf, size_t bufsiz)
+{ errno = ENOSYS; return -1; }
+static inline int symlink(const char *oldpath, const char *newpath)
+{ errno = ENOSYS; return -1; }
+static inline int fchmod(int fildes, mode_t mode)
+{ errno = ENOSYS; return -1; }
+#ifndef __MINGW64_VERSION_MAJOR
+static inline pid_t fork(void)
+{ errno = ENOSYS; return -1; }
+#endif
+static inline unsigned int alarm(unsigned int seconds)
+{ return 0; }
+static inline int fsync(int fd)
+{ return _commit(fd); }
+static inline void sync(void)
+{}
+static inline uid_t getuid(void)
+{ return 1; }
+static inline struct passwd *getpwnam(const char *name)
+{ return NULL; }
+static inline int fcntl(int fd, int cmd, ...)
+{
+	if (cmd == F_GETFD || cmd == F_SETFD)
+		return 0;
+	errno = EINVAL;
+	return -1;
+}
+
+#define sigemptyset(x) (void)0
+static inline int sigaddset(sigset_t *set, int signum)
+{ return 0; }
+#define SIG_BLOCK 0
+#define SIG_UNBLOCK 0
+static inline int sigprocmask(int how, const sigset_t *set, sigset_t *oldset)
+{ return 0; }
+static inline pid_t getppid(void)
+{ return 1; }
+static inline pid_t getpgid(pid_t pid)
+{ return pid == 0 ? getpid() : pid; }
+static inline pid_t tcgetpgrp(int fd)
+{ return getpid(); }
+
+/*
+ * simple adaptors
+ */
+
+int mingw_mkdir(const char *path, int mode);
+#define mkdir mingw_mkdir
+
+#define WNOHANG 1
+pid_t waitpid(pid_t pid, int *status, int options);
+
+#define kill mingw_kill
+int mingw_kill(pid_t pid, int sig);
+
+#ifndef NO_OPENSSL
+#include <openssl/ssl.h>
+static inline int mingw_SSL_set_fd(SSL *ssl, int fd)
+{
+	return SSL_set_fd(ssl, _get_osfhandle(fd));
+}
+#define SSL_set_fd mingw_SSL_set_fd
+
+static inline int mingw_SSL_set_rfd(SSL *ssl, int fd)
+{
+	return SSL_set_rfd(ssl, _get_osfhandle(fd));
+}
+#define SSL_set_rfd mingw_SSL_set_rfd
+
+static inline int mingw_SSL_set_wfd(SSL *ssl, int fd)
+{
+	return SSL_set_wfd(ssl, _get_osfhandle(fd));
+}
+#define SSL_set_wfd mingw_SSL_set_wfd
+#endif
+
+/*
+ * implementations of missing functions
+ */
+
+int pipe(int filedes[2]);
+unsigned int sleep (unsigned int seconds);
+int mkstemp(char *template);
+int gettimeofday(struct timeval *tv, void *tz);
+#ifndef __MINGW64_VERSION_MAJOR
+struct tm *gmtime_r(const time_t *timep, struct tm *result);
+struct tm *localtime_r(const time_t *timep, struct tm *result);
+#endif
+int getpagesize(void);	/* defined in MinGW's libgcc.a */
+struct passwd *getpwuid(uid_t uid);
+int setitimer(int type, struct itimerval *in, struct itimerval *out);
+int sigaction(int sig, struct sigaction *in, struct sigaction *out);
+int link(const char *oldpath, const char *newpath);
+int uname(struct utsname *buf);
+
+/*
+ * replacements of existing functions
+ */
+
+int mingw_unlink(const char *pathname);
+#define unlink mingw_unlink
+
+int mingw_rmdir(const char *path);
+#define rmdir mingw_rmdir
+
+int mingw_open (const char *filename, int oflags, ...);
+#define open mingw_open
+
+int mingw_fgetc(FILE *stream);
+#define fgetc mingw_fgetc
+
+FILE *mingw_fopen (const char *filename, const char *otype);
+#define fopen mingw_fopen
+
+FILE *mingw_freopen (const char *filename, const char *otype, FILE *stream);
+#define freopen mingw_freopen
+
+int mingw_fflush(FILE *stream);
+#define fflush mingw_fflush
+
+ssize_t mingw_write(int fd, const void *buf, size_t len);
+#define write mingw_write
+
+int mingw_access(const char *filename, int mode);
+#undef access
+#define access mingw_access
+
+int mingw_chdir(const char *dirname);
+#define chdir mingw_chdir
+
+int mingw_chmod(const char *filename, int mode);
+#define chmod mingw_chmod
+
+char *mingw_mktemp(char *template);
+#define mktemp mingw_mktemp
+
+char *mingw_getcwd(char *pointer, int len);
+#define getcwd mingw_getcwd
+
+#ifdef NO_UNSETENV
+#error "NO_UNSETENV is incompatible with the Windows-specific startup code!"
+#endif
+
+/*
+ * We bind *env() routines (even the mingw_ ones) to private mingw_ versions.
+ * These talk to the CRT using UNICODE/wchar_t, but maintain the original
+ * narrow-char API.
+ *
+ * Note that the MSCRT maintains both ANSI (getenv()) and UNICODE (_wgetenv())
+ * routines and stores both versions of each environment variable in parallel
+ * (and secretly updates both when you set one or the other), but it uses CP_ACP
+ * to do the conversion rather than CP_UTF8.
+ *
+ * Since everything in the git code base is UTF8, we define the mingw_ routines
+ * to access the CRT using the UNICODE routines and manually convert them to
+ * UTF8.  This also avoids round-trip problems.
+ *
+ * This also helps with our linkage, since "_wenviron" is publicly exported
+ * from the CRT.  But to access "_environ" we would have to statically link
+ * to the CRT (/MT).
+ *
+ * We require NO_SETENV (and let gitsetenv() call our mingw_putenv).
+ */
+#define getenv       mingw_getenv
+#define putenv       mingw_putenv
+#define unsetenv     mingw_putenv
+char *mingw_getenv(const char *name);
+int   mingw_putenv(const char *name);
+
+int mingw_gethostname(char *host, int namelen);
+#define gethostname mingw_gethostname
+
+struct hostent *mingw_gethostbyname(const char *host);
+#define gethostbyname mingw_gethostbyname
+
+int mingw_getaddrinfo(const char *node, const char *service,
+		      const struct addrinfo *hints, struct addrinfo **res);
+#define getaddrinfo mingw_getaddrinfo
+
+int mingw_socket(int domain, int type, int protocol);
+#define socket mingw_socket
+
+int mingw_connect(int sockfd, struct sockaddr *sa, size_t sz);
+#define connect mingw_connect
+
+int mingw_bind(int sockfd, struct sockaddr *sa, size_t sz);
+#define bind mingw_bind
+
+int mingw_setsockopt(int sockfd, int lvl, int optname, void *optval, int optlen);
+#define setsockopt mingw_setsockopt
+
+int mingw_shutdown(int sockfd, int how);
+#define shutdown mingw_shutdown
+
+int mingw_listen(int sockfd, int backlog);
+#define listen mingw_listen
+
+int mingw_accept(int sockfd, struct sockaddr *sa, socklen_t *sz);
+#define accept mingw_accept
+
+int mingw_rename(const char*, const char*);
+#define rename mingw_rename
+
+#if defined(USE_WIN32_MMAP) || defined(_MSC_VER)
+int mingw_getpagesize(void);
+#define getpagesize mingw_getpagesize
+#endif
+
+struct rlimit {
+	unsigned int rlim_cur;
+};
+#define RLIMIT_NOFILE 0
+
+static inline int getrlimit(int resource, struct rlimit *rlp)
+{
+	if (resource != RLIMIT_NOFILE) {
+		errno = EINVAL;
+		return -1;
+	}
+
+	rlp->rlim_cur = 2048;
+	return 0;
+}
+
+/*
+ * Use mingw specific stat()/lstat()/fstat() implementations on Windows,
+ * including our own struct stat with 64 bit st_size and nanosecond-precision
+ * file times.
+ */
+#ifndef __MINGW64_VERSION_MAJOR
+#define off_t off64_t
+#define lseek _lseeki64
+#ifndef _MSC_VER
+struct timespec {
+	time_t tv_sec;
+	long tv_nsec;
+};
+#endif
+#endif
+
+struct mingw_stat {
+    _dev_t st_dev;
+    _ino_t st_ino;
+    _mode_t st_mode;
+    short st_nlink;
+    short st_uid;
+    short st_gid;
+    _dev_t st_rdev;
+    off64_t st_size;
+    struct timespec st_atim;
+    struct timespec st_mtim;
+    struct timespec st_ctim;
+};
+
+#define st_atime st_atim.tv_sec
+#define st_mtime st_mtim.tv_sec
+#define st_ctime st_ctim.tv_sec
+
+#ifdef stat
+#undef stat
+#endif
+#define stat mingw_stat
+int mingw_lstat(const char *file_name, struct stat *buf);
+int mingw_stat(const char *file_name, struct stat *buf);
+int mingw_fstat(int fd, struct stat *buf);
+#ifdef fstat
+#undef fstat
+#endif
+#define fstat mingw_fstat
+#ifdef lstat
+#undef lstat
+#endif
+#define lstat mingw_lstat
+
+
+int mingw_utime(const char *file_name, const struct utimbuf *times);
+#define utime mingw_utime
+size_t mingw_strftime(char *s, size_t max,
+		   const char *format, const struct tm *tm);
+#define strftime mingw_strftime
+
+pid_t mingw_spawnvpe(const char *cmd, const char **argv, char **env,
+		     const char *dir,
+		     int fhin, int fhout, int fherr);
+int mingw_execvp(const char *cmd, char *const *argv);
+#define execvp mingw_execvp
+int mingw_execv(const char *cmd, char *const *argv);
+#define execv mingw_execv
+
+static inline unsigned int git_ntohl(unsigned int x)
+{ return (unsigned int)ntohl(x); }
+#define ntohl git_ntohl
+
+sig_handler_t mingw_signal(int sig, sig_handler_t handler);
+#define signal mingw_signal
+
+int mingw_raise(int sig);
+#define raise mingw_raise
+
+/*
+ * ANSI emulation wrappers
+ */
+
+int winansi_isatty(int fd);
+#define isatty winansi_isatty
+
+int winansi_dup2(int oldfd, int newfd);
+#define dup2 winansi_dup2
+
+void winansi_init(void);
+HANDLE winansi_get_osfhandle(int fd);
+
+/*
+ * git specific compatibility
+ */
+
+static inline void convert_slashes(char *path)
+{
+	for (; *path; path++)
+		if (*path == '\\')
+			*path = '/';
+}
+#define PATH_SEP ';'
+extern char *mingw_query_user_email(void);
+#define query_user_email mingw_query_user_email
+#if !defined(__MINGW64_VERSION_MAJOR) && (!defined(_MSC_VER) || _MSC_VER < 1800)
+#define PRIuMAX "I64u"
+#define PRId64 "I64d"
+#else
+#include <inttypes.h>
+#endif
+
+/**
+ * Converts UTF-8 encoded string to UTF-16LE.
+ *
+ * To support repositories with legacy-encoded file names, invalid UTF-8 bytes
+ * 0xa0 - 0xff are converted to corresponding printable Unicode chars \u00a0 -
+ * \u00ff, and invalid UTF-8 bytes 0x80 - 0x9f (which would make non-printable
+ * Unicode) are converted to hex-code.
+ *
+ * Lead-bytes not followed by an appropriate number of trail-bytes, over-long
+ * encodings and 4-byte encodings > \u10ffff are detected as invalid UTF-8.
+ *
+ * Maximum space requirement for the target buffer is two wide chars per UTF-8
+ * char (((strlen(utf) * 2) + 1) [* sizeof(wchar_t)]).
+ *
+ * The maximum space is needed only if the entire input string consists of
+ * invalid UTF-8 bytes in range 0x80-0x9f, as per the following table:
+ *
+ *               |                   | UTF-8 | UTF-16 |
+ *   Code point  |  UTF-8 sequence   | bytes | words  | ratio
+ * --------------+-------------------+-------+--------+-------
+ * 000000-00007f | 0-7f              |   1   |   1    |  1
+ * 000080-0007ff | c2-df + 80-bf     |   2   |   1    |  0.5
+ * 000800-00ffff | e0-ef + 2 * 80-bf |   3   |   1    |  0.33
+ * 010000-10ffff | f0-f4 + 3 * 80-bf |   4   |  2 (a) |  0.5
+ * invalid       | 80-9f             |   1   |  2 (b) |  2
+ * invalid       | a0-ff             |   1   |   1    |  1
+ *
+ * (a) encoded as UTF-16 surrogate pair
+ * (b) encoded as two hex digits
+ *
+ * Note that, while the UTF-8 encoding scheme can be extended to 5-byte, 6-byte
+ * or even indefinite-byte sequences, the largest valid code point \u10ffff
+ * encodes as only 4 UTF-8 bytes.
+ *
+ * Parameters:
+ * wcs: wide char target buffer
+ * utf: string to convert
+ * wcslen: size of target buffer (in wchar_t's)
+ * utflen: size of string to convert, or -1 if 0-terminated
+ *
+ * Returns:
+ * length of converted string (_wcslen(wcs)), or -1 on failure
+ *
+ * Errors:
+ * EINVAL: one of the input parameters is invalid (e.g. NULL)
+ * ERANGE: the output buffer is too small
+ */
+int xutftowcsn(wchar_t *wcs, const char *utf, size_t wcslen, int utflen);
+
+/**
+ * Simplified variant of xutftowcsn, assumes input string is \0-terminated.
+ */
+static inline int xutftowcs(wchar_t *wcs, const char *utf, size_t wcslen)
+{
+	return xutftowcsn(wcs, utf, wcslen, -1);
+}
+
+/**
+ * Simplified file system specific variant of xutftowcsn, assumes output
+ * buffer size is MAX_PATH wide chars and input string is \0-terminated,
+ * fails with ENAMETOOLONG if input string is too long.
+ */
+static inline int xutftowcs_path(wchar_t *wcs, const char *utf)
+{
+	int result = xutftowcsn(wcs, utf, MAX_PATH, -1);
+	if (result < 0 && errno == ERANGE)
+		errno = ENAMETOOLONG;
+	return result;
+}
+
+/**
+ * Converts UTF-16LE encoded string to UTF-8.
+ *
+ * Maximum space requirement for the target buffer is three UTF-8 chars per
+ * wide char ((_wcslen(wcs) * 3) + 1).
+ *
+ * The maximum space is needed only if the entire input string consists of
+ * UTF-16 words in range 0x0800-0xd7ff or 0xe000-0xffff (i.e. \u0800-\uffff
+ * modulo surrogate pairs), as per the following table:
+ *
+ *               |                       | UTF-16 | UTF-8 |
+ *   Code point  |  UTF-16 sequence      | words  | bytes | ratio
+ * --------------+-----------------------+--------+-------+-------
+ * 000000-00007f | 0000-007f             |   1    |   1   |  1
+ * 000080-0007ff | 0080-07ff             |   1    |   2   |  2
+ * 000800-00ffff | 0800-d7ff / e000-ffff |   1    |   3   |  3
+ * 010000-10ffff | d800-dbff + dc00-dfff |   2    |   4   |  2
+ *
+ * Note that invalid code points > 10ffff cannot be represented in UTF-16.
+ *
+ * Parameters:
+ * utf: target buffer
+ * wcs: wide string to convert
+ * utflen: size of target buffer
+ *
+ * Returns:
+ * length of converted string, or -1 on failure
+ *
+ * Errors:
+ * EINVAL: one of the input parameters is invalid (e.g. NULL)
+ * ERANGE: the output buffer is too small
+ */
+int xwcstoutf(char *utf, const wchar_t *wcs, size_t utflen);
+
+/*
+ * A critical section used in the implementation of the spawn
+ * functions (mingw_spawnv[p]e()) and waitpid(). Intialised in
+ * the replacement main() macro below.
+ */
+extern CRITICAL_SECTION pinfo_cs;
+
+/*
+ * Git, like most portable C applications, implements a main() function. On
+ * Windows, this main() function would receive parameters encoded in the
+ * current locale, but Git for Windows would prefer UTF-8 encoded  parameters.
+ *
+ * To make that happen, we still declare main() here, and then declare and
+ * implement wmain() (which is the Unicode variant of main()) and compile with
+ * -municode. This wmain() function reencodes the parameters from UTF-16 to
+ * UTF-8 format, sets up a couple of other things as required on Windows, and
+ * then hands off to the main() function.
+ */
+int wmain(int argc, const wchar_t **w_argv);
+int main(int argc, const char **argv);
+
+/*
+ * Used by Pthread API implementation for Windows
+ */
+extern int err_win_to_posix(DWORD winerr);
diff --git a/compat/mkdir.c b/compat/mkdir.c
new file mode 100644
index 000000000000..9e253fb72f2d
--- /dev/null
+++ b/compat/mkdir.c
@@ -0,0 +1,24 @@
+#include "../git-compat-util.h"
+#undef mkdir
+
+/* for platforms that can't deal with a trailing '/' */
+int compat_mkdir_wo_trailing_slash(const char *dir, mode_t mode)
+{
+	int retval;
+	char *tmp_dir = NULL;
+	size_t len = strlen(dir);
+
+	if (len && dir[len-1] == '/') {
+		if ((tmp_dir = strdup(dir)) == NULL)
+			return -1;
+		tmp_dir[len-1] = '\0';
+	}
+	else
+		tmp_dir = (char *)dir;
+
+	retval = mkdir(tmp_dir, mode);
+	if (tmp_dir != dir)
+		free(tmp_dir);
+
+	return retval;
+}
diff --git a/compat/mkdtemp.c b/compat/mkdtemp.c
new file mode 100644
index 000000000000..11361195925c
--- /dev/null
+++ b/compat/mkdtemp.c
@@ -0,0 +1,8 @@
+#include "../git-compat-util.h"
+
+char *gitmkdtemp(char *template)
+{
+	if (!*mktemp(template) || mkdir(template, 0700))
+		return NULL;
+	return template;
+}
diff --git a/compat/mmap.c b/compat/mmap.c
new file mode 100644
index 000000000000..14d31010dfe5
--- /dev/null
+++ b/compat/mmap.c
@@ -0,0 +1,40 @@
+#include "../git-compat-util.h"
+
+void *git_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset)
+{
+	size_t n = 0;
+
+	if (start != NULL || flags != MAP_PRIVATE || prot != PROT_READ)
+		die("Invalid usage of mmap when built with NO_MMAP");
+
+	start = xmalloc(length);
+	if (start == NULL) {
+		errno = ENOMEM;
+		return MAP_FAILED;
+	}
+
+	while (n < length) {
+		ssize_t count = xpread(fd, (char *)start + n, length - n, offset + n);
+
+		if (count == 0) {
+			memset((char *)start+n, 0, length-n);
+			break;
+		}
+
+		if (count < 0) {
+			free(start);
+			errno = EACCES;
+			return MAP_FAILED;
+		}
+
+		n += count;
+	}
+
+	return start;
+}
+
+int git_munmap(void *start, size_t length)
+{
+	free(start);
+	return 0;
+}
diff --git a/compat/msvc.c b/compat/msvc.c
new file mode 100644
index 000000000000..71843d7eefaf
--- /dev/null
+++ b/compat/msvc.c
@@ -0,0 +1,6 @@
+#include "../git-compat-util.h"
+#include "win32.h"
+#include <conio.h>
+#include "../strbuf.h"
+
+#include "mingw.c"
diff --git a/compat/msvc.h b/compat/msvc.h
new file mode 100644
index 000000000000..1d7a8c614565
--- /dev/null
+++ b/compat/msvc.h
@@ -0,0 +1,33 @@
+#ifndef __MSVC__HEAD
+#define __MSVC__HEAD
+
+#include <direct.h>
+#include <process.h>
+#include <malloc.h>
+#include <io.h>
+
+#pragma warning(disable: 4018) /* signed/unsigned comparison */
+#pragma warning(disable: 4244) /* type conversion, possible loss of data */
+#pragma warning(disable: 4090) /* 'function' : different 'const' qualifiers (ALLOC_GROW etc.)*/
+
+/* porting function */
+#define inline __inline
+#define __inline__ __inline
+#define __attribute__(x)
+#define strcasecmp   _stricmp
+#define strncasecmp  _strnicmp
+#define ftruncate    _chsize
+#define strtoull     _strtoui64
+#define strtoll      _strtoi64
+
+#undef ERROR
+
+#define ftello _ftelli64
+
+typedef int sigset_t;
+/* open for reading, writing, or both (not in fcntl.h) */
+#define O_ACCMODE     (_O_RDONLY | _O_WRONLY | _O_RDWR)
+
+#include "compat/mingw.h"
+
+#endif
diff --git a/compat/nedmalloc/License.txt b/compat/nedmalloc/License.txt
new file mode 100644
index 000000000000..36b7cd93cdfb
--- /dev/null
+++ b/compat/nedmalloc/License.txt
@@ -0,0 +1,23 @@
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
diff --git a/compat/nedmalloc/Readme.txt b/compat/nedmalloc/Readme.txt
new file mode 100644
index 000000000000..07cbf50c0f9a
--- /dev/null
+++ b/compat/nedmalloc/Readme.txt
@@ -0,0 +1,136 @@
+nedalloc v1.05 15th June 2008:
+-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
+
+by Niall Douglas (http://www.nedprod.com/programs/portable/nedmalloc/)
+
+Enclosed is nedalloc, an alternative malloc implementation for multiple
+threads without lock contention based on dlmalloc v2.8.4. It is more
+or less a newer implementation of ptmalloc2, the standard allocator in
+Linux (which is based on dlmalloc v2.7.0) but also contains a per-thread
+cache for maximum CPU scalability.
+
+It is licensed under the Boost Software License which basically means
+you can do anything you like with it. This does not apply to the malloc.c.h
+file which remains copyright to others.
+
+It has been tested on win32 (x86), win64 (x64), Linux (x64), FreeBSD (x64)
+and Apple MacOS X (x86). It works very well on all of these and is very
+significantly faster than the system allocator on all of these platforms.
+
+By literally dropping in this allocator as a replacement for your system
+allocator, you can see real world improvements of up to three times in normal
+code!
+
+To use:
+-=-=-=-
+Drop in nedmalloc.h, nedmalloc.c and malloc.c.h into your project.
+Configure using the instructions in nedmalloc.h. Run and enjoy.
+
+To test, compile test.c. It will run a comparison between your system
+allocator and nedalloc and tell you how much faster nedalloc is. It also
+serves as an example of usage.
+
+Notes:
+-=-=-=
+If you want the very latest version of this allocator, get it from the
+TnFOX SVN repository at svn://svn.berlios.de/viewcvs/tnfox/trunk/src/nedmalloc
+
+Because of how nedalloc allocates an mspace per thread, it can cause
+severe bloating of memory usage under certain allocation patterns.
+You can substantially reduce this wastage by setting MAXTHREADSINPOOL
+or the threads parameter to nedcreatepool() to a fraction of the number of
+threads which would normally be in a pool at once. This will reduce
+bloating at the cost of an increase in lock contention. If allocated size
+is less than THREADCACHEMAX, locking is avoided 90-99% of the time and
+if most of your allocations are below this value, you can safely set
+MAXTHREADSINPOOL to one.
+
+You will suffer memory leakage unless you call neddisablethreadcache()
+per pool for every thread which exits. This is because nedalloc cannot
+portably know when a thread exits and thus when its thread cache can
+be returned for use by other code. Don't forget pool zero, the system pool.
+
+For C++ type allocation patterns (where the same sizes of memory are
+regularly allocated and deallocated as objects are created and destroyed),
+the threadcache always benefits performance. If however your allocation
+patterns are different, searching the threadcache may significantly slow
+down your code - as a rule of thumb, if cache utilisation is below 80%
+(see the source for neddisablethreadcache() for how to enable debug
+printing in release mode) then you should disable the thread cache for
+that thread. You can compile out the threadcache code by setting
+THREADCACHEMAX to zero.
+
+Speed comparisons:
+-=-=-=-=-=-=-=-=-=
+See Benchmarks.xls for details.
+
+The enclosed test.c can do two things: it can be a torture test or a speed
+test. The speed test is designed to be a representative synthetic
+memory allocator test. It works by randomly mixing allocations with frees
+with half of the allocation sizes being a two power multiple less than
+512 bytes (to mimic C++ stack instantiated objects) and the other half
+being a simple random value less than 16Kb.
+
+The real world code results are from Tn's TestIO benchmark. This is a
+heavily multithreaded and memory intensive benchmark with a lot of branching
+and other stuff modern processors don't like so much. As you'll note, the
+test doesn't show the benefits of the threadcache mostly due to the saturation
+of the memory bus being the limiting factor.
+
+ChangeLog:
+-=-=-=-=-=
+v1.05 15th June 2008:
+ * { 1042 } Added error check for TLSSET() and TLSFREE() macros. Thanks to
+Markus Elfring for reporting this.
+ * { 1043 } Fixed a segfault when freeing memory allocated using
+nedindependent_comalloc(). Thanks to Pavel Vozenilek for reporting this.
+
+v1.04 14th July 2007:
+ * Fixed a bug with the new optimised implementation that failed to lock
+on a realloc under certain conditions.
+ * Fixed lack of thread synchronisation in InitPool() causing pool corruption
+ * Fixed a memory leak of thread cache contents on disabling. Thanks to Earl
+Chew for reporting this.
+ * Added a sanity check for freed blocks being valid.
+ * Reworked test.c into being a torture test.
+ * Fixed GCC assembler optimisation misspecification
+
+v1.04alpha_svn915 7th October 2006:
+ * Fixed failure to unlock thread cache list if allocating a new list failed.
+Thanks to Dmitry Chichkov for reporting this. Further thanks to Aleksey Sanin.
+ * Fixed realloc(0, <size>) segfaulting. Thanks to Dmitry Chichkov for
+reporting this.
+ * Made config defines #ifndef so they can be overridden by the build system.
+Thanks to Aleksey Sanin for suggesting this.
+ * Fixed deadlock in nedprealloc() due to unnecessary locking of preferred
+thread mspace when mspace_realloc() always uses the original block's mspace
+anyway. Thanks to Aleksey Sanin for reporting this.
+ * Made some speed improvements by hacking mspace_malloc() to no longer lock
+its mspace, thus allowing the recursive mutex implementation to be removed
+with an associated speed increase. Thanks to Aleksey Sanin for suggesting this.
+ * Fixed a bug where allocating mspaces overran its max limit. Thanks to
+Aleksey Sanin for reporting this.
+
+v1.03 10th July 2006:
+ * Fixed memory corruption bug in threadcache code which only appeared with >4
+threads and in heavy use of the threadcache.
+
+v1.02 15th May 2006:
+ * Integrated dlmalloc v2.8.4, fixing the win32 memory release problem and
+improving performance still further. Speed is now up to twice the speed of v1.01
+(average is 67% faster).
+ * Fixed win32 critical section implementation. Thanks to Pavel Kuznetsov
+for reporting this.
+ * Wasn't locking mspace if all mspaces were locked. Thanks to Pavel Kuznetsov
+for reporting this.
+ * Added Apple Mac OS X support.
+
+v1.01 24th February 2006:
+ * Fixed multiprocessor scaling problems by removing sources of cache sloshing
+ * Earl Chew <earl_chew <at> agilent <dot> com> sent patches for the following:
+   1. size2binidx() wasn't working for default code path (non x86)
+   2. Fixed failure to release mspace lock under certain circumstances which
+      caused a deadlock
+
+v1.00 1st January 2006:
+ * First release
diff --git a/compat/nedmalloc/malloc.c.h b/compat/nedmalloc/malloc.c.h
new file mode 100644
index 000000000000..b833ff9225e7
--- /dev/null
+++ b/compat/nedmalloc/malloc.c.h
@@ -0,0 +1,5761 @@
+/*
+  This is a version (aka dlmalloc) of malloc/free/realloc written by
+  Doug Lea and released to the public domain, as explained at
+  http://creativecommons.org/licenses/publicdomain.  Send questions,
+  comments, complaints, performance data, etc to dl@cs.oswego.edu
+
+* Version pre-2.8.4 Mon Nov 27 11:22:37 2006    (dl at gee)
+
+   Note: There may be an updated version of this malloc obtainable at
+	   ftp://gee.cs.oswego.edu/pub/misc/malloc.c
+	 Check before installing!
+
+* Quickstart
+
+  This library is all in one file to simplify the most common usage:
+  ftp it, compile it (-O3), and link it into another program. All of
+  the compile-time options default to reasonable values for use on
+  most platforms.  You might later want to step through various
+  compile-time and dynamic tuning options.
+
+  For convenience, an include file for code using this malloc is at:
+     ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.4.h
+  You don't really need this .h file unless you call functions not
+  defined in your system include files.  The .h file contains only the
+  excerpts from this file needed for using this malloc on ANSI C/C++
+  systems, so long as you haven't changed compile-time options about
+  naming and tuning parameters.  If you do, then you can create your
+  own malloc.h that does include all settings by cutting at the point
+  indicated below. Note that you may already by default be using a C
+  library containing a malloc that is based on some version of this
+  malloc (for example in linux). You might still want to use the one
+  in this file to customize settings or to avoid overheads associated
+  with library versions.
+
+* Vital statistics:
+
+  Supported pointer/size_t representation:       4 or 8 bytes
+       size_t MUST be an unsigned type of the same width as
+       pointers. (If you are using an ancient system that declares
+       size_t as a signed type, or need it to be a different width
+       than pointers, you can use a previous release of this malloc
+       (e.g. 2.7.2) supporting these.)
+
+  Alignment:                                     8 bytes (default)
+       This suffices for nearly all current machines and C compilers.
+       However, you can define MALLOC_ALIGNMENT to be wider than this
+       if necessary (up to 128bytes), at the expense of using more space.
+
+  Minimum overhead per allocated chunk:   4 or  8 bytes (if 4byte sizes)
+					  8 or 16 bytes (if 8byte sizes)
+       Each malloced chunk has a hidden word of overhead holding size
+       and status information, and additional cross-check word
+       if FOOTERS is defined.
+
+  Minimum allocated size: 4-byte ptrs:  16 bytes    (including overhead)
+			  8-byte ptrs:  32 bytes    (including overhead)
+
+       Even a request for zero bytes (i.e., malloc(0)) returns a
+       pointer to something of the minimum allocatable size.
+       The maximum overhead wastage (i.e., number of extra bytes
+       allocated than were requested in malloc) is less than or equal
+       to the minimum size, except for requests >= mmap_threshold that
+       are serviced via mmap(), where the worst case wastage is about
+       32 bytes plus the remainder from a system page (the minimal
+       mmap unit); typically 4096 or 8192 bytes.
+
+  Security: static-safe; optionally more or less
+       The "security" of malloc refers to the ability of malicious
+       code to accentuate the effects of errors (for example, freeing
+       space that is not currently malloc'ed or overwriting past the
+       ends of chunks) in code that calls malloc.  This malloc
+       guarantees not to modify any memory locations below the base of
+       heap, i.e., static variables, even in the presence of usage
+       errors.  The routines additionally detect most improper frees
+       and reallocs.  All this holds as long as the static bookkeeping
+       for malloc itself is not corrupted by some other means.  This
+       is only one aspect of security -- these checks do not, and
+       cannot, detect all possible programming errors.
+
+       If FOOTERS is defined nonzero, then each allocated chunk
+       carries an additional check word to verify that it was malloced
+       from its space.  These check words are the same within each
+       execution of a program using malloc, but differ across
+       executions, so externally crafted fake chunks cannot be
+       freed. This improves security by rejecting frees/reallocs that
+       could corrupt heap memory, in addition to the checks preventing
+       writes to statics that are always on.  This may further improve
+       security at the expense of time and space overhead.  (Note that
+       FOOTERS may also be worth using with MSPACES.)
+
+       By default detected errors cause the program to abort (calling
+       "abort()"). You can override this to instead proceed past
+       errors by defining PROCEED_ON_ERROR.  In this case, a bad free
+       has no effect, and a malloc that encounters a bad address
+       caused by user overwrites will ignore the bad address by
+       dropping pointers and indices to all known memory. This may
+       be appropriate for programs that should continue if at all
+       possible in the face of programming errors, although they may
+       run out of memory because dropped memory is never reclaimed.
+
+       If you don't like either of these options, you can define
+       CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
+       else. And if you are sure that your program using malloc has
+       no errors or vulnerabilities, you can define INSECURE to 1,
+       which might (or might not) provide a small performance improvement.
+
+  Thread-safety: NOT thread-safe unless USE_LOCKS defined
+       When USE_LOCKS is defined, each public call to malloc, free,
+       etc is surrounded with either a pthread mutex or a win32
+       spinlock (depending on WIN32). This is not especially fast, and
+       can be a major bottleneck.  It is designed only to provide
+       minimal protection in concurrent environments, and to provide a
+       basis for extensions.  If you are using malloc in a concurrent
+       program, consider instead using nedmalloc
+       (http://www.nedprod.com/programs/portable/nedmalloc/) or
+       ptmalloc (See http://www.malloc.de), which are derived
+       from versions of this malloc.
+
+  System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
+       This malloc can use unix sbrk or any emulation (invoked using
+       the CALL_MORECORE macro) and/or mmap/munmap or any emulation
+       (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
+       memory.  On most unix systems, it tends to work best if both
+       MORECORE and MMAP are enabled.  On Win32, it uses emulations
+       based on VirtualAlloc. It also uses common C library functions
+       like memset.
+
+  Compliance: I believe it is compliant with the Single Unix Specification
+       (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
+       others as well.
+
+* Overview of algorithms
+
+  This is not the fastest, most space-conserving, most portable, or
+  most tunable malloc ever written. However it is among the fastest
+  while also being among the most space-conserving, portable and
+  tunable.  Consistent balance across these factors results in a good
+  general-purpose allocator for malloc-intensive programs.
+
+  In most ways, this malloc is a best-fit allocator. Generally, it
+  chooses the best-fitting existing chunk for a request, with ties
+  broken in approximately least-recently-used order. (This strategy
+  normally maintains low fragmentation.) However, for requests less
+  than 256bytes, it deviates from best-fit when there is not an
+  exactly fitting available chunk by preferring to use space adjacent
+  to that used for the previous small request, as well as by breaking
+  ties in approximately most-recently-used order. (These enhance
+  locality of series of small allocations.)  And for very large requests
+  (>= 256Kb by default), it relies on system memory mapping
+  facilities, if supported.  (This helps avoid carrying around and
+  possibly fragmenting memory used only for large chunks.)
+
+  All operations (except malloc_stats and mallinfo) have execution
+  times that are bounded by a constant factor of the number of bits in
+  a size_t, not counting any clearing in calloc or copying in realloc,
+  or actions surrounding MORECORE and MMAP that have times
+  proportional to the number of non-contiguous regions returned by
+  system allocation routines, which is often just 1. In real-time
+  applications, you can optionally suppress segment traversals using
+  NO_SEGMENT_TRAVERSAL, which assures bounded execution even when
+  system allocators return non-contiguous spaces, at the typical
+  expense of carrying around more memory and increased fragmentation.
+
+  The implementation is not very modular and seriously overuses
+  macros. Perhaps someday all C compilers will do as good a job
+  inlining modular code as can now be done by brute-force expansion,
+  but now, enough of them seem not to.
+
+  Some compilers issue a lot of warnings about code that is
+  dead/unreachable only on some platforms, and also about intentional
+  uses of negation on unsigned types. All known cases of each can be
+  ignored.
+
+  For a longer but out of date high-level description, see
+     http://gee.cs.oswego.edu/dl/html/malloc.html
+
+* MSPACES
+  If MSPACES is defined, then in addition to malloc, free, etc.,
+  this file also defines mspace_malloc, mspace_free, etc. These
+  are versions of malloc routines that take an "mspace" argument
+  obtained using create_mspace, to control all internal bookkeeping.
+  If ONLY_MSPACES is defined, only these versions are compiled.
+  So if you would like to use this allocator for only some allocations,
+  and your system malloc for others, you can compile with
+  ONLY_MSPACES and then do something like...
+    static mspace mymspace = create_mspace(0,0); // for example
+    #define mymalloc(bytes)  mspace_malloc(mymspace, bytes)
+
+  (Note: If you only need one instance of an mspace, you can instead
+  use "USE_DL_PREFIX" to relabel the global malloc.)
+
+  You can similarly create thread-local allocators by storing
+  mspaces as thread-locals. For example:
+    static __thread mspace tlms = 0;
+    void*  tlmalloc(size_t bytes) {
+      if (tlms == 0) tlms = create_mspace(0, 0);
+      return mspace_malloc(tlms, bytes);
+    }
+    void  tlfree(void* mem) { mspace_free(tlms, mem); }
+
+  Unless FOOTERS is defined, each mspace is completely independent.
+  You cannot allocate from one and free to another (although
+  conformance is only weakly checked, so usage errors are not always
+  caught). If FOOTERS is defined, then each chunk carries around a tag
+  indicating its originating mspace, and frees are directed to their
+  originating spaces.
+
+ -------------------------  Compile-time options ---------------------------
+
+Be careful in setting #define values for numerical constants of type
+size_t. On some systems, literal values are not automatically extended
+to size_t precision unless they are explicitly casted. You can also
+use the symbolic values MAX_SIZE_T, SIZE_T_ONE, etc below.
+
+WIN32                    default: defined if _WIN32 defined
+  Defining WIN32 sets up defaults for MS environment and compilers.
+  Otherwise defaults are for unix. Beware that there seem to be some
+  cases where this malloc might not be a pure drop-in replacement for
+  Win32 malloc: Random-looking failures from Win32 GDI API's (eg;
+  SetDIBits()) may be due to bugs in some video driver implementations
+  when pixel buffers are malloc()ed, and the region spans more than
+  one VirtualAlloc()ed region. Because dlmalloc uses a small (64Kb)
+  default granularity, pixel buffers may straddle virtual allocation
+  regions more often than when using the Microsoft allocator.  You can
+  avoid this by using VirtualAlloc() and VirtualFree() for all pixel
+  buffers rather than using malloc().  If this is not possible,
+  recompile this malloc with a larger DEFAULT_GRANULARITY.
+
+MALLOC_ALIGNMENT         default: (size_t)8
+  Controls the minimum alignment for malloc'ed chunks.  It must be a
+  power of two and at least 8, even on machines for which smaller
+  alignments would suffice. It may be defined as larger than this
+  though. Note however that code and data structures are optimized for
+  the case of 8-byte alignment.
+
+MSPACES                  default: 0 (false)
+  If true, compile in support for independent allocation spaces.
+  This is only supported if HAVE_MMAP is true.
+
+ONLY_MSPACES             default: 0 (false)
+  If true, only compile in mspace versions, not regular versions.
+
+USE_LOCKS                default: 0 (false)
+  Causes each call to each public routine to be surrounded with
+  pthread or WIN32 mutex lock/unlock. (If set true, this can be
+  overridden on a per-mspace basis for mspace versions.) If set to a
+  non-zero value other than 1, locks are used, but their
+  implementation is left out, so lock functions must be supplied manually.
+
+USE_SPIN_LOCKS           default: 1 iff USE_LOCKS and on x86 using gcc or MSC
+  If true, uses custom spin locks for locking. This is currently
+  supported only for x86 platforms using gcc or recent MS compilers.
+  Otherwise, posix locks or win32 critical sections are used.
+
+FOOTERS                  default: 0
+  If true, provide extra checking and dispatching by placing
+  information in the footers of allocated chunks. This adds
+  space and time overhead.
+
+INSECURE                 default: 0
+  If true, omit checks for usage errors and heap space overwrites.
+
+USE_DL_PREFIX            default: NOT defined
+  Causes compiler to prefix all public routines with the string 'dl'.
+  This can be useful when you only want to use this malloc in one part
+  of a program, using your regular system malloc elsewhere.
+
+ABORT                    default: defined as abort()
+  Defines how to abort on failed checks.  On most systems, a failed
+  check cannot die with an "assert" or even print an informative
+  message, because the underlying print routines in turn call malloc,
+  which will fail again.  Generally, the best policy is to simply call
+  abort(). It's not very useful to do more than this because many
+  errors due to overwriting will show up as address faults (null, odd
+  addresses etc) rather than malloc-triggered checks, so will also
+  abort.  Also, most compilers know that abort() does not return, so
+  can better optimize code conditionally calling it.
+
+PROCEED_ON_ERROR           default: defined as 0 (false)
+  Controls whether detected bad addresses cause them to bypassed
+  rather than aborting. If set, detected bad arguments to free and
+  realloc are ignored. And all bookkeeping information is zeroed out
+  upon a detected overwrite of freed heap space, thus losing the
+  ability to ever return it from malloc again, but enabling the
+  application to proceed. If PROCEED_ON_ERROR is defined, the
+  static variable malloc_corruption_error_count is compiled in
+  and can be examined to see if errors have occurred. This option
+  generates slower code than the default abort policy.
+
+DEBUG                    default: NOT defined
+  The DEBUG setting is mainly intended for people trying to modify
+  this code or diagnose problems when porting to new platforms.
+  However, it may also be able to better isolate user errors than just
+  using runtime checks.  The assertions in the check routines spell
+  out in more detail the assumptions and invariants underlying the
+  algorithms.  The checking is fairly extensive, and will slow down
+  execution noticeably. Calling malloc_stats or mallinfo with DEBUG
+  set will attempt to check every non-mmapped allocated and free chunk
+  in the course of computing the summaries.
+
+ABORT_ON_ASSERT_FAILURE   default: defined as 1 (true)
+  Debugging assertion failures can be nearly impossible if your
+  version of the assert macro causes malloc to be called, which will
+  lead to a cascade of further failures, blowing the runtime stack.
+  ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
+  which will usually make debugging easier.
+
+MALLOC_FAILURE_ACTION     default: sets errno to ENOMEM, or no-op on win32
+  The action to take before "return 0" when malloc fails to be able to
+  return memory because there is none available.
+
+HAVE_MORECORE             default: 1 (true) unless win32 or ONLY_MSPACES
+  True if this system supports sbrk or an emulation of it.
+
+MORECORE                  default: sbrk
+  The name of the sbrk-style system routine to call to obtain more
+  memory.  See below for guidance on writing custom MORECORE
+  functions. The type of the argument to sbrk/MORECORE varies across
+  systems.  It cannot be size_t, because it supports negative
+  arguments, so it is normally the signed type of the same width as
+  size_t (sometimes declared as "intptr_t").  It doesn't much matter
+  though. Internally, we only call it with arguments less than half
+  the max value of a size_t, which should work across all reasonable
+  possibilities, although sometimes generating compiler warnings.
+
+MORECORE_CONTIGUOUS       default: 1 (true) if HAVE_MORECORE
+  If true, take advantage of fact that consecutive calls to MORECORE
+  with positive arguments always return contiguous increasing
+  addresses.  This is true of unix sbrk. It does not hurt too much to
+  set it true anyway, since malloc copes with non-contiguities.
+  Setting it false when definitely non-contiguous saves time
+  and possibly wasted space it would take to discover this though.
+
+MORECORE_CANNOT_TRIM      default: NOT defined
+  True if MORECORE cannot release space back to the system when given
+  negative arguments. This is generally necessary only if you are
+  using a hand-crafted MORECORE function that cannot handle negative
+  arguments.
+
+NO_SEGMENT_TRAVERSAL       default: 0
+  If non-zero, suppresses traversals of memory segments
+  returned by either MORECORE or CALL_MMAP. This disables
+  merging of segments that are contiguous, and selectively
+  releasing them to the OS if unused, but bounds execution times.
+
+HAVE_MMAP                 default: 1 (true)
+  True if this system supports mmap or an emulation of it.  If so, and
+  HAVE_MORECORE is not true, MMAP is used for all system
+  allocation. If set and HAVE_MORECORE is true as well, MMAP is
+  primarily used to directly allocate very large blocks. It is also
+  used as a backup strategy in cases where MORECORE fails to provide
+  space from system. Note: A single call to MUNMAP is assumed to be
+  able to unmap memory that may have be allocated using multiple calls
+  to MMAP, so long as they are adjacent.
+
+HAVE_MREMAP               default: 1 on linux, else 0
+  If true realloc() uses mremap() to re-allocate large blocks and
+  extend or shrink allocation spaces.
+
+MMAP_CLEARS               default: 1 except on WINCE.
+  True if mmap clears memory so calloc doesn't need to. This is true
+  for standard unix mmap using /dev/zero and on WIN32 except for WINCE.
+
+USE_BUILTIN_FFS            default: 0 (i.e., not used)
+  Causes malloc to use the builtin ffs() function to compute indices.
+  Some compilers may recognize and intrinsify ffs to be faster than the
+  supplied C version. Also, the case of x86 using gcc is special-cased
+  to an asm instruction, so is already as fast as it can be, and so
+  this setting has no effect. Similarly for Win32 under recent MS compilers.
+  (On most x86s, the asm version is only slightly faster than the C version.)
+
+malloc_getpagesize         default: derive from system includes, or 4096.
+  The system page size. To the extent possible, this malloc manages
+  memory from the system in page-size units.  This may be (and
+  usually is) a function rather than a constant. This is ignored
+  if WIN32, where page size is determined using getSystemInfo during
+  initialization.
+
+USE_DEV_RANDOM             default: 0 (i.e., not used)
+  Causes malloc to use /dev/random to initialize secure magic seed for
+  stamping footers. Otherwise, the current time is used.
+
+NO_MALLINFO                default: 0
+  If defined, don't compile "mallinfo". This can be a simple way
+  of dealing with mismatches between system declarations and
+  those in this file.
+
+MALLINFO_FIELD_TYPE        default: size_t
+  The type of the fields in the mallinfo struct. This was originally
+  defined as "int" in SVID etc, but is more usefully defined as
+  size_t. The value is used only if  HAVE_USR_INCLUDE_MALLOC_H is not set
+
+REALLOC_ZERO_BYTES_FREES    default: not defined
+  This should be set if a call to realloc with zero bytes should
+  be the same as a call to free. Some people think it should. Otherwise,
+  since this malloc returns a unique pointer for malloc(0), so does
+  realloc(p, 0).
+
+LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
+LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H,  LACKS_ERRNO_H
+LACKS_STDLIB_H                default: NOT defined unless on WIN32
+  Define these if your system does not have these header files.
+  You might need to manually insert some of the declarations they provide.
+
+DEFAULT_GRANULARITY        default: page size if MORECORE_CONTIGUOUS,
+				system_info.dwAllocationGranularity in WIN32,
+				otherwise 64K.
+      Also settable using mallopt(M_GRANULARITY, x)
+  The unit for allocating and deallocating memory from the system.  On
+  most systems with contiguous MORECORE, there is no reason to
+  make this more than a page. However, systems with MMAP tend to
+  either require or encourage larger granularities.  You can increase
+  this value to prevent system allocation functions to be called so
+  often, especially if they are slow.  The value must be at least one
+  page and must be a power of two.  Setting to 0 causes initialization
+  to either page size or win32 region size.  (Note: In previous
+  versions of malloc, the equivalent of this option was called
+  "TOP_PAD")
+
+DEFAULT_TRIM_THRESHOLD    default: 2MB
+      Also settable using mallopt(M_TRIM_THRESHOLD, x)
+  The maximum amount of unused top-most memory to keep before
+  releasing via malloc_trim in free().  Automatic trimming is mainly
+  useful in long-lived programs using contiguous MORECORE.  Because
+  trimming via sbrk can be slow on some systems, and can sometimes be
+  wasteful (in cases where programs immediately afterward allocate
+  more large chunks) the value should be high enough so that your
+  overall system performance would improve by releasing this much
+  memory.  As a rough guide, you might set to a value close to the
+  average size of a process (program) running on your system.
+  Releasing this much memory would allow such a process to run in
+  memory.  Generally, it is worth tuning trim thresholds when a
+  program undergoes phases where several large chunks are allocated
+  and released in ways that can reuse each other's storage, perhaps
+  mixed with phases where there are no such chunks at all. The trim
+  value must be greater than page size to have any useful effect.  To
+  disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
+  some people use of mallocing a huge space and then freeing it at
+  program startup, in an attempt to reserve system memory, doesn't
+  have the intended effect under automatic trimming, since that memory
+  will immediately be returned to the system.
+
+DEFAULT_MMAP_THRESHOLD       default: 256K
+      Also settable using mallopt(M_MMAP_THRESHOLD, x)
+  The request size threshold for using MMAP to directly service a
+  request. Requests of at least this size that cannot be allocated
+  using already-existing space will be serviced via mmap.  (If enough
+  normal freed space already exists it is used instead.)  Using mmap
+  segregates relatively large chunks of memory so that they can be
+  individually obtained and released from the host system. A request
+  serviced through mmap is never reused by any other request (at least
+  not directly; the system may just so happen to remap successive
+  requests to the same locations).  Segregating space in this way has
+  the benefits that: Mmapped space can always be individually released
+  back to the system, which helps keep the system level memory demands
+  of a long-lived program low.  Also, mapped memory doesn't become
+  `locked' between other chunks, as can happen with normally allocated
+  chunks, which means that even trimming via malloc_trim would not
+  release them.  However, it has the disadvantage that the space
+  cannot be reclaimed, consolidated, and then used to service later
+  requests, as happens with normal chunks.  The advantages of mmap
+  nearly always outweigh disadvantages for "large" chunks, but the
+  value of "large" may vary across systems.  The default is an
+  empirically derived value that works well in most systems. You can
+  disable mmap by setting to MAX_SIZE_T.
+
+MAX_RELEASE_CHECK_RATE   default: 4095 unless not HAVE_MMAP
+  The number of consolidated frees between checks to release
+  unused segments when freeing. When using non-contiguous segments,
+  especially with multiple mspaces, checking only for topmost space
+  doesn't always suffice to trigger trimming. To compensate for this,
+  free() will, with a period of MAX_RELEASE_CHECK_RATE (or the
+  current number of segments, if greater) try to release unused
+  segments to the OS when freeing chunks that result in
+  consolidation. The best value for this parameter is a compromise
+  between slowing down frees with relatively costly checks that
+  rarely trigger versus holding on to unused memory. To effectively
+  disable, set to MAX_SIZE_T. This may lead to a very slight speed
+  improvement at the expense of carrying around more memory.
+*/
+
+/* Version identifier to allow people to support multiple versions */
+#ifndef DLMALLOC_VERSION
+#define DLMALLOC_VERSION 20804
+#endif /* DLMALLOC_VERSION */
+
+#if defined(linux)
+#define _GNU_SOURCE 1
+#endif
+
+#ifndef WIN32
+#ifdef _WIN32
+#define WIN32 1
+#endif  /* _WIN32 */
+#ifdef _WIN32_WCE
+#define LACKS_FCNTL_H
+#define WIN32 1
+#endif /* _WIN32_WCE */
+#endif  /* WIN32 */
+#ifdef WIN32
+#define WIN32_LEAN_AND_MEAN
+#ifndef _WIN32_WINNT
+#define _WIN32_WINNT 0x403
+#endif
+#include <windows.h>
+#define HAVE_MMAP 1
+#define HAVE_MORECORE 0
+#define LACKS_UNISTD_H
+#define LACKS_SYS_PARAM_H
+#define LACKS_SYS_MMAN_H
+#define LACKS_STRING_H
+#define LACKS_STRINGS_H
+#define LACKS_SYS_TYPES_H
+#define LACKS_ERRNO_H
+#ifndef MALLOC_FAILURE_ACTION
+#define MALLOC_FAILURE_ACTION
+#endif /* MALLOC_FAILURE_ACTION */
+#ifdef _WIN32_WCE /* WINCE reportedly does not clear */
+#define MMAP_CLEARS 0
+#else
+#define MMAP_CLEARS 1
+#endif /* _WIN32_WCE */
+#endif  /* WIN32 */
+
+#if defined(DARWIN) || defined(_DARWIN)
+/* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
+#ifndef HAVE_MORECORE
+#define HAVE_MORECORE 0
+#define HAVE_MMAP 1
+/* OSX allocators provide 16 byte alignment */
+#ifndef MALLOC_ALIGNMENT
+#define MALLOC_ALIGNMENT ((size_t)16U)
+#endif
+#endif  /* HAVE_MORECORE */
+#endif  /* DARWIN */
+
+#ifndef LACKS_SYS_TYPES_H
+#include <sys/types.h>  /* For size_t */
+#endif  /* LACKS_SYS_TYPES_H */
+
+/* The maximum possible size_t value has all bits set */
+#define MAX_SIZE_T           (~(size_t)0)
+
+#ifndef ONLY_MSPACES
+#define ONLY_MSPACES 0     /* define to a value */
+#else
+#define ONLY_MSPACES 1
+#endif  /* ONLY_MSPACES */
+#ifndef MSPACES
+#if ONLY_MSPACES
+#define MSPACES 1
+#else   /* ONLY_MSPACES */
+#define MSPACES 0
+#endif  /* ONLY_MSPACES */
+#endif  /* MSPACES */
+#ifndef MALLOC_ALIGNMENT
+#define MALLOC_ALIGNMENT ((size_t)8U)
+#endif  /* MALLOC_ALIGNMENT */
+#ifndef FOOTERS
+#define FOOTERS 0
+#endif  /* FOOTERS */
+#ifndef ABORT
+#define ABORT  abort()
+#endif  /* ABORT */
+#ifndef ABORT_ON_ASSERT_FAILURE
+#define ABORT_ON_ASSERT_FAILURE 1
+#endif  /* ABORT_ON_ASSERT_FAILURE */
+#ifndef PROCEED_ON_ERROR
+#define PROCEED_ON_ERROR 0
+#endif  /* PROCEED_ON_ERROR */
+#ifndef USE_LOCKS
+#define USE_LOCKS 0
+#endif  /* USE_LOCKS */
+#ifndef USE_SPIN_LOCKS
+#if USE_LOCKS && (defined(__GNUC__) && ((defined(__i386__) || defined(__x86_64__)))) || (defined(_MSC_VER) && _MSC_VER>=1310)
+#define USE_SPIN_LOCKS 1
+#else
+#define USE_SPIN_LOCKS 0
+#endif /* USE_LOCKS && ... */
+#endif /* USE_SPIN_LOCKS */
+#ifndef INSECURE
+#define INSECURE 0
+#endif  /* INSECURE */
+#ifndef HAVE_MMAP
+#define HAVE_MMAP 1
+#endif  /* HAVE_MMAP */
+#ifndef MMAP_CLEARS
+#define MMAP_CLEARS 1
+#endif  /* MMAP_CLEARS */
+#ifndef HAVE_MREMAP
+#ifdef linux
+#define HAVE_MREMAP 1
+#else   /* linux */
+#define HAVE_MREMAP 0
+#endif  /* linux */
+#endif  /* HAVE_MREMAP */
+#ifndef MALLOC_FAILURE_ACTION
+#define MALLOC_FAILURE_ACTION  errno = ENOMEM;
+#endif  /* MALLOC_FAILURE_ACTION */
+#ifndef HAVE_MORECORE
+#if ONLY_MSPACES
+#define HAVE_MORECORE 0
+#else   /* ONLY_MSPACES */
+#define HAVE_MORECORE 1
+#endif  /* ONLY_MSPACES */
+#endif  /* HAVE_MORECORE */
+#if !HAVE_MORECORE
+#define MORECORE_CONTIGUOUS 0
+#else   /* !HAVE_MORECORE */
+#define MORECORE_DEFAULT sbrk
+#ifndef MORECORE_CONTIGUOUS
+#define MORECORE_CONTIGUOUS 1
+#endif  /* MORECORE_CONTIGUOUS */
+#endif  /* HAVE_MORECORE */
+#ifndef DEFAULT_GRANULARITY
+#if (MORECORE_CONTIGUOUS || defined(WIN32))
+#define DEFAULT_GRANULARITY (0)  /* 0 means to compute in init_mparams */
+#else   /* MORECORE_CONTIGUOUS */
+#define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
+#endif  /* MORECORE_CONTIGUOUS */
+#endif  /* DEFAULT_GRANULARITY */
+#ifndef DEFAULT_TRIM_THRESHOLD
+#ifndef MORECORE_CANNOT_TRIM
+#define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
+#else   /* MORECORE_CANNOT_TRIM */
+#define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
+#endif  /* MORECORE_CANNOT_TRIM */
+#endif  /* DEFAULT_TRIM_THRESHOLD */
+#ifndef DEFAULT_MMAP_THRESHOLD
+#if HAVE_MMAP
+#define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
+#else   /* HAVE_MMAP */
+#define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
+#endif  /* HAVE_MMAP */
+#endif  /* DEFAULT_MMAP_THRESHOLD */
+#ifndef MAX_RELEASE_CHECK_RATE
+#if HAVE_MMAP
+#define MAX_RELEASE_CHECK_RATE 4095
+#else
+#define MAX_RELEASE_CHECK_RATE MAX_SIZE_T
+#endif /* HAVE_MMAP */
+#endif /* MAX_RELEASE_CHECK_RATE */
+#ifndef USE_BUILTIN_FFS
+#define USE_BUILTIN_FFS 0
+#endif  /* USE_BUILTIN_FFS */
+#ifndef USE_DEV_RANDOM
+#define USE_DEV_RANDOM 0
+#endif  /* USE_DEV_RANDOM */
+#ifndef NO_MALLINFO
+#define NO_MALLINFO 0
+#endif  /* NO_MALLINFO */
+#ifndef MALLINFO_FIELD_TYPE
+#define MALLINFO_FIELD_TYPE size_t
+#endif  /* MALLINFO_FIELD_TYPE */
+#ifndef NO_SEGMENT_TRAVERSAL
+#define NO_SEGMENT_TRAVERSAL 0
+#endif /* NO_SEGMENT_TRAVERSAL */
+
+/*
+  mallopt tuning options.  SVID/XPG defines four standard parameter
+  numbers for mallopt, normally defined in malloc.h.  None of these
+  are used in this malloc, so setting them has no effect. But this
+  malloc does support the following options.
+*/
+
+#define M_TRIM_THRESHOLD     (-1)
+#define M_GRANULARITY        (-2)
+#define M_MMAP_THRESHOLD     (-3)
+
+/* ------------------------ Mallinfo declarations ------------------------ */
+
+#if !NO_MALLINFO
+/*
+  This version of malloc supports the standard SVID/XPG mallinfo
+  routine that returns a struct containing usage properties and
+  statistics. It should work on any system that has a
+  /usr/include/malloc.h defining struct mallinfo.  The main
+  declaration needed is the mallinfo struct that is returned (by-copy)
+  by mallinfo().  The malloinfo struct contains a bunch of fields that
+  are not even meaningful in this version of malloc.  These fields are
+  are instead filled by mallinfo() with other numbers that might be of
+  interest.
+
+  HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
+  /usr/include/malloc.h file that includes a declaration of struct
+  mallinfo.  If so, it is included; else a compliant version is
+  declared below.  These must be precisely the same for mallinfo() to
+  work.  The original SVID version of this struct, defined on most
+  systems with mallinfo, declares all fields as ints. But some others
+  define as unsigned long. If your system defines the fields using a
+  type of different width than listed here, you MUST #include your
+  system version and #define HAVE_USR_INCLUDE_MALLOC_H.
+*/
+
+/* #define HAVE_USR_INCLUDE_MALLOC_H */
+
+#ifdef HAVE_USR_INCLUDE_MALLOC_H
+#include "/usr/include/malloc.h"
+#else /* HAVE_USR_INCLUDE_MALLOC_H */
+#ifndef STRUCT_MALLINFO_DECLARED
+#define STRUCT_MALLINFO_DECLARED 1
+struct mallinfo {
+  MALLINFO_FIELD_TYPE arena;    /* non-mmapped space allocated from system */
+  MALLINFO_FIELD_TYPE ordblks;  /* number of free chunks */
+  MALLINFO_FIELD_TYPE smblks;   /* always 0 */
+  MALLINFO_FIELD_TYPE hblks;    /* always 0 */
+  MALLINFO_FIELD_TYPE hblkhd;   /* space in mmapped regions */
+  MALLINFO_FIELD_TYPE usmblks;  /* maximum total allocated space */
+  MALLINFO_FIELD_TYPE fsmblks;  /* always 0 */
+  MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
+  MALLINFO_FIELD_TYPE fordblks; /* total free space */
+  MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
+};
+#endif /* STRUCT_MALLINFO_DECLARED */
+#endif /* HAVE_USR_INCLUDE_MALLOC_H */
+#endif /* NO_MALLINFO */
+
+/*
+  Try to persuade compilers to inline. The most critical functions for
+  inlining are defined as macros, so these aren't used for them.
+*/
+
+#ifdef __MINGW64_VERSION_MAJOR
+#undef FORCEINLINE
+#endif
+#ifndef FORCEINLINE
+  #if defined(__GNUC__)
+#define FORCEINLINE __inline __attribute__ ((always_inline))
+  #elif defined(_MSC_VER)
+    #define FORCEINLINE __forceinline
+  #endif
+#endif
+#ifndef NOINLINE
+  #if defined(__GNUC__)
+    #define NOINLINE __attribute__ ((noinline))
+  #elif defined(_MSC_VER)
+    #define NOINLINE __declspec(noinline)
+  #else
+    #define NOINLINE
+  #endif
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#ifndef FORCEINLINE
+ #define FORCEINLINE inline
+#endif
+#endif /* __cplusplus */
+#ifndef FORCEINLINE
+ #define FORCEINLINE
+#endif
+
+#if !ONLY_MSPACES
+
+/* ------------------- Declarations of public routines ------------------- */
+
+#ifndef USE_DL_PREFIX
+#define dlcalloc               calloc
+#define dlfree                 free
+#define dlmalloc               malloc
+#define dlmemalign             memalign
+#define dlrealloc              realloc
+#define dlvalloc               valloc
+#define dlpvalloc              pvalloc
+#define dlmallinfo             mallinfo
+#define dlmallopt              mallopt
+#define dlmalloc_trim          malloc_trim
+#define dlmalloc_stats         malloc_stats
+#define dlmalloc_usable_size   malloc_usable_size
+#define dlmalloc_footprint     malloc_footprint
+#define dlmalloc_max_footprint malloc_max_footprint
+#define dlindependent_calloc   independent_calloc
+#define dlindependent_comalloc independent_comalloc
+#endif /* USE_DL_PREFIX */
+
+
+/*
+  malloc(size_t n)
+  Returns a pointer to a newly allocated chunk of at least n bytes, or
+  null if no space is available, in which case errno is set to ENOMEM
+  on ANSI C systems.
+
+  If n is zero, malloc returns a minimum-sized chunk. (The minimum
+  size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
+  systems.)  Note that size_t is an unsigned type, so calls with
+  arguments that would be negative if signed are interpreted as
+  requests for huge amounts of space, which will often fail. The
+  maximum supported value of n differs across systems, but is in all
+  cases less than the maximum representable value of a size_t.
+*/
+void* dlmalloc(size_t);
+
+/*
+  free(void* p)
+  Releases the chunk of memory pointed to by p, that had been previously
+  allocated using malloc or a related routine such as realloc.
+  It has no effect if p is null. If p was not malloced or already
+  freed, free(p) will by default cause the current program to abort.
+*/
+void  dlfree(void*);
+
+/*
+  calloc(size_t n_elements, size_t element_size);
+  Returns a pointer to n_elements * element_size bytes, with all locations
+  set to zero.
+*/
+void* dlcalloc(size_t, size_t);
+
+/*
+  realloc(void* p, size_t n)
+  Returns a pointer to a chunk of size n that contains the same data
+  as does chunk p up to the minimum of (n, p's size) bytes, or null
+  if no space is available.
+
+  The returned pointer may or may not be the same as p. The algorithm
+  prefers extending p in most cases when possible, otherwise it
+  employs the equivalent of a malloc-copy-free sequence.
+
+  If p is null, realloc is equivalent to malloc.
+
+  If space is not available, realloc returns null, errno is set (if on
+  ANSI) and p is NOT freed.
+
+  if n is for fewer bytes than already held by p, the newly unused
+  space is lopped off and freed if possible.  realloc with a size
+  argument of zero (re)allocates a minimum-sized chunk.
+
+  The old unix realloc convention of allowing the last-free'd chunk
+  to be used as an argument to realloc is not supported.
+*/
+
+void* dlrealloc(void*, size_t);
+
+/*
+  memalign(size_t alignment, size_t n);
+  Returns a pointer to a newly allocated chunk of n bytes, aligned
+  in accord with the alignment argument.
+
+  The alignment argument should be a power of two. If the argument is
+  not a power of two, the nearest greater power is used.
+  8-byte alignment is guaranteed by normal malloc calls, so don't
+  bother calling memalign with an argument of 8 or less.
+
+  Overreliance on memalign is a sure way to fragment space.
+*/
+void* dlmemalign(size_t, size_t);
+
+/*
+  valloc(size_t n);
+  Equivalent to memalign(pagesize, n), where pagesize is the page
+  size of the system. If the pagesize is unknown, 4096 is used.
+*/
+void* dlvalloc(size_t);
+
+/*
+  mallopt(int parameter_number, int parameter_value)
+  Sets tunable parameters The format is to provide a
+  (parameter-number, parameter-value) pair.  mallopt then sets the
+  corresponding parameter to the argument value if it can (i.e., so
+  long as the value is meaningful), and returns 1 if successful else
+  0.  To workaround the fact that mallopt is specified to use int,
+  not size_t parameters, the value -1 is specially treated as the
+  maximum unsigned size_t value.
+
+  SVID/XPG/ANSI defines four standard param numbers for mallopt,
+  normally defined in malloc.h.  None of these are use in this malloc,
+  so setting them has no effect. But this malloc also supports other
+  options in mallopt. See below for details.  Briefly, supported
+  parameters are as follows (listed defaults are for "typical"
+  configurations).
+
+  Symbol            param #  default    allowed param values
+  M_TRIM_THRESHOLD     -1   2*1024*1024   any   (-1 disables)
+  M_GRANULARITY        -2     page size   any power of 2 >= page size
+  M_MMAP_THRESHOLD     -3      256*1024   any   (or 0 if no MMAP support)
+*/
+int dlmallopt(int, int);
+
+/*
+  malloc_footprint();
+  Returns the number of bytes obtained from the system.  The total
+  number of bytes allocated by malloc, realloc etc., is less than this
+  value. Unlike mallinfo, this function returns only a precomputed
+  result, so can be called frequently to monitor memory consumption.
+  Even if locks are otherwise defined, this function does not use them,
+  so results might not be up to date.
+*/
+size_t dlmalloc_footprint(void);
+
+/*
+  malloc_max_footprint();
+  Returns the maximum number of bytes obtained from the system. This
+  value will be greater than current footprint if deallocated space
+  has been reclaimed by the system. The peak number of bytes allocated
+  by malloc, realloc etc., is less than this value. Unlike mallinfo,
+  this function returns only a precomputed result, so can be called
+  frequently to monitor memory consumption.  Even if locks are
+  otherwise defined, this function does not use them, so results might
+  not be up to date.
+*/
+size_t dlmalloc_max_footprint(void);
+
+#if !NO_MALLINFO
+/*
+  mallinfo()
+  Returns (by copy) a struct containing various summary statistics:
+
+  arena:     current total non-mmapped bytes allocated from system
+  ordblks:   the number of free chunks
+  smblks:    always zero.
+  hblks:     current number of mmapped regions
+  hblkhd:    total bytes held in mmapped regions
+  usmblks:   the maximum total allocated space. This will be greater
+		than current total if trimming has occurred.
+  fsmblks:   always zero
+  uordblks:  current total allocated space (normal or mmapped)
+  fordblks:  total free space
+  keepcost:  the maximum number of bytes that could ideally be released
+	       back to system via malloc_trim. ("ideally" means that
+	       it ignores page restrictions etc.)
+
+  Because these fields are ints, but internal bookkeeping may
+  be kept as longs, the reported values may wrap around zero and
+  thus be inaccurate.
+*/
+struct mallinfo dlmallinfo(void);
+#endif /* NO_MALLINFO */
+
+/*
+  independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
+
+  independent_calloc is similar to calloc, but instead of returning a
+  single cleared space, it returns an array of pointers to n_elements
+  independent elements that can hold contents of size elem_size, each
+  of which starts out cleared, and can be independently freed,
+  realloc'ed etc. The elements are guaranteed to be adjacently
+  allocated (this is not guaranteed to occur with multiple callocs or
+  mallocs), which may also improve cache locality in some
+  applications.
+
+  The "chunks" argument is optional (i.e., may be null, which is
+  probably the most typical usage). If it is null, the returned array
+  is itself dynamically allocated and should also be freed when it is
+  no longer needed. Otherwise, the chunks array must be of at least
+  n_elements in length. It is filled in with the pointers to the
+  chunks.
+
+  In either case, independent_calloc returns this pointer array, or
+  null if the allocation failed.  If n_elements is zero and "chunks"
+  is null, it returns a chunk representing an array with zero elements
+  (which should be freed if not wanted).
+
+  Each element must be individually freed when it is no longer
+  needed. If you'd like to instead be able to free all at once, you
+  should instead use regular calloc and assign pointers into this
+  space to represent elements.  (In this case though, you cannot
+  independently free elements.)
+
+  independent_calloc simplifies and speeds up implementations of many
+  kinds of pools.  It may also be useful when constructing large data
+  structures that initially have a fixed number of fixed-sized nodes,
+  but the number is not known at compile time, and some of the nodes
+  may later need to be freed. For example:
+
+  struct Node { int item; struct Node* next; };
+
+  struct Node* build_list() {
+    struct Node** pool;
+    int n = read_number_of_nodes_needed();
+    if (n <= 0) return 0;
+    pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
+    if (pool == 0) die();
+    // organize into a linked list...
+    struct Node* first = pool[0];
+    for (i = 0; i < n-1; ++i)
+      pool[i]->next = pool[i+1];
+    free(pool);     // Can now free the array (or not, if it is needed later)
+    return first;
+  }
+*/
+void** dlindependent_calloc(size_t, size_t, void**);
+
+/*
+  independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
+
+  independent_comalloc allocates, all at once, a set of n_elements
+  chunks with sizes indicated in the "sizes" array.    It returns
+  an array of pointers to these elements, each of which can be
+  independently freed, realloc'ed etc. The elements are guaranteed to
+  be adjacently allocated (this is not guaranteed to occur with
+  multiple callocs or mallocs), which may also improve cache locality
+  in some applications.
+
+  The "chunks" argument is optional (i.e., may be null). If it is null
+  the returned array is itself dynamically allocated and should also
+  be freed when it is no longer needed. Otherwise, the chunks array
+  must be of at least n_elements in length. It is filled in with the
+  pointers to the chunks.
+
+  In either case, independent_comalloc returns this pointer array, or
+  null if the allocation failed.  If n_elements is zero and chunks is
+  null, it returns a chunk representing an array with zero elements
+  (which should be freed if not wanted).
+
+  Each element must be individually freed when it is no longer
+  needed. If you'd like to instead be able to free all at once, you
+  should instead use a single regular malloc, and assign pointers at
+  particular offsets in the aggregate space. (In this case though, you
+  cannot independently free elements.)
+
+  independent_comallac differs from independent_calloc in that each
+  element may have a different size, and also that it does not
+  automatically clear elements.
+
+  independent_comalloc can be used to speed up allocation in cases
+  where several structs or objects must always be allocated at the
+  same time.  For example:
+
+  struct Head { ... }
+  struct Foot { ... }
+
+  void send_message(char* msg) {
+    int msglen = strlen(msg);
+    size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
+    void* chunks[3];
+    if (independent_comalloc(3, sizes, chunks) == 0)
+      die();
+    struct Head* head = (struct Head*)(chunks[0]);
+    char*        body = (char*)(chunks[1]);
+    struct Foot* foot = (struct Foot*)(chunks[2]);
+    // ...
+  }
+
+  In general though, independent_comalloc is worth using only for
+  larger values of n_elements. For small values, you probably won't
+  detect enough difference from series of malloc calls to bother.
+
+  Overuse of independent_comalloc can increase overall memory usage,
+  since it cannot reuse existing noncontiguous small chunks that
+  might be available for some of the elements.
+*/
+void** dlindependent_comalloc(size_t, size_t*, void**);
+
+
+/*
+  pvalloc(size_t n);
+  Equivalent to valloc(minimum-page-that-holds(n)), that is,
+  round up n to nearest pagesize.
+ */
+void*  dlpvalloc(size_t);
+
+/*
+  malloc_trim(size_t pad);
+
+  If possible, gives memory back to the system (via negative arguments
+  to sbrk) if there is unused memory at the `high' end of the malloc
+  pool or in unused MMAP segments. You can call this after freeing
+  large blocks of memory to potentially reduce the system-level memory
+  requirements of a program. However, it cannot guarantee to reduce
+  memory. Under some allocation patterns, some large free blocks of
+  memory will be locked between two used chunks, so they cannot be
+  given back to the system.
+
+  The `pad' argument to malloc_trim represents the amount of free
+  trailing space to leave untrimmed. If this argument is zero, only
+  the minimum amount of memory to maintain internal data structures
+  will be left. Non-zero arguments can be supplied to maintain enough
+  trailing space to service future expected allocations without having
+  to re-obtain memory from the system.
+
+  Malloc_trim returns 1 if it actually released any memory, else 0.
+*/
+int  dlmalloc_trim(size_t);
+
+/*
+  malloc_stats();
+  Prints on stderr the amount of space obtained from the system (both
+  via sbrk and mmap), the maximum amount (which may be more than
+  current if malloc_trim and/or munmap got called), and the current
+  number of bytes allocated via malloc (or realloc, etc) but not yet
+  freed. Note that this is the number of bytes allocated, not the
+  number requested. It will be larger than the number requested
+  because of alignment and bookkeeping overhead. Because it includes
+  alignment wastage as being in use, this figure may be greater than
+  zero even when no user-level chunks are allocated.
+
+  The reported current and maximum system memory can be inaccurate if
+  a program makes other calls to system memory allocation functions
+  (normally sbrk) outside of malloc.
+
+  malloc_stats prints only the most commonly interesting statistics.
+  More information can be obtained by calling mallinfo.
+*/
+void  dlmalloc_stats(void);
+
+#endif /* ONLY_MSPACES */
+
+/*
+  malloc_usable_size(void* p);
+
+  Returns the number of bytes you can actually use in
+  an allocated chunk, which may be more than you requested (although
+  often not) due to alignment and minimum size constraints.
+  You can use this many bytes without worrying about
+  overwriting other allocated objects. This is not a particularly great
+  programming practice. malloc_usable_size can be more useful in
+  debugging and assertions, for example:
+
+  p = malloc(n);
+  assert(malloc_usable_size(p) >= 256);
+*/
+size_t dlmalloc_usable_size(void*);
+
+
+#if MSPACES
+
+/*
+  mspace is an opaque type representing an independent
+  region of space that supports mspace_malloc, etc.
+*/
+typedef void* mspace;
+
+/*
+  create_mspace creates and returns a new independent space with the
+  given initial capacity, or, if 0, the default granularity size.  It
+  returns null if there is no system memory available to create the
+  space.  If argument locked is non-zero, the space uses a separate
+  lock to control access. The capacity of the space will grow
+  dynamically as needed to service mspace_malloc requests.  You can
+  control the sizes of incremental increases of this space by
+  compiling with a different DEFAULT_GRANULARITY or dynamically
+  setting with mallopt(M_GRANULARITY, value).
+*/
+mspace create_mspace(size_t capacity, int locked);
+
+/*
+  destroy_mspace destroys the given space, and attempts to return all
+  of its memory back to the system, returning the total number of
+  bytes freed. After destruction, the results of access to all memory
+  used by the space become undefined.
+*/
+size_t destroy_mspace(mspace msp);
+
+/*
+  create_mspace_with_base uses the memory supplied as the initial base
+  of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
+  space is used for bookkeeping, so the capacity must be at least this
+  large. (Otherwise 0 is returned.) When this initial space is
+  exhausted, additional memory will be obtained from the system.
+  Destroying this space will deallocate all additionally allocated
+  space (if possible) but not the initial base.
+*/
+mspace create_mspace_with_base(void* base, size_t capacity, int locked);
+
+/*
+  mspace_mmap_large_chunks controls whether requests for large chunks
+  are allocated in their own mmapped regions, separate from others in
+  this mspace. By default this is enabled, which reduces
+  fragmentation. However, such chunks are not necessarily released to
+  the system upon destroy_mspace.  Disabling by setting to false may
+  increase fragmentation, but avoids leakage when relying on
+  destroy_mspace to release all memory allocated using this space.
+*/
+int mspace_mmap_large_chunks(mspace msp, int enable);
+
+
+/*
+  mspace_malloc behaves as malloc, but operates within
+  the given space.
+*/
+void* mspace_malloc(mspace msp, size_t bytes);
+
+/*
+  mspace_free behaves as free, but operates within
+  the given space.
+
+  If compiled with FOOTERS==1, mspace_free is not actually needed.
+  free may be called instead of mspace_free because freed chunks from
+  any space are handled by their originating spaces.
+*/
+void mspace_free(mspace msp, void* mem);
+
+/*
+  mspace_realloc behaves as realloc, but operates within
+  the given space.
+
+  If compiled with FOOTERS==1, mspace_realloc is not actually
+  needed.  realloc may be called instead of mspace_realloc because
+  realloced chunks from any space are handled by their originating
+  spaces.
+*/
+void* mspace_realloc(mspace msp, void* mem, size_t newsize);
+
+/*
+  mspace_calloc behaves as calloc, but operates within
+  the given space.
+*/
+void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
+
+/*
+  mspace_memalign behaves as memalign, but operates within
+  the given space.
+*/
+void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
+
+/*
+  mspace_independent_calloc behaves as independent_calloc, but
+  operates within the given space.
+*/
+void** mspace_independent_calloc(mspace msp, size_t n_elements,
+				 size_t elem_size, void* chunks[]);
+
+/*
+  mspace_independent_comalloc behaves as independent_comalloc, but
+  operates within the given space.
+*/
+void** mspace_independent_comalloc(mspace msp, size_t n_elements,
+				   size_t sizes[], void* chunks[]);
+
+/*
+  mspace_footprint() returns the number of bytes obtained from the
+  system for this space.
+*/
+size_t mspace_footprint(mspace msp);
+
+/*
+  mspace_max_footprint() returns the peak number of bytes obtained from the
+  system for this space.
+*/
+size_t mspace_max_footprint(mspace msp);
+
+
+#if !NO_MALLINFO
+/*
+  mspace_mallinfo behaves as mallinfo, but reports properties of
+  the given space.
+*/
+struct mallinfo mspace_mallinfo(mspace msp);
+#endif /* NO_MALLINFO */
+
+/*
+  malloc_usable_size(void* p) behaves the same as malloc_usable_size;
+*/
+  size_t mspace_usable_size(void* mem);
+
+/*
+  mspace_malloc_stats behaves as malloc_stats, but reports
+  properties of the given space.
+*/
+void mspace_malloc_stats(mspace msp);
+
+/*
+  mspace_trim behaves as malloc_trim, but
+  operates within the given space.
+*/
+int mspace_trim(mspace msp, size_t pad);
+
+/*
+  An alias for mallopt.
+*/
+int mspace_mallopt(int, int);
+
+#endif /* MSPACES */
+
+#ifdef __cplusplus
+};  /* end of extern "C" */
+#endif /* __cplusplus */
+
+/*
+  ========================================================================
+  To make a fully customizable malloc.h header file, cut everything
+  above this line, put into file malloc.h, edit to suit, and #include it
+  on the next line, as well as in programs that use this malloc.
+  ========================================================================
+*/
+
+/* #include "malloc.h" */
+
+/*------------------------------ internal #includes ---------------------- */
+
+#ifdef WIN32
+#ifndef __GNUC__
+#pragma warning( disable : 4146 ) /* no "unsigned" warnings */
+#endif
+#endif /* WIN32 */
+
+#include <stdio.h>       /* for printing in malloc_stats */
+
+#ifndef LACKS_ERRNO_H
+#include <errno.h>       /* for MALLOC_FAILURE_ACTION */
+#endif /* LACKS_ERRNO_H */
+#if FOOTERS
+#include <time.h>        /* for magic initialization */
+#endif /* FOOTERS */
+#ifndef LACKS_STDLIB_H
+#include <stdlib.h>      /* for abort() */
+#endif /* LACKS_STDLIB_H */
+#ifdef DEBUG
+#if ABORT_ON_ASSERT_FAILURE
+#define assert(x) if(!(x)) ABORT
+#else /* ABORT_ON_ASSERT_FAILURE */
+#include <assert.h>
+#endif /* ABORT_ON_ASSERT_FAILURE */
+#else  /* DEBUG */
+#ifndef assert
+#define assert(x)
+#endif
+#define DEBUG 0
+#endif /* DEBUG */
+#ifndef LACKS_STRING_H
+#include <string.h>      /* for memset etc */
+#endif  /* LACKS_STRING_H */
+#if USE_BUILTIN_FFS
+#ifndef LACKS_STRINGS_H
+#include <strings.h>     /* for ffs */
+#endif /* LACKS_STRINGS_H */
+#endif /* USE_BUILTIN_FFS */
+#if HAVE_MMAP
+#ifndef LACKS_SYS_MMAN_H
+#include <sys/mman.h>    /* for mmap */
+#endif /* LACKS_SYS_MMAN_H */
+#ifndef LACKS_FCNTL_H
+#include <fcntl.h>
+#endif /* LACKS_FCNTL_H */
+#endif /* HAVE_MMAP */
+#ifndef LACKS_UNISTD_H
+#include <unistd.h>     /* for sbrk, sysconf */
+#else /* LACKS_UNISTD_H */
+#if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
+extern void*     sbrk(ptrdiff_t);
+#endif /* FreeBSD etc */
+#endif /* LACKS_UNISTD_H */
+
+/* Declarations for locking */
+#if USE_LOCKS
+#ifndef WIN32
+#include <pthread.h>
+#if defined (__SVR4) && defined (__sun)  /* solaris */
+#include <thread.h>
+#endif /* solaris */
+#else
+#ifndef _M_AMD64
+/* These are already defined on AMD64 builds */
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+#ifndef __MINGW32__
+LONG __cdecl _InterlockedCompareExchange(LONG volatile *Dest, LONG Exchange, LONG Comp);
+LONG __cdecl _InterlockedExchange(LONG volatile *Target, LONG Value);
+#endif
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+#endif /* _M_AMD64 */
+#ifndef __MINGW32__
+#pragma intrinsic (_InterlockedCompareExchange)
+#pragma intrinsic (_InterlockedExchange)
+#else
+  /* --[ start GCC compatibility ]----------------------------------------------
+   * Compatibility <intrin_x86.h> header for GCC -- GCC equivalents of intrinsic
+   * Microsoft Visual C++ functions. Originally developed for the ReactOS
+   * (<http://www.reactos.org/>) and TinyKrnl (<http://www.tinykrnl.org/>)
+   * projects.
+   *
+   * Copyright (c) 2006 KJK::Hyperion <hackbunny@reactos.com>
+   *
+   * Permission is hereby granted, free of charge, to any person obtaining a
+   * copy of this software and associated documentation files (the "Software"),
+   * to deal in the Software without restriction, including without limitation
+   * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   * and/or sell copies of the Software, and to permit persons to whom the
+   * Software is furnished to do so, subject to the following conditions:
+   *
+   * The above copyright notice and this permission notice shall be included in
+   * all copies or substantial portions of the Software.
+   *
+   * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   * DEALINGS IN THE SOFTWARE.
+   */
+
+  /*** Atomic operations ***/
+  #if (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__) > 40100
+    #undef _ReadWriteBarrier
+    #define _ReadWriteBarrier() __sync_synchronize()
+  #else
+    static __inline__ __attribute__((always_inline)) long __sync_lock_test_and_set(volatile long * const Target, const long Value)
+    {
+      long res;
+      __asm__ __volatile__("xchg%z0 %2, %0" : "=g" (*(Target)), "=r" (res) : "1" (Value));
+      return res;
+    }
+    static void __inline__ __attribute__((always_inline)) _MemoryBarrier(void)
+    {
+      __asm__ __volatile__("" : : : "memory");
+    }
+    #define _ReadWriteBarrier() _MemoryBarrier()
+  #endif
+  /* BUGBUG: GCC only supports full barriers */
+  static __inline__ __attribute__((always_inline)) long _InterlockedExchange(volatile long * const Target, const long Value)
+  {
+    /* NOTE: __sync_lock_test_and_set would be an acquire barrier, so we force a full barrier */
+    _ReadWriteBarrier();
+    return __sync_lock_test_and_set(Target, Value);
+  }
+  /* --[ end GCC compatibility ]---------------------------------------------- */
+#endif
+#define interlockedcompareexchange _InterlockedCompareExchange
+#define interlockedexchange _InterlockedExchange
+#endif /* Win32 */
+#endif /* USE_LOCKS */
+
+/* Declarations for bit scanning on win32 */
+#if defined(_MSC_VER) && _MSC_VER>=1300
+#ifndef BitScanForward	/* Try to avoid pulling in WinNT.h */
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+unsigned char _BitScanForward(unsigned long *index, unsigned long mask);
+unsigned char _BitScanReverse(unsigned long *index, unsigned long mask);
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+
+#define BitScanForward _BitScanForward
+#define BitScanReverse _BitScanReverse
+#pragma intrinsic(_BitScanForward)
+#pragma intrinsic(_BitScanReverse)
+#endif /* BitScanForward */
+#endif /* defined(_MSC_VER) && _MSC_VER>=1300 */
+
+#ifndef WIN32
+#ifndef malloc_getpagesize
+#  ifdef _SC_PAGESIZE         /* some SVR4 systems omit an underscore */
+#    ifndef _SC_PAGE_SIZE
+#      define _SC_PAGE_SIZE _SC_PAGESIZE
+#    endif
+#  endif
+#  ifdef _SC_PAGE_SIZE
+#    define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
+#  else
+#    if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
+       extern size_t getpagesize();
+#      define malloc_getpagesize getpagesize()
+#    else
+#      ifdef WIN32 /* use supplied emulation of getpagesize */
+#        define malloc_getpagesize getpagesize()
+#      else
+#        ifndef LACKS_SYS_PARAM_H
+#          include <sys/param.h>
+#        endif
+#        ifdef EXEC_PAGESIZE
+#          define malloc_getpagesize EXEC_PAGESIZE
+#        else
+#          ifdef NBPG
+#            ifndef CLSIZE
+#              define malloc_getpagesize NBPG
+#            else
+#              define malloc_getpagesize (NBPG * CLSIZE)
+#            endif
+#          else
+#            ifdef NBPC
+#              define malloc_getpagesize NBPC
+#            else
+#              ifdef PAGESIZE
+#                define malloc_getpagesize PAGESIZE
+#              else /* just guess */
+#                define malloc_getpagesize ((size_t)4096U)
+#              endif
+#            endif
+#          endif
+#        endif
+#      endif
+#    endif
+#  endif
+#endif
+#endif
+
+
+
+/* ------------------- size_t and alignment properties -------------------- */
+
+/* The byte and bit size of a size_t */
+#define SIZE_T_SIZE         (sizeof(size_t))
+#define SIZE_T_BITSIZE      (sizeof(size_t) << 3)
+
+/* Some constants coerced to size_t */
+/* Annoying but necessary to avoid errors on some platforms */
+#define SIZE_T_ZERO         ((size_t)0)
+#define SIZE_T_ONE          ((size_t)1)
+#define SIZE_T_TWO          ((size_t)2)
+#define SIZE_T_FOUR         ((size_t)4)
+#define TWO_SIZE_T_SIZES    (SIZE_T_SIZE<<1)
+#define FOUR_SIZE_T_SIZES   (SIZE_T_SIZE<<2)
+#define SIX_SIZE_T_SIZES    (FOUR_SIZE_T_SIZES+TWO_SIZE_T_SIZES)
+#define HALF_MAX_SIZE_T     (MAX_SIZE_T / 2U)
+
+/* The bit mask value corresponding to MALLOC_ALIGNMENT */
+#define CHUNK_ALIGN_MASK    (MALLOC_ALIGNMENT - SIZE_T_ONE)
+
+/* True if address a has acceptable alignment */
+#define is_aligned(A)       (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
+
+/* the number of bytes to offset an address to align it */
+#define align_offset(A)\
+ ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)? 0 :\
+  ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & CHUNK_ALIGN_MASK))
+
+/* -------------------------- MMAP preliminaries ------------------------- */
+
+/*
+   If HAVE_MORECORE or HAVE_MMAP are false, we just define calls and
+   checks to fail so compiler optimizer can delete code rather than
+   using so many "#if"s.
+*/
+
+
+/* MORECORE and MMAP must return MFAIL on failure */
+#define MFAIL                ((void*)(MAX_SIZE_T))
+#define CMFAIL               ((char*)(MFAIL)) /* defined for convenience */
+
+#if HAVE_MMAP
+
+#ifndef WIN32
+#define MUNMAP_DEFAULT(a, s)  munmap((a), (s))
+#define MMAP_PROT            (PROT_READ|PROT_WRITE)
+#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
+#define MAP_ANONYMOUS        MAP_ANON
+#endif /* MAP_ANON */
+#ifdef MAP_ANONYMOUS
+#define MMAP_FLAGS           (MAP_PRIVATE|MAP_ANONYMOUS)
+#define MMAP_DEFAULT(s)       mmap(0, (s), MMAP_PROT, MMAP_FLAGS, -1, 0)
+#else /* MAP_ANONYMOUS */
+/*
+   Nearly all versions of mmap support MAP_ANONYMOUS, so the following
+   is unlikely to be needed, but is supplied just in case.
+*/
+#define MMAP_FLAGS           (MAP_PRIVATE)
+static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */
+#define MMAP_DEFAULT(s) ((dev_zero_fd < 0) ? \
+	   (dev_zero_fd = open("/dev/zero", O_RDWR), \
+	    mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0)) : \
+	    mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0))
+#endif /* MAP_ANONYMOUS */
+
+#define DIRECT_MMAP_DEFAULT(s) MMAP_DEFAULT(s)
+
+#else /* WIN32 */
+
+/* Win32 MMAP via VirtualAlloc */
+static FORCEINLINE void* win32mmap(size_t size) {
+  void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
+  return (ptr != 0)? ptr: MFAIL;
+}
+
+/* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
+static FORCEINLINE void* win32direct_mmap(size_t size) {
+  void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
+			   PAGE_READWRITE);
+  return (ptr != 0)? ptr: MFAIL;
+}
+
+/* This function supports releasing coalesed segments */
+static FORCEINLINE int win32munmap(void* ptr, size_t size) {
+  MEMORY_BASIC_INFORMATION minfo;
+  char* cptr = (char*)ptr;
+  while (size) {
+    if (VirtualQuery(cptr, &minfo, sizeof(minfo)) == 0)
+      return -1;
+    if (minfo.BaseAddress != cptr || minfo.AllocationBase != cptr ||
+	minfo.State != MEM_COMMIT || minfo.RegionSize > size)
+      return -1;
+    if (VirtualFree(cptr, 0, MEM_RELEASE) == 0)
+      return -1;
+    cptr += minfo.RegionSize;
+    size -= minfo.RegionSize;
+  }
+  return 0;
+}
+
+#define MMAP_DEFAULT(s)             win32mmap(s)
+#define MUNMAP_DEFAULT(a, s)        win32munmap((a), (s))
+#define DIRECT_MMAP_DEFAULT(s)      win32direct_mmap(s)
+#endif /* WIN32 */
+#endif /* HAVE_MMAP */
+
+#if HAVE_MREMAP
+#ifndef WIN32
+#define MREMAP_DEFAULT(addr, osz, nsz, mv) mremap((addr), (osz), (nsz), (mv))
+#endif /* WIN32 */
+#endif /* HAVE_MREMAP */
+
+
+/**
+ * Define CALL_MORECORE
+ */
+#if HAVE_MORECORE
+    #ifdef MORECORE
+	#define CALL_MORECORE(S)    MORECORE(S)
+    #else  /* MORECORE */
+	#define CALL_MORECORE(S)    MORECORE_DEFAULT(S)
+    #endif /* MORECORE */
+#else  /* HAVE_MORECORE */
+    #define CALL_MORECORE(S)        MFAIL
+#endif /* HAVE_MORECORE */
+
+/**
+ * Define CALL_MMAP/CALL_MUNMAP/CALL_DIRECT_MMAP
+ */
+#if HAVE_MMAP
+    #define IS_MMAPPED_BIT          (SIZE_T_ONE)
+    #define USE_MMAP_BIT            (SIZE_T_ONE)
+
+    #ifdef MMAP
+	#define CALL_MMAP(s)        MMAP(s)
+    #else /* MMAP */
+	#define CALL_MMAP(s)        MMAP_DEFAULT(s)
+    #endif /* MMAP */
+    #ifdef MUNMAP
+	#define CALL_MUNMAP(a, s)   MUNMAP((a), (s))
+    #else /* MUNMAP */
+	#define CALL_MUNMAP(a, s)   MUNMAP_DEFAULT((a), (s))
+    #endif /* MUNMAP */
+    #ifdef DIRECT_MMAP
+	#define CALL_DIRECT_MMAP(s) DIRECT_MMAP(s)
+    #else /* DIRECT_MMAP */
+	#define CALL_DIRECT_MMAP(s) DIRECT_MMAP_DEFAULT(s)
+    #endif /* DIRECT_MMAP */
+#else  /* HAVE_MMAP */
+    #define IS_MMAPPED_BIT          (SIZE_T_ZERO)
+    #define USE_MMAP_BIT            (SIZE_T_ZERO)
+
+    #define MMAP(s)                 MFAIL
+    #define MUNMAP(a, s)            (-1)
+    #define DIRECT_MMAP(s)          MFAIL
+    #define CALL_DIRECT_MMAP(s)     DIRECT_MMAP(s)
+    #define CALL_MMAP(s)            MMAP(s)
+    #define CALL_MUNMAP(a, s)       MUNMAP((a), (s))
+#endif /* HAVE_MMAP */
+
+/**
+ * Define CALL_MREMAP
+ */
+#if HAVE_MMAP && HAVE_MREMAP
+    #ifdef MREMAP
+	#define CALL_MREMAP(addr, osz, nsz, mv) MREMAP((addr), (osz), (nsz), (mv))
+    #else /* MREMAP */
+	#define CALL_MREMAP(addr, osz, nsz, mv) MREMAP_DEFAULT((addr), (osz), (nsz), (mv))
+    #endif /* MREMAP */
+#else  /* HAVE_MMAP && HAVE_MREMAP */
+    #define CALL_MREMAP(addr, osz, nsz, mv)     MFAIL
+#endif /* HAVE_MMAP && HAVE_MREMAP */
+
+/* mstate bit set if continguous morecore disabled or failed */
+#define USE_NONCONTIGUOUS_BIT (4U)
+
+/* segment bit set in create_mspace_with_base */
+#define EXTERN_BIT            (8U)
+
+
+/* --------------------------- Lock preliminaries ------------------------ */
+
+/*
+  When locks are defined, there is one global lock, plus
+  one per-mspace lock.
+
+  The global lock_ensures that mparams.magic and other unique
+  mparams values are initialized only once. It also protects
+  sequences of calls to MORECORE.  In many cases sys_alloc requires
+  two calls, that should not be interleaved with calls by other
+  threads.  This does not protect against direct calls to MORECORE
+  by other threads not using this lock, so there is still code to
+  cope the best we can on interference.
+
+  Per-mspace locks surround calls to malloc, free, etc.  To enable use
+  in layered extensions, per-mspace locks are reentrant.
+
+  Because lock-protected regions generally have bounded times, it is
+  OK to use the supplied simple spinlocks in the custom versions for
+  x86.
+
+  If USE_LOCKS is > 1, the definitions of lock routines here are
+  bypassed, in which case you will need to define at least
+  INITIAL_LOCK, ACQUIRE_LOCK, RELEASE_LOCK and possibly TRY_LOCK
+  (which is not used in this malloc, but commonly needed in
+  extensions.)
+*/
+
+#if USE_LOCKS == 1
+
+#if USE_SPIN_LOCKS
+#ifndef WIN32
+
+/* Custom pthread-style spin locks on x86 and x64 for gcc */
+struct pthread_mlock_t {
+  volatile unsigned int l;
+  volatile unsigned int c;
+  volatile pthread_t threadid;
+};
+#define MLOCK_T struct        pthread_mlock_t
+#define CURRENT_THREAD        pthread_self()
+#define INITIAL_LOCK(sl)      (memset(sl, 0, sizeof(MLOCK_T)), 0)
+#define ACQUIRE_LOCK(sl)      pthread_acquire_lock(sl)
+#define RELEASE_LOCK(sl)      pthread_release_lock(sl)
+#define TRY_LOCK(sl)          pthread_try_lock(sl)
+#define SPINS_PER_YIELD       63
+
+static MLOCK_T malloc_global_mutex = { 0, 0, 0};
+
+static FORCEINLINE int pthread_acquire_lock (MLOCK_T *sl) {
+  int spins = 0;
+  volatile unsigned int* lp = &sl->l;
+  for (;;) {
+    if (*lp != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 0;
+      }
+    }
+    else {
+      /* place args to cmpxchgl in locals to evade oddities in some gccs */
+      int cmp = 0;
+      int val = 1;
+      int ret;
+      __asm__ __volatile__  ("lock; cmpxchgl %1, %2"
+			     : "=a" (ret)
+			     : "r" (val), "m" (*(lp)), "0"(cmp)
+			     : "memory", "cc");
+      if (!ret) {
+	assert(!sl->threadid);
+	sl->c = 1;
+	sl->threadid = CURRENT_THREAD;
+	return 0;
+      }
+      if ((++spins & SPINS_PER_YIELD) == 0) {
+#if defined (__SVR4) && defined (__sun) /* solaris */
+	thr_yield();
+#else
+#if defined(__linux__) || defined(__FreeBSD__) || defined(__APPLE__)
+	sched_yield();
+#else  /* no-op yield on unknown systems */
+	;
+#endif /* __linux__ || __FreeBSD__ || __APPLE__ */
+#endif /* solaris */
+      }
+    }
+  }
+}
+
+static FORCEINLINE void pthread_release_lock (MLOCK_T *sl) {
+  assert(sl->l != 0);
+  assert(sl->threadid == CURRENT_THREAD);
+  if (--sl->c == 0) {
+    sl->threadid = 0;
+    volatile unsigned int* lp = &sl->l;
+    int prev = 0;
+    int ret;
+    __asm__ __volatile__ ("lock; xchgl %0, %1"
+			  : "=r" (ret)
+			  : "m" (*(lp)), "0"(prev)
+			  : "memory");
+  }
+}
+
+static FORCEINLINE int pthread_try_lock (MLOCK_T *sl) {
+  volatile unsigned int* lp = &sl->l;
+  if (*lp != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 1;
+      }
+  }
+  else {
+    int cmp = 0;
+    int val = 1;
+    int ret;
+    __asm__ __volatile__  ("lock; cmpxchgl %1, %2"
+			   : "=a" (ret)
+			   : "r" (val), "m" (*(lp)), "0"(cmp)
+			   : "memory", "cc");
+    if (!ret) {
+      assert(!sl->threadid);
+      sl->c = 1;
+      sl->threadid = CURRENT_THREAD;
+      return 1;
+    }
+  }
+  return 0;
+}
+
+
+#else /* WIN32 */
+/* Custom win32-style spin locks on x86 and x64 for MSC */
+struct win32_mlock_t
+{
+  volatile long l;
+  volatile unsigned int c;
+  volatile long threadid;
+};
+
+static inline int return_0(int i) { return 0; }
+#define MLOCK_T               struct win32_mlock_t
+#define CURRENT_THREAD        win32_getcurrentthreadid()
+#define INITIAL_LOCK(sl)      (memset(sl, 0, sizeof(MLOCK_T)), return_0(0))
+#define ACQUIRE_LOCK(sl)      win32_acquire_lock(sl)
+#define RELEASE_LOCK(sl)      win32_release_lock(sl)
+#define TRY_LOCK(sl)          win32_try_lock(sl)
+#define SPINS_PER_YIELD       63
+
+static MLOCK_T malloc_global_mutex = { 0, 0, 0};
+
+static FORCEINLINE long win32_getcurrentthreadid(void) {
+#ifdef _MSC_VER
+#if defined(_M_IX86)
+  long *threadstruct=(long *)__readfsdword(0x18);
+  long threadid=threadstruct[0x24/sizeof(long)];
+  return threadid;
+#elif defined(_M_X64)
+  /* todo */
+  return GetCurrentThreadId();
+#else
+  return GetCurrentThreadId();
+#endif
+#else
+  return GetCurrentThreadId();
+#endif
+}
+
+static FORCEINLINE int win32_acquire_lock (MLOCK_T *sl) {
+  int spins = 0;
+  for (;;) {
+    if (sl->l != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 0;
+      }
+    }
+    else {
+      if (!interlockedexchange(&sl->l, 1)) {
+	assert(!sl->threadid);
+		sl->c=CURRENT_THREAD;
+	sl->threadid = CURRENT_THREAD;
+	sl->c = 1;
+	return 0;
+      }
+    }
+    if ((++spins & SPINS_PER_YIELD) == 0)
+      SleepEx(0, FALSE);
+  }
+}
+
+static FORCEINLINE void win32_release_lock (MLOCK_T *sl) {
+  assert(sl->threadid == CURRENT_THREAD);
+  assert(sl->l != 0);
+  if (--sl->c == 0) {
+    sl->threadid = 0;
+    interlockedexchange (&sl->l, 0);
+  }
+}
+
+static FORCEINLINE int win32_try_lock (MLOCK_T *sl) {
+  if(sl->l != 0) {
+      if (sl->threadid == CURRENT_THREAD) {
+	++sl->c;
+	return 1;
+      }
+  }
+  else {
+    if (!interlockedexchange(&sl->l, 1)){
+      assert(!sl->threadid);
+      sl->threadid = CURRENT_THREAD;
+      sl->c = 1;
+      return 1;
+    }
+  }
+  return 0;
+}
+
+#endif /* WIN32 */
+#else /* USE_SPIN_LOCKS */
+
+#ifndef WIN32
+/* pthreads-based locks */
+
+#define MLOCK_T               pthread_mutex_t
+#define CURRENT_THREAD        pthread_self()
+#define INITIAL_LOCK(sl)      pthread_init_lock(sl)
+#define ACQUIRE_LOCK(sl)      pthread_mutex_lock(sl)
+#define RELEASE_LOCK(sl)      pthread_mutex_unlock(sl)
+#define TRY_LOCK(sl)          (!pthread_mutex_trylock(sl))
+
+static MLOCK_T malloc_global_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+/* Cope with old-style linux recursive lock initialization by adding */
+/* skipped internal declaration from pthread.h */
+#ifdef linux
+#ifndef PTHREAD_MUTEX_RECURSIVE
+extern int pthread_mutexattr_setkind_np __P ((pthread_mutexattr_t *__attr,
+					   int __kind));
+#define PTHREAD_MUTEX_RECURSIVE PTHREAD_MUTEX_RECURSIVE_NP
+#define pthread_mutexattr_settype(x,y) pthread_mutexattr_setkind_np(x,y)
+#endif
+#endif
+
+static int pthread_init_lock (MLOCK_T *sl) {
+  pthread_mutexattr_t attr;
+  if (pthread_mutexattr_init(&attr)) return 1;
+  if (pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE)) return 1;
+  if (pthread_mutex_init(sl, &attr)) return 1;
+  if (pthread_mutexattr_destroy(&attr)) return 1;
+  return 0;
+}
+
+#else /* WIN32 */
+/* Win32 critical sections */
+#define MLOCK_T               CRITICAL_SECTION
+#define CURRENT_THREAD        GetCurrentThreadId()
+#define INITIAL_LOCK(s)       (!InitializeCriticalSectionAndSpinCount((s), 0x80000000|4000))
+#define ACQUIRE_LOCK(s)       (EnterCriticalSection(s), 0)
+#define RELEASE_LOCK(s)       LeaveCriticalSection(s)
+#define TRY_LOCK(s)           TryEnterCriticalSection(s)
+#define NEED_GLOBAL_LOCK_INIT
+
+static MLOCK_T malloc_global_mutex;
+static volatile long malloc_global_mutex_status;
+
+/* Use spin loop to initialize global lock */
+static void init_malloc_global_mutex() {
+  for (;;) {
+    long stat = malloc_global_mutex_status;
+    if (stat > 0)
+      return;
+    /* transition to < 0 while initializing, then to > 0) */
+    if (stat == 0 &&
+	interlockedcompareexchange(&malloc_global_mutex_status, -1, 0) == 0) {
+      InitializeCriticalSection(&malloc_global_mutex);
+      interlockedexchange(&malloc_global_mutex_status,1);
+      return;
+    }
+    SleepEx(0, FALSE);
+  }
+}
+
+#endif /* WIN32 */
+#endif /* USE_SPIN_LOCKS */
+#endif /* USE_LOCKS == 1 */
+
+/* -----------------------  User-defined locks ------------------------ */
+
+#if USE_LOCKS > 1
+/* Define your own lock implementation here */
+/* #define INITIAL_LOCK(sl)  ... */
+/* #define ACQUIRE_LOCK(sl)  ... */
+/* #define RELEASE_LOCK(sl)  ... */
+/* #define TRY_LOCK(sl) ... */
+/* static MLOCK_T malloc_global_mutex = ... */
+#endif /* USE_LOCKS > 1 */
+
+/* -----------------------  Lock-based state ------------------------ */
+
+#if USE_LOCKS
+#define USE_LOCK_BIT               (2U)
+#else  /* USE_LOCKS */
+#define USE_LOCK_BIT               (0U)
+#define INITIAL_LOCK(l)
+#endif /* USE_LOCKS */
+
+#if USE_LOCKS
+#define ACQUIRE_MALLOC_GLOBAL_LOCK()  ACQUIRE_LOCK(&malloc_global_mutex);
+#define RELEASE_MALLOC_GLOBAL_LOCK()  RELEASE_LOCK(&malloc_global_mutex);
+#else  /* USE_LOCKS */
+#define ACQUIRE_MALLOC_GLOBAL_LOCK()
+#define RELEASE_MALLOC_GLOBAL_LOCK()
+#endif /* USE_LOCKS */
+
+
+/* -----------------------  Chunk representations ------------------------ */
+
+/*
+  (The following includes lightly edited explanations by Colin Plumb.)
+
+  The malloc_chunk declaration below is misleading (but accurate and
+  necessary).  It declares a "view" into memory allowing access to
+  necessary fields at known offsets from a given base.
+
+  Chunks of memory are maintained using a `boundary tag' method as
+  originally described by Knuth.  (See the paper by Paul Wilson
+  ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a survey of such
+  techniques.)  Sizes of free chunks are stored both in the front of
+  each chunk and at the end.  This makes consolidating fragmented
+  chunks into bigger chunks fast.  The head fields also hold bits
+  representing whether chunks are free or in use.
+
+  Here are some pictures to make it clearer.  They are "exploded" to
+  show that the state of a chunk can be thought of as extending from
+  the high 31 bits of the head field of its header through the
+  prev_foot and PINUSE_BIT bit of the following chunk header.
+
+  A chunk that's in use looks like:
+
+   chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	   | Size of previous chunk (if P = 0)                             |
+	   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
+	 | Size of this chunk                                         1| +-+
+   mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 |                                                               |
+	 +-                                                             -+
+	 |                                                               |
+	 +-                                                             -+
+	 |                                                               :
+	 +-      size - sizeof(size_t) available payload bytes          -+
+	 :                                                               |
+ chunk-> +-                                                             -+
+	 |                                                               |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|
+       | Size of next chunk (may or may not be in use)               | +-+
+ mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+    And if it's free, it looks like this:
+
+   chunk-> +-                                                             -+
+	   | User payload (must be in use, or we would have merged!)       |
+	   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
+	 | Size of this chunk                                         0| +-+
+   mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 | Next pointer                                                  |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 | Prev pointer                                                  |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 |                                                               :
+	 +-      size - sizeof(struct chunk) unused bytes               -+
+	 :                                                               |
+ chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	 | Size of this chunk                                            |
+	 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|
+       | Size of next chunk (must be in use, or we would have merged)| +-+
+ mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+       |                                                               :
+       +- User payload                                                -+
+       :                                                               |
+       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+								     |0|
+								     +-+
+  Note that since we always merge adjacent free chunks, the chunks
+  adjacent to a free chunk must be in use.
+
+  Given a pointer to a chunk (which can be derived trivially from the
+  payload pointer) we can, in O(1) time, find out whether the adjacent
+  chunks are free, and if so, unlink them from the lists that they
+  are on and merge them with the current chunk.
+
+  Chunks always begin on even word boundaries, so the mem portion
+  (which is returned to the user) is also on an even word boundary, and
+  thus at least double-word aligned.
+
+  The P (PINUSE_BIT) bit, stored in the unused low-order bit of the
+  chunk size (which is always a multiple of two words), is an in-use
+  bit for the *previous* chunk.  If that bit is *clear*, then the
+  word before the current chunk size contains the previous chunk
+  size, and can be used to find the front of the previous chunk.
+  The very first chunk allocated always has this bit set, preventing
+  access to non-existent (or non-owned) memory. If pinuse is set for
+  any given chunk, then you CANNOT determine the size of the
+  previous chunk, and might even get a memory addressing fault when
+  trying to do so.
+
+  The C (CINUSE_BIT) bit, stored in the unused second-lowest bit of
+  the chunk size redundantly records whether the current chunk is
+  inuse. This redundancy enables usage checks within free and realloc,
+  and reduces indirection when freeing and consolidating chunks.
+
+  Each freshly allocated chunk must have both cinuse and pinuse set.
+  That is, each allocated chunk borders either a previously allocated
+  and still in-use chunk, or the base of its memory arena. This is
+  ensured by making all allocations from the `lowest' part of any
+  found chunk.  Further, no free chunk physically borders another one,
+  so each free chunk is known to be preceded and followed by either
+  inuse chunks or the ends of memory.
+
+  Note that the `foot' of the current chunk is actually represented
+  as the prev_foot of the NEXT chunk. This makes it easier to
+  deal with alignments etc but can be very confusing when trying
+  to extend or adapt this code.
+
+  The exceptions to all this are
+
+     1. The special chunk `top' is the top-most available chunk (i.e.,
+	the one bordering the end of available memory). It is treated
+	specially.  Top is never included in any bin, is used only if
+	no other chunk is available, and is released back to the
+	system if it is very large (see M_TRIM_THRESHOLD).  In effect,
+	the top chunk is treated as larger (and thus less well
+	fitting) than any other available chunk.  The top chunk
+	doesn't update its trailing size field since there is no next
+	contiguous chunk that would have to index off it. However,
+	space is still allocated for it (TOP_FOOT_SIZE) to enable
+	separation or merging when space is extended.
+
+     3. Chunks allocated via mmap, which have the lowest-order bit
+	(IS_MMAPPED_BIT) set in their prev_foot fields, and do not set
+	PINUSE_BIT in their head fields.  Because they are allocated
+	one-by-one, each must carry its own prev_foot field, which is
+	also used to hold the offset this chunk has within its mmapped
+	region, which is needed to preserve alignment. Each mmapped
+	chunk is trailed by the first two fields of a fake next-chunk
+	for sake of usage checks.
+
+*/
+
+struct malloc_chunk {
+  size_t               prev_foot;  /* Size of previous chunk (if free).  */
+  size_t               head;       /* Size and inuse bits. */
+  struct malloc_chunk* fd;         /* double links -- used only if free. */
+  struct malloc_chunk* bk;
+};
+
+typedef struct malloc_chunk  mchunk;
+typedef struct malloc_chunk* mchunkptr;
+typedef struct malloc_chunk* sbinptr;  /* The type of bins of chunks */
+typedef unsigned int bindex_t;         /* Described below */
+typedef unsigned int binmap_t;         /* Described below */
+typedef unsigned int flag_t;           /* The type of various bit flag sets */
+
+/* ------------------- Chunks sizes and alignments ----------------------- */
+
+#define MCHUNK_SIZE         (sizeof(mchunk))
+
+#if FOOTERS
+#define CHUNK_OVERHEAD      (TWO_SIZE_T_SIZES)
+#else /* FOOTERS */
+#define CHUNK_OVERHEAD      (SIZE_T_SIZE)
+#endif /* FOOTERS */
+
+/* MMapped chunks need a second word of overhead ... */
+#define MMAP_CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
+/* ... and additional padding for fake next-chunk at foot */
+#define MMAP_FOOT_PAD       (FOUR_SIZE_T_SIZES)
+
+/* The smallest size we can malloc is an aligned minimal chunk */
+#define MIN_CHUNK_SIZE\
+  ((MCHUNK_SIZE + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
+
+/* conversion from malloc headers to user pointers, and back */
+#define chunk2mem(p)        ((void*)((char*)(p)       + TWO_SIZE_T_SIZES))
+#define mem2chunk(mem)      ((mchunkptr)((char*)(mem) - TWO_SIZE_T_SIZES))
+/* chunk associated with aligned address A */
+#define align_as_chunk(A)   (mchunkptr)((A) + align_offset(chunk2mem(A)))
+
+/* Bounds on request (not chunk) sizes. */
+#define MAX_REQUEST         ((-MIN_CHUNK_SIZE) << 2)
+#define MIN_REQUEST         (MIN_CHUNK_SIZE - CHUNK_OVERHEAD - SIZE_T_ONE)
+
+/* pad request bytes into a usable size */
+#define pad_request(req) \
+   (((req) + CHUNK_OVERHEAD + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
+
+/* pad request, checking for minimum (but not maximum) */
+#define request2size(req) \
+  (((req) < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(req))
+
+
+/* ------------------ Operations on head and foot fields ----------------- */
+
+/*
+  The head field of a chunk is or'ed with PINUSE_BIT when previous
+  adjacent chunk in use, and or'ed with CINUSE_BIT if this chunk is in
+  use. If the chunk was obtained with mmap, the prev_foot field has
+  IS_MMAPPED_BIT set, otherwise holding the offset of the base of the
+  mmapped region to the base of the chunk.
+
+  FLAG4_BIT is not used by this malloc, but might be useful in extensions.
+*/
+
+#define PINUSE_BIT          (SIZE_T_ONE)
+#define CINUSE_BIT          (SIZE_T_TWO)
+#define FLAG4_BIT           (SIZE_T_FOUR)
+#define INUSE_BITS          (PINUSE_BIT|CINUSE_BIT)
+#define FLAG_BITS           (PINUSE_BIT|CINUSE_BIT|FLAG4_BIT)
+
+/* Head value for fenceposts */
+#define FENCEPOST_HEAD      (INUSE_BITS|SIZE_T_SIZE)
+
+/* extraction of fields from head words */
+#define cinuse(p)           ((p)->head & CINUSE_BIT)
+#define pinuse(p)           ((p)->head & PINUSE_BIT)
+#define chunksize(p)        ((p)->head & ~(FLAG_BITS))
+
+#define clear_pinuse(p)     ((p)->head &= ~PINUSE_BIT)
+#define clear_cinuse(p)     ((p)->head &= ~CINUSE_BIT)
+
+/* Treat space at ptr +/- offset as a chunk */
+#define chunk_plus_offset(p, s)  ((mchunkptr)(((char*)(p)) + (s)))
+#define chunk_minus_offset(p, s) ((mchunkptr)(((char*)(p)) - (s)))
+
+/* Ptr to next or previous physical malloc_chunk. */
+#define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->head & ~FLAG_BITS)))
+#define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_foot) ))
+
+/* extract next chunk's pinuse bit */
+#define next_pinuse(p)  ((next_chunk(p)->head) & PINUSE_BIT)
+
+/* Get/set size at footer */
+#define get_foot(p, s)  (((mchunkptr)((char*)(p) + (s)))->prev_foot)
+#define set_foot(p, s)  (((mchunkptr)((char*)(p) + (s)))->prev_foot = (s))
+
+/* Set size, pinuse bit, and foot */
+#define set_size_and_pinuse_of_free_chunk(p, s)\
+  ((p)->head = (s|PINUSE_BIT), set_foot(p, s))
+
+/* Set size, pinuse bit, foot, and clear next pinuse */
+#define set_free_with_pinuse(p, s, n)\
+  (clear_pinuse(n), set_size_and_pinuse_of_free_chunk(p, s))
+
+#define is_mmapped(p)\
+  (!((p)->head & PINUSE_BIT) && ((p)->prev_foot & IS_MMAPPED_BIT))
+
+/* Get the internal overhead associated with chunk p */
+#define overhead_for(p)\
+ (is_mmapped(p)? MMAP_CHUNK_OVERHEAD : CHUNK_OVERHEAD)
+
+/* Return true if malloced space is not necessarily cleared */
+#if MMAP_CLEARS
+#define calloc_must_clear(p) (!is_mmapped(p))
+#else /* MMAP_CLEARS */
+#define calloc_must_clear(p) (1)
+#endif /* MMAP_CLEARS */
+
+/* ---------------------- Overlaid data structures ----------------------- */
+
+/*
+  When chunks are not in use, they are treated as nodes of either
+  lists or trees.
+
+  "Small"  chunks are stored in circular doubly-linked lists, and look
+  like this:
+
+    chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Size of previous chunk                            |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `head:' |             Size of chunk, in bytes                         |P|
+      mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Forward pointer to next chunk in list             |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Back pointer to previous chunk in list            |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Unused space (may be 0 bytes long)                .
+	    .                                                               .
+	    .                                                               |
+nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `foot:' |             Size of chunk, in bytes                           |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+  Larger chunks are kept in a form of bitwise digital trees (aka
+  tries) keyed on chunksizes.  Because malloc_tree_chunks are only for
+  free chunks greater than 256 bytes, their size doesn't impose any
+  constraints on user chunk sizes.  Each node looks like:
+
+    chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Size of previous chunk                            |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `head:' |             Size of chunk, in bytes                         |P|
+      mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Forward pointer to next chunk of same size        |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Back pointer to previous chunk of same size       |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Pointer to left child (child[0])                  |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Pointer to right child (child[1])                 |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Pointer to parent                                 |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             bin index of this chunk                           |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+	    |             Unused space                                      .
+	    .                                                               |
+nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+    `foot:' |             Size of chunk, in bytes                           |
+	    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+
+  Each tree holding treenodes is a tree of unique chunk sizes.  Chunks
+  of the same size are arranged in a circularly-linked list, with only
+  the oldest chunk (the next to be used, in our FIFO ordering)
+  actually in the tree.  (Tree members are distinguished by a non-null
+  parent pointer.)  If a chunk with the same size as an existing node
+  is inserted, it is linked off the existing node using pointers that
+  work in the same way as fd/bk pointers of small chunks.
+
+  Each tree contains a power of 2 sized range of chunk sizes (the
+  smallest is 0x100 <= x < 0x180), which is divided in half at each
+  tree level, with the chunks in the smaller half of the range (0x100
+  <= x < 0x140 for the top nose) in the left subtree and the larger
+  half (0x140 <= x < 0x180) in the right subtree.  This is, of course,
+  done by inspecting individual bits.
+
+  Using these rules, each node's left subtree contains all smaller
+  sizes than its right subtree.  However, the node at the root of each
+  subtree has no particular ordering relationship to either.  (The
+  dividing line between the subtree sizes is based on trie relation.)
+  If we remove the last chunk of a given size from the interior of the
+  tree, we need to replace it with a leaf node.  The tree ordering
+  rules permit a node to be replaced by any leaf below it.
+
+  The smallest chunk in a tree (a common operation in a best-fit
+  allocator) can be found by walking a path to the leftmost leaf in
+  the tree.  Unlike a usual binary tree, where we follow left child
+  pointers until we reach a null, here we follow the right child
+  pointer any time the left one is null, until we reach a leaf with
+  both child pointers null. The smallest chunk in the tree will be
+  somewhere along that path.
+
+  The worst case number of steps to add, find, or remove a node is
+  bounded by the number of bits differentiating chunks within
+  bins. Under current bin calculations, this ranges from 6 up to 21
+  (for 32 bit sizes) or up to 53 (for 64 bit sizes). The typical case
+  is of course much better.
+*/
+
+struct malloc_tree_chunk {
+  /* The first four fields must be compatible with malloc_chunk */
+  size_t                    prev_foot;
+  size_t                    head;
+  struct malloc_tree_chunk* fd;
+  struct malloc_tree_chunk* bk;
+
+  struct malloc_tree_chunk* child[2];
+  struct malloc_tree_chunk* parent;
+  bindex_t                  index;
+};
+
+typedef struct malloc_tree_chunk  tchunk;
+typedef struct malloc_tree_chunk* tchunkptr;
+typedef struct malloc_tree_chunk* tbinptr; /* The type of bins of trees */
+
+/* A little helper macro for trees */
+#define leftmost_child(t) ((t)->child[0] != 0? (t)->child[0] : (t)->child[1])
+
+/* ----------------------------- Segments -------------------------------- */
+
+/*
+  Each malloc space may include non-contiguous segments, held in a
+  list headed by an embedded malloc_segment record representing the
+  top-most space. Segments also include flags holding properties of
+  the space. Large chunks that are directly allocated by mmap are not
+  included in this list. They are instead independently created and
+  destroyed without otherwise keeping track of them.
+
+  Segment management mainly comes into play for spaces allocated by
+  MMAP.  Any call to MMAP might or might not return memory that is
+  adjacent to an existing segment.  MORECORE normally contiguously
+  extends the current space, so this space is almost always adjacent,
+  which is simpler and faster to deal with. (This is why MORECORE is
+  used preferentially to MMAP when both are available -- see
+  sys_alloc.)  When allocating using MMAP, we don't use any of the
+  hinting mechanisms (inconsistently) supported in various
+  implementations of unix mmap, or distinguish reserving from
+  committing memory. Instead, we just ask for space, and exploit
+  contiguity when we get it.  It is probably possible to do
+  better than this on some systems, but no general scheme seems
+  to be significantly better.
+
+  Management entails a simpler variant of the consolidation scheme
+  used for chunks to reduce fragmentation -- new adjacent memory is
+  normally prepended or appended to an existing segment. However,
+  there are limitations compared to chunk consolidation that mostly
+  reflect the fact that segment processing is relatively infrequent
+  (occurring only when getting memory from system) and that we
+  don't expect to have huge numbers of segments:
+
+  * Segments are not indexed, so traversal requires linear scans.  (It
+    would be possible to index these, but is not worth the extra
+    overhead and complexity for most programs on most platforms.)
+  * New segments are only appended to old ones when holding top-most
+    memory; if they cannot be prepended to others, they are held in
+    different segments.
+
+  Except for the top-most segment of an mstate, each segment record
+  is kept at the tail of its segment. Segments are added by pushing
+  segment records onto the list headed by &mstate.seg for the
+  containing mstate.
+
+  Segment flags control allocation/merge/deallocation policies:
+  * If EXTERN_BIT set, then we did not allocate this segment,
+    and so should not try to deallocate or merge with others.
+    (This currently holds only for the initial segment passed
+    into create_mspace_with_base.)
+  * If IS_MMAPPED_BIT set, the segment may be merged with
+    other surrounding mmapped segments and trimmed/de-allocated
+    using munmap.
+  * If neither bit is set, then the segment was obtained using
+    MORECORE so can be merged with surrounding MORECORE'd segments
+    and deallocated/trimmed using MORECORE with negative arguments.
+*/
+
+struct malloc_segment {
+  char*        base;             /* base address */
+  size_t       size;             /* allocated size */
+  struct malloc_segment* next;   /* ptr to next segment */
+  flag_t       sflags;           /* mmap and extern flag */
+};
+
+#define is_mmapped_segment(S)  ((S)->sflags & IS_MMAPPED_BIT)
+#define is_extern_segment(S)   ((S)->sflags & EXTERN_BIT)
+
+typedef struct malloc_segment  msegment;
+typedef struct malloc_segment* msegmentptr;
+
+/* ---------------------------- malloc_state ----------------------------- */
+
+/*
+   A malloc_state holds all of the bookkeeping for a space.
+   The main fields are:
+
+  Top
+    The topmost chunk of the currently active segment. Its size is
+    cached in topsize.  The actual size of topmost space is
+    topsize+TOP_FOOT_SIZE, which includes space reserved for adding
+    fenceposts and segment records if necessary when getting more
+    space from the system.  The size at which to autotrim top is
+    cached from mparams in trim_check, except that it is disabled if
+    an autotrim fails.
+
+  Designated victim (dv)
+    This is the preferred chunk for servicing small requests that
+    don't have exact fits.  It is normally the chunk split off most
+    recently to service another small request.  Its size is cached in
+    dvsize. The link fields of this chunk are not maintained since it
+    is not kept in a bin.
+
+  SmallBins
+    An array of bin headers for free chunks.  These bins hold chunks
+    with sizes less than MIN_LARGE_SIZE bytes. Each bin contains
+    chunks of all the same size, spaced 8 bytes apart.  To simplify
+    use in double-linked lists, each bin header acts as a malloc_chunk
+    pointing to the real first node, if it exists (else pointing to
+    itself).  This avoids special-casing for headers.  But to avoid
+    waste, we allocate only the fd/bk pointers of bins, and then use
+    repositioning tricks to treat these as the fields of a chunk.
+
+  TreeBins
+    Treebins are pointers to the roots of trees holding a range of
+    sizes. There are 2 equally spaced treebins for each power of two
+    from TREE_SHIFT to TREE_SHIFT+16. The last bin holds anything
+    larger.
+
+  Bin maps
+    There is one bit map for small bins ("smallmap") and one for
+    treebins ("treemap).  Each bin sets its bit when non-empty, and
+    clears the bit when empty.  Bit operations are then used to avoid
+    bin-by-bin searching -- nearly all "search" is done without ever
+    looking at bins that won't be selected.  The bit maps
+    conservatively use 32 bits per map word, even if on 64bit system.
+    For a good description of some of the bit-based techniques used
+    here, see Henry S. Warren Jr's book "Hacker's Delight" (and
+    supplement at http://hackersdelight.org/). Many of these are
+    intended to reduce the branchiness of paths through malloc etc, as
+    well as to reduce the number of memory locations read or written.
+
+  Segments
+    A list of segments headed by an embedded malloc_segment record
+    representing the initial space.
+
+  Address check support
+    The least_addr field is the least address ever obtained from
+    MORECORE or MMAP. Attempted frees and reallocs of any address less
+    than this are trapped (unless INSECURE is defined).
+
+  Magic tag
+    A cross-check field that should always hold same value as mparams.magic.
+
+  Flags
+    Bits recording whether to use MMAP, locks, or contiguous MORECORE
+
+  Statistics
+    Each space keeps track of current and maximum system memory
+    obtained via MORECORE or MMAP.
+
+  Trim support
+    Fields holding the amount of unused topmost memory that should trigger
+    timming, and a counter to force periodic scanning to release unused
+    non-topmost segments.
+
+  Locking
+    If USE_LOCKS is defined, the "mutex" lock is acquired and released
+    around every public call using this mspace.
+
+  Extension support
+    A void* pointer and a size_t field that can be used to help implement
+    extensions to this malloc.
+*/
+
+/* Bin types, widths and sizes */
+#define NSMALLBINS        (32U)
+#define NTREEBINS         (32U)
+#define SMALLBIN_SHIFT    (3U)
+#define SMALLBIN_WIDTH    (SIZE_T_ONE << SMALLBIN_SHIFT)
+#define TREEBIN_SHIFT     (8U)
+#define MIN_LARGE_SIZE    (SIZE_T_ONE << TREEBIN_SHIFT)
+#define MAX_SMALL_SIZE    (MIN_LARGE_SIZE - SIZE_T_ONE)
+#define MAX_SMALL_REQUEST (MAX_SMALL_SIZE - CHUNK_ALIGN_MASK - CHUNK_OVERHEAD)
+
+struct malloc_state {
+  binmap_t   smallmap;
+  binmap_t   treemap;
+  size_t     dvsize;
+  size_t     topsize;
+  char*      least_addr;
+  mchunkptr  dv;
+  mchunkptr  top;
+  size_t     trim_check;
+  size_t     release_checks;
+  size_t     magic;
+  mchunkptr  smallbins[(NSMALLBINS+1)*2];
+  tbinptr    treebins[NTREEBINS];
+  size_t     footprint;
+  size_t     max_footprint;
+  flag_t     mflags;
+#if USE_LOCKS
+  MLOCK_T    mutex;     /* locate lock among fields that rarely change */
+#endif /* USE_LOCKS */
+  msegment   seg;
+  void*      extp;      /* Unused but available for extensions */
+  size_t     exts;
+};
+
+typedef struct malloc_state*    mstate;
+
+/* ------------- Global malloc_state and malloc_params ------------------- */
+
+/*
+  malloc_params holds global properties, including those that can be
+  dynamically set using mallopt. There is a single instance, mparams,
+  initialized in init_mparams. Note that the non-zeroness of "magic"
+  also serves as an initialization flag.
+*/
+
+struct malloc_params {
+  volatile size_t magic;
+  size_t page_size;
+  size_t granularity;
+  size_t mmap_threshold;
+  size_t trim_threshold;
+  flag_t default_mflags;
+};
+
+static struct malloc_params mparams;
+
+/* Ensure mparams initialized */
+#define ensure_initialization() ((void)(mparams.magic != 0 || init_mparams()))
+
+#if !ONLY_MSPACES
+
+/* The global malloc_state used for all non-"mspace" calls */
+static struct malloc_state _gm_;
+#define gm                 (&_gm_)
+#define is_global(M)       ((M) == &_gm_)
+
+#endif /* !ONLY_MSPACES */
+
+#define is_initialized(M)  ((M)->top != 0)
+
+/* -------------------------- system alloc setup ------------------------- */
+
+/* Operations on mflags */
+
+#define use_lock(M)           ((M)->mflags &   USE_LOCK_BIT)
+#define enable_lock(M)        ((M)->mflags |=  USE_LOCK_BIT)
+#define disable_lock(M)       ((M)->mflags &= ~USE_LOCK_BIT)
+
+#define use_mmap(M)           ((M)->mflags &   USE_MMAP_BIT)
+#define enable_mmap(M)        ((M)->mflags |=  USE_MMAP_BIT)
+#define disable_mmap(M)       ((M)->mflags &= ~USE_MMAP_BIT)
+
+#define use_noncontiguous(M)  ((M)->mflags &   USE_NONCONTIGUOUS_BIT)
+#define disable_contiguous(M) ((M)->mflags |=  USE_NONCONTIGUOUS_BIT)
+
+#define set_lock(M,L)\
+ ((M)->mflags = (L)?\
+  ((M)->mflags | USE_LOCK_BIT) :\
+  ((M)->mflags & ~USE_LOCK_BIT))
+
+/* page-align a size */
+#define page_align(S)\
+ (((S) + (mparams.page_size - SIZE_T_ONE)) & ~(mparams.page_size - SIZE_T_ONE))
+
+/* granularity-align a size */
+#define granularity_align(S)\
+  (((S) + (mparams.granularity - SIZE_T_ONE))\
+   & ~(mparams.granularity - SIZE_T_ONE))
+
+
+/* For mmap, use granularity alignment on windows, else page-align */
+#ifdef WIN32
+#define mmap_align(S) granularity_align(S)
+#else
+#define mmap_align(S) page_align(S)
+#endif
+
+/* For sys_alloc, enough padding to ensure can malloc request on success */
+#define SYS_ALLOC_PADDING (TOP_FOOT_SIZE + MALLOC_ALIGNMENT)
+
+#define is_page_aligned(S)\
+   (((size_t)(S) & (mparams.page_size - SIZE_T_ONE)) == 0)
+#define is_granularity_aligned(S)\
+   (((size_t)(S) & (mparams.granularity - SIZE_T_ONE)) == 0)
+
+/*  True if segment S holds address A */
+#define segment_holds(S, A)\
+  ((char*)(A) >= S->base && (char*)(A) < S->base + S->size)
+
+/* Return segment holding given address */
+static msegmentptr segment_holding(mstate m, char* addr) {
+  msegmentptr sp = &m->seg;
+  for (;;) {
+    if (addr >= sp->base && addr < sp->base + sp->size)
+      return sp;
+    if ((sp = sp->next) == 0)
+      return 0;
+  }
+}
+
+/* Return true if segment contains a segment link */
+static int has_segment_link(mstate m, msegmentptr ss) {
+  msegmentptr sp = &m->seg;
+  for (;;) {
+    if ((char*)sp >= ss->base && (char*)sp < ss->base + ss->size)
+      return 1;
+    if ((sp = sp->next) == 0)
+      return 0;
+  }
+}
+
+#ifndef MORECORE_CANNOT_TRIM
+#define should_trim(M,s)  ((s) > (M)->trim_check)
+#else  /* MORECORE_CANNOT_TRIM */
+#define should_trim(M,s)  (0)
+#endif /* MORECORE_CANNOT_TRIM */
+
+/*
+  TOP_FOOT_SIZE is padding at the end of a segment, including space
+  that may be needed to place segment records and fenceposts when new
+  noncontiguous segments are added.
+*/
+#define TOP_FOOT_SIZE\
+  (align_offset(chunk2mem(0))+pad_request(sizeof(struct malloc_segment))+MIN_CHUNK_SIZE)
+
+
+/* -------------------------------  Hooks -------------------------------- */
+
+/*
+  PREACTION should be defined to return 0 on success, and nonzero on
+  failure. If you are not using locking, you can redefine these to do
+  anything you like.
+*/
+
+#if USE_LOCKS
+
+#define PREACTION(M)  ((use_lock(M))? ACQUIRE_LOCK(&(M)->mutex) : 0)
+#define POSTACTION(M) { if (use_lock(M)) RELEASE_LOCK(&(M)->mutex); }
+#else /* USE_LOCKS */
+
+#ifndef PREACTION
+#define PREACTION(M) (0)
+#endif  /* PREACTION */
+
+#ifndef POSTACTION
+#define POSTACTION(M)
+#endif  /* POSTACTION */
+
+#endif /* USE_LOCKS */
+
+/*
+  CORRUPTION_ERROR_ACTION is triggered upon detected bad addresses.
+  USAGE_ERROR_ACTION is triggered on detected bad frees and
+  reallocs. The argument p is an address that might have triggered the
+  fault. It is ignored by the two predefined actions, but might be
+  useful in custom actions that try to help diagnose errors.
+*/
+
+#if PROCEED_ON_ERROR
+
+/* A count of the number of corruption errors causing resets */
+int malloc_corruption_error_count;
+
+/* default corruption action */
+static void reset_on_error(mstate m);
+
+#define CORRUPTION_ERROR_ACTION(m)  reset_on_error(m)
+#define USAGE_ERROR_ACTION(m, p)
+
+#else /* PROCEED_ON_ERROR */
+
+#ifndef CORRUPTION_ERROR_ACTION
+#define CORRUPTION_ERROR_ACTION(m) ABORT
+#endif /* CORRUPTION_ERROR_ACTION */
+
+#ifndef USAGE_ERROR_ACTION
+#define USAGE_ERROR_ACTION(m,p) ABORT
+#endif /* USAGE_ERROR_ACTION */
+
+#endif /* PROCEED_ON_ERROR */
+
+/* -------------------------- Debugging setup ---------------------------- */
+
+#if ! DEBUG
+
+#define check_free_chunk(M,P)
+#define check_inuse_chunk(M,P)
+#define check_malloced_chunk(M,P,N)
+#define check_mmapped_chunk(M,P)
+#define check_malloc_state(M)
+#define check_top_chunk(M,P)
+
+#else /* DEBUG */
+#define check_free_chunk(M,P)       do_check_free_chunk(M,P)
+#define check_inuse_chunk(M,P)      do_check_inuse_chunk(M,P)
+#define check_top_chunk(M,P)        do_check_top_chunk(M,P)
+#define check_malloced_chunk(M,P,N) do_check_malloced_chunk(M,P,N)
+#define check_mmapped_chunk(M,P)    do_check_mmapped_chunk(M,P)
+#define check_malloc_state(M)       do_check_malloc_state(M)
+
+static void   do_check_any_chunk(mstate m, mchunkptr p);
+static void   do_check_top_chunk(mstate m, mchunkptr p);
+static void   do_check_mmapped_chunk(mstate m, mchunkptr p);
+static void   do_check_inuse_chunk(mstate m, mchunkptr p);
+static void   do_check_free_chunk(mstate m, mchunkptr p);
+static void   do_check_malloced_chunk(mstate m, void* mem, size_t s);
+static void   do_check_tree(mstate m, tchunkptr t);
+static void   do_check_treebin(mstate m, bindex_t i);
+static void   do_check_smallbin(mstate m, bindex_t i);
+static void   do_check_malloc_state(mstate m);
+static int    bin_find(mstate m, mchunkptr x);
+static size_t traverse_and_check(mstate m);
+#endif /* DEBUG */
+
+/* ---------------------------- Indexing Bins ---------------------------- */
+
+#define is_small(s)         (((s) >> SMALLBIN_SHIFT) < NSMALLBINS)
+#define small_index(s)      ((s)  >> SMALLBIN_SHIFT)
+#define small_index2size(i) ((i)  << SMALLBIN_SHIFT)
+#define MIN_SMALL_INDEX     (small_index(MIN_CHUNK_SIZE))
+
+/* addressing by index. See above about smallbin repositioning */
+#define smallbin_at(M, i)   ((sbinptr)((char*)&((M)->smallbins[(i)<<1])))
+#define treebin_at(M,i)     (&((M)->treebins[i]))
+
+/* assign tree index for size S to variable I. Use x86 asm if possible  */
+#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
+#define compute_tree_index(S, I)\
+{\
+  unsigned int X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int K;\
+    __asm__("bsrl\t%1, %0\n\t" : "=r" (K) : "rm"  (X));\
+    I =  (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
+  }\
+}
+
+#elif defined (__INTEL_COMPILER)
+#define compute_tree_index(S, I)\
+{\
+  size_t X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int K = _bit_scan_reverse (X); \
+    I =  (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
+  }\
+}
+
+#elif defined(_MSC_VER) && _MSC_VER>=1300
+#define compute_tree_index(S, I)\
+{\
+  size_t X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int K;\
+    _BitScanReverse((DWORD *) &K, X);\
+    I =  (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
+  }\
+}
+
+#else /* GNUC */
+#define compute_tree_index(S, I)\
+{\
+  size_t X = S >> TREEBIN_SHIFT;\
+  if (X == 0)\
+    I = 0;\
+  else if (X > 0xFFFF)\
+    I = NTREEBINS-1;\
+  else {\
+    unsigned int Y = (unsigned int)X;\
+    unsigned int N = ((Y - 0x100) >> 16) & 8;\
+    unsigned int K = (((Y <<= N) - 0x1000) >> 16) & 4;\
+    N += K;\
+    N += K = (((Y <<= K) - 0x4000) >> 16) & 2;\
+    K = 14 - N + ((Y <<= K) >> 15);\
+    I = (K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1));\
+  }\
+}
+#endif /* GNUC */
+
+/* Bit representing maximum resolved size in a treebin at i */
+#define bit_for_tree_index(i) \
+   (i == NTREEBINS-1)? (SIZE_T_BITSIZE-1) : (((i) >> 1) + TREEBIN_SHIFT - 2)
+
+/* Shift placing maximum resolved bit in a treebin at i as sign bit */
+#define leftshift_for_tree_index(i) \
+   ((i == NTREEBINS-1)? 0 : \
+    ((SIZE_T_BITSIZE-SIZE_T_ONE) - (((i) >> 1) + TREEBIN_SHIFT - 2)))
+
+/* The size of the smallest chunk held in bin with index i */
+#define minsize_for_tree_index(i) \
+   ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) |  \
+   (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
+
+
+/* ------------------------ Operations on bin maps ----------------------- */
+
+/* bit corresponding to given index */
+#define idx2bit(i)              ((binmap_t)(1) << (i))
+
+/* Mark/Clear bits with given index */
+#define mark_smallmap(M,i)      ((M)->smallmap |=  idx2bit(i))
+#define clear_smallmap(M,i)     ((M)->smallmap &= ~idx2bit(i))
+#define smallmap_is_marked(M,i) ((M)->smallmap &   idx2bit(i))
+
+#define mark_treemap(M,i)       ((M)->treemap  |=  idx2bit(i))
+#define clear_treemap(M,i)      ((M)->treemap  &= ~idx2bit(i))
+#define treemap_is_marked(M,i)  ((M)->treemap  &   idx2bit(i))
+
+/* isolate the least set bit of a bitmap */
+#define least_bit(x)         ((x) & -(x))
+
+/* mask with all bits to left of least bit of x on */
+#define left_bits(x)         ((x<<1) | -(x<<1))
+
+/* mask with all bits to left of or equal to least bit of x on */
+#define same_or_left_bits(x) ((x) | -(x))
+
+/* index corresponding to given bit. Use x86 asm if possible */
+
+#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int J;\
+  __asm__("bsfl\t%1, %0\n\t" : "=r" (J) : "rm" (X));\
+  I = (bindex_t)J;\
+}
+
+#elif defined (__INTEL_COMPILER)
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int J;\
+  J = _bit_scan_forward (X); \
+  I = (bindex_t)J;\
+}
+
+#elif defined(_MSC_VER) && _MSC_VER>=1300
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int J;\
+  _BitScanForward((DWORD *) &J, X);\
+  I = (bindex_t)J;\
+}
+
+#elif USE_BUILTIN_FFS
+#define compute_bit2idx(X, I) I = ffs(X)-1
+
+#else
+#define compute_bit2idx(X, I)\
+{\
+  unsigned int Y = X - 1;\
+  unsigned int K = Y >> (16-4) & 16;\
+  unsigned int N = K;        Y >>= K;\
+  N += K = Y >> (8-3) &  8;  Y >>= K;\
+  N += K = Y >> (4-2) &  4;  Y >>= K;\
+  N += K = Y >> (2-1) &  2;  Y >>= K;\
+  N += K = Y >> (1-0) &  1;  Y >>= K;\
+  I = (bindex_t)(N + Y);\
+}
+#endif /* GNUC */
+
+
+/* ----------------------- Runtime Check Support ------------------------- */
+
+/*
+  For security, the main invariant is that malloc/free/etc never
+  writes to a static address other than malloc_state, unless static
+  malloc_state itself has been corrupted, which cannot occur via
+  malloc (because of these checks). In essence this means that we
+  believe all pointers, sizes, maps etc held in malloc_state, but
+  check all of those linked or offsetted from other embedded data
+  structures.  These checks are interspersed with main code in a way
+  that tends to minimize their run-time cost.
+
+  When FOOTERS is defined, in addition to range checking, we also
+  verify footer fields of inuse chunks, which can be used guarantee
+  that the mstate controlling malloc/free is intact.  This is a
+  streamlined version of the approach described by William Robertson
+  et al in "Run-time Detection of Heap-based Overflows" LISA'03
+  http://www.usenix.org/events/lisa03/tech/robertson.html The footer
+  of an inuse chunk holds the xor of its mstate and a random seed,
+  that is checked upon calls to free() and realloc().  This is
+  (probablistically) unguessable from outside the program, but can be
+  computed by any code successfully malloc'ing any chunk, so does not
+  itself provide protection against code that has already broken
+  security through some other means.  Unlike Robertson et al, we
+  always dynamically check addresses of all offset chunks (previous,
+  next, etc). This turns out to be cheaper than relying on hashes.
+*/
+
+#if !INSECURE
+/* Check if address a is at least as high as any from MORECORE or MMAP */
+#define ok_address(M, a) ((char*)(a) >= (M)->least_addr)
+/* Check if address of next chunk n is higher than base chunk p */
+#define ok_next(p, n)    ((char*)(p) < (char*)(n))
+/* Check if p has its cinuse bit on */
+#define ok_cinuse(p)     cinuse(p)
+/* Check if p has its pinuse bit on */
+#define ok_pinuse(p)     pinuse(p)
+
+#else /* !INSECURE */
+#define ok_address(M, a) (1)
+#define ok_next(b, n)    (1)
+#define ok_cinuse(p)     (1)
+#define ok_pinuse(p)     (1)
+#endif /* !INSECURE */
+
+#if (FOOTERS && !INSECURE)
+/* Check if (alleged) mstate m has expected magic field */
+#define ok_magic(M)      ((M)->magic == mparams.magic)
+#else  /* (FOOTERS && !INSECURE) */
+#define ok_magic(M)      (1)
+#endif /* (FOOTERS && !INSECURE) */
+
+
+/* In gcc, use __builtin_expect to minimize impact of checks */
+#if !INSECURE
+#if defined(__GNUC__) && __GNUC__ >= 3
+#define RTCHECK(e)  __builtin_expect(e, 1)
+#else /* GNUC */
+#define RTCHECK(e)  (e)
+#endif /* GNUC */
+#else /* !INSECURE */
+#define RTCHECK(e)  (1)
+#endif /* !INSECURE */
+
+/* macros to set up inuse chunks with or without footers */
+
+#if !FOOTERS
+
+#define mark_inuse_foot(M,p,s)
+
+/* Set cinuse bit and pinuse bit of next chunk */
+#define set_inuse(M,p,s)\
+  ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
+  ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
+
+/* Set cinuse and pinuse of this chunk and pinuse of next chunk */
+#define set_inuse_and_pinuse(M,p,s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
+  ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
+
+/* Set size, cinuse and pinuse bit of this chunk */
+#define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT))
+
+#else /* FOOTERS */
+
+/* Set foot of inuse chunk to be xor of mstate and seed */
+#define mark_inuse_foot(M,p,s)\
+  (((mchunkptr)((char*)(p) + (s)))->prev_foot = ((size_t)(M) ^ mparams.magic))
+
+#define get_mstate_for(p)\
+  ((mstate)(((mchunkptr)((char*)(p) +\
+    (chunksize(p))))->prev_foot ^ mparams.magic))
+
+#define set_inuse(M,p,s)\
+  ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
+  (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT), \
+  mark_inuse_foot(M,p,s))
+
+#define set_inuse_and_pinuse(M,p,s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
+  (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT),\
+ mark_inuse_foot(M,p,s))
+
+#define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
+  ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
+  mark_inuse_foot(M, p, s))
+
+#endif /* !FOOTERS */
+
+/* ---------------------------- setting mparams -------------------------- */
+
+/* Initialize mparams */
+static int init_mparams(void) {
+#ifdef NEED_GLOBAL_LOCK_INIT
+  if (malloc_global_mutex_status <= 0)
+    init_malloc_global_mutex();
+#endif
+
+  ACQUIRE_MALLOC_GLOBAL_LOCK();
+  if (mparams.magic == 0) {
+    size_t magic;
+    size_t psize;
+    size_t gsize;
+
+#ifndef WIN32
+    psize = malloc_getpagesize;
+    gsize = ((DEFAULT_GRANULARITY != 0)? DEFAULT_GRANULARITY : psize);
+#else /* WIN32 */
+    {
+      SYSTEM_INFO system_info;
+      GetSystemInfo(&system_info);
+      psize = system_info.dwPageSize;
+      gsize = ((DEFAULT_GRANULARITY != 0)?
+	       DEFAULT_GRANULARITY : system_info.dwAllocationGranularity);
+    }
+#endif /* WIN32 */
+
+    /* Sanity-check configuration:
+       size_t must be unsigned and as wide as pointer type.
+       ints must be at least 4 bytes.
+       alignment must be at least 8.
+       Alignment, min chunk size, and page size must all be powers of 2.
+    */
+    if ((sizeof(size_t) != sizeof(char*)) ||
+	(MAX_SIZE_T < MIN_CHUNK_SIZE)  ||
+	(sizeof(int) < 4)  ||
+	(MALLOC_ALIGNMENT < (size_t)8U) ||
+	((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-SIZE_T_ONE)) != 0) ||
+	((MCHUNK_SIZE      & (MCHUNK_SIZE-SIZE_T_ONE))      != 0) ||
+	((gsize            & (gsize-SIZE_T_ONE))            != 0) ||
+	((psize            & (psize-SIZE_T_ONE))            != 0))
+      ABORT;
+
+    mparams.granularity = gsize;
+    mparams.page_size = psize;
+    mparams.mmap_threshold = DEFAULT_MMAP_THRESHOLD;
+    mparams.trim_threshold = DEFAULT_TRIM_THRESHOLD;
+#if MORECORE_CONTIGUOUS
+    mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT;
+#else  /* MORECORE_CONTIGUOUS */
+    mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT|USE_NONCONTIGUOUS_BIT;
+#endif /* MORECORE_CONTIGUOUS */
+
+#if !ONLY_MSPACES
+    /* Set up lock for main malloc area */
+    gm->mflags = mparams.default_mflags;
+    INITIAL_LOCK(&gm->mutex);
+#endif
+
+#if (FOOTERS && !INSECURE)
+    {
+#if USE_DEV_RANDOM
+      int fd;
+      unsigned char buf[sizeof(size_t)];
+      /* Try to use /dev/urandom, else fall back on using time */
+      if ((fd = open("/dev/urandom", O_RDONLY)) >= 0 &&
+	  read(fd, buf, sizeof(buf)) == sizeof(buf)) {
+	magic = *((size_t *) buf);
+	close(fd);
+      }
+      else
+#endif /* USE_DEV_RANDOM */
+#ifdef WIN32
+	magic = (size_t)(GetTickCount() ^ (size_t)0x55555555U);
+#else
+      magic = (size_t)(time(0) ^ (size_t)0x55555555U);
+#endif
+      magic |= (size_t)8U;    /* ensure nonzero */
+      magic &= ~(size_t)7U;   /* improve chances of fault for bad values */
+    }
+#else /* (FOOTERS && !INSECURE) */
+    magic = (size_t)0x58585858U;
+#endif /* (FOOTERS && !INSECURE) */
+
+    mparams.magic = magic;
+  }
+
+  RELEASE_MALLOC_GLOBAL_LOCK();
+  return 1;
+}
+
+/* support for mallopt */
+static int change_mparam(int param_number, int value) {
+  size_t val = (value == -1)? MAX_SIZE_T : (size_t)value;
+  ensure_initialization();
+  switch(param_number) {
+  case M_TRIM_THRESHOLD:
+    mparams.trim_threshold = val;
+    return 1;
+  case M_GRANULARITY:
+    if (val >= mparams.page_size && ((val & (val-1)) == 0)) {
+      mparams.granularity = val;
+      return 1;
+    }
+    else
+      return 0;
+  case M_MMAP_THRESHOLD:
+    mparams.mmap_threshold = val;
+    return 1;
+  default:
+    return 0;
+  }
+}
+
+#if DEBUG
+/* ------------------------- Debugging Support --------------------------- */
+
+/* Check properties of any chunk, whether free, inuse, mmapped etc  */
+static void do_check_any_chunk(mstate m, mchunkptr p) {
+  assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
+  assert(ok_address(m, p));
+}
+
+/* Check properties of top chunk */
+static void do_check_top_chunk(mstate m, mchunkptr p) {
+  msegmentptr sp = segment_holding(m, (char*)p);
+  size_t  sz = p->head & ~INUSE_BITS; /* third-lowest bit can be set! */
+  assert(sp != 0);
+  assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
+  assert(ok_address(m, p));
+  assert(sz == m->topsize);
+  assert(sz > 0);
+  assert(sz == ((sp->base + sp->size) - (char*)p) - TOP_FOOT_SIZE);
+  assert(pinuse(p));
+  assert(!pinuse(chunk_plus_offset(p, sz)));
+}
+
+/* Check properties of (inuse) mmapped chunks */
+static void do_check_mmapped_chunk(mstate m, mchunkptr p) {
+  size_t  sz = chunksize(p);
+  size_t len = (sz + (p->prev_foot & ~IS_MMAPPED_BIT) + MMAP_FOOT_PAD);
+  assert(is_mmapped(p));
+  assert(use_mmap(m));
+  assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
+  assert(ok_address(m, p));
+  assert(!is_small(sz));
+  assert((len & (mparams.page_size-SIZE_T_ONE)) == 0);
+  assert(chunk_plus_offset(p, sz)->head == FENCEPOST_HEAD);
+  assert(chunk_plus_offset(p, sz+SIZE_T_SIZE)->head == 0);
+}
+
+/* Check properties of inuse chunks */
+static void do_check_inuse_chunk(mstate m, mchunkptr p) {
+  do_check_any_chunk(m, p);
+  assert(cinuse(p));
+  assert(next_pinuse(p));
+  /* If not pinuse and not mmapped, previous chunk has OK offset */
+  assert(is_mmapped(p) || pinuse(p) || next_chunk(prev_chunk(p)) == p);
+  if (is_mmapped(p))
+    do_check_mmapped_chunk(m, p);
+}
+
+/* Check properties of free chunks */
+static void do_check_free_chunk(mstate m, mchunkptr p) {
+  size_t sz = chunksize(p);
+  mchunkptr next = chunk_plus_offset(p, sz);
+  do_check_any_chunk(m, p);
+  assert(!cinuse(p));
+  assert(!next_pinuse(p));
+  assert (!is_mmapped(p));
+  if (p != m->dv && p != m->top) {
+    if (sz >= MIN_CHUNK_SIZE) {
+      assert((sz & CHUNK_ALIGN_MASK) == 0);
+      assert(is_aligned(chunk2mem(p)));
+      assert(next->prev_foot == sz);
+      assert(pinuse(p));
+      assert (next == m->top || cinuse(next));
+      assert(p->fd->bk == p);
+      assert(p->bk->fd == p);
+    }
+    else  /* markers are always of size SIZE_T_SIZE */
+      assert(sz == SIZE_T_SIZE);
+  }
+}
+
+/* Check properties of malloced chunks at the point they are malloced */
+static void do_check_malloced_chunk(mstate m, void* mem, size_t s) {
+  if (mem != 0) {
+    mchunkptr p = mem2chunk(mem);
+    size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
+    do_check_inuse_chunk(m, p);
+    assert((sz & CHUNK_ALIGN_MASK) == 0);
+    assert(sz >= MIN_CHUNK_SIZE);
+    assert(sz >= s);
+    /* unless mmapped, size is less than MIN_CHUNK_SIZE more than request */
+    assert(is_mmapped(p) || sz < (s + MIN_CHUNK_SIZE));
+  }
+}
+
+/* Check a tree and its subtrees.  */
+static void do_check_tree(mstate m, tchunkptr t) {
+  tchunkptr head = 0;
+  tchunkptr u = t;
+  bindex_t tindex = t->index;
+  size_t tsize = chunksize(t);
+  bindex_t idx;
+  compute_tree_index(tsize, idx);
+  assert(tindex == idx);
+  assert(tsize >= MIN_LARGE_SIZE);
+  assert(tsize >= minsize_for_tree_index(idx));
+  assert((idx == NTREEBINS-1) || (tsize < minsize_for_tree_index((idx+1))));
+
+  do { /* traverse through chain of same-sized nodes */
+    do_check_any_chunk(m, ((mchunkptr)u));
+    assert(u->index == tindex);
+    assert(chunksize(u) == tsize);
+    assert(!cinuse(u));
+    assert(!next_pinuse(u));
+    assert(u->fd->bk == u);
+    assert(u->bk->fd == u);
+    if (u->parent == 0) {
+      assert(u->child[0] == 0);
+      assert(u->child[1] == 0);
+    }
+    else {
+      assert(head == 0); /* only one node on chain has parent */
+      head = u;
+      assert(u->parent != u);
+      assert (u->parent->child[0] == u ||
+	      u->parent->child[1] == u ||
+	      *((tbinptr*)(u->parent)) == u);
+      if (u->child[0] != 0) {
+	assert(u->child[0]->parent == u);
+	assert(u->child[0] != u);
+	do_check_tree(m, u->child[0]);
+      }
+      if (u->child[1] != 0) {
+	assert(u->child[1]->parent == u);
+	assert(u->child[1] != u);
+	do_check_tree(m, u->child[1]);
+      }
+      if (u->child[0] != 0 && u->child[1] != 0) {
+	assert(chunksize(u->child[0]) < chunksize(u->child[1]));
+      }
+    }
+    u = u->fd;
+  } while (u != t);
+  assert(head != 0);
+}
+
+/*  Check all the chunks in a treebin.  */
+static void do_check_treebin(mstate m, bindex_t i) {
+  tbinptr* tb = treebin_at(m, i);
+  tchunkptr t = *tb;
+  int empty = (m->treemap & (1U << i)) == 0;
+  if (t == 0)
+    assert(empty);
+  if (!empty)
+    do_check_tree(m, t);
+}
+
+/*  Check all the chunks in a smallbin.  */
+static void do_check_smallbin(mstate m, bindex_t i) {
+  sbinptr b = smallbin_at(m, i);
+  mchunkptr p = b->bk;
+  unsigned int empty = (m->smallmap & (1U << i)) == 0;
+  if (p == b)
+    assert(empty);
+  if (!empty) {
+    for (; p != b; p = p->bk) {
+      size_t size = chunksize(p);
+      mchunkptr q;
+      /* each chunk claims to be free */
+      do_check_free_chunk(m, p);
+      /* chunk belongs in bin */
+      assert(small_index(size) == i);
+      assert(p->bk == b || chunksize(p->bk) == chunksize(p));
+      /* chunk is followed by an inuse chunk */
+      q = next_chunk(p);
+      if (q->head != FENCEPOST_HEAD)
+	do_check_inuse_chunk(m, q);
+    }
+  }
+}
+
+/* Find x in a bin. Used in other check functions. */
+static int bin_find(mstate m, mchunkptr x) {
+  size_t size = chunksize(x);
+  if (is_small(size)) {
+    bindex_t sidx = small_index(size);
+    sbinptr b = smallbin_at(m, sidx);
+    if (smallmap_is_marked(m, sidx)) {
+      mchunkptr p = b;
+      do {
+	if (p == x)
+	  return 1;
+      } while ((p = p->fd) != b);
+    }
+  }
+  else {
+    bindex_t tidx;
+    compute_tree_index(size, tidx);
+    if (treemap_is_marked(m, tidx)) {
+      tchunkptr t = *treebin_at(m, tidx);
+      size_t sizebits = size << leftshift_for_tree_index(tidx);
+      while (t != 0 && chunksize(t) != size) {
+	t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
+	sizebits <<= 1;
+      }
+      if (t != 0) {
+	tchunkptr u = t;
+	do {
+	  if (u == (tchunkptr)x)
+	    return 1;
+	} while ((u = u->fd) != t);
+      }
+    }
+  }
+  return 0;
+}
+
+/* Traverse each chunk and check it; return total */
+static size_t traverse_and_check(mstate m) {
+  size_t sum = 0;
+  if (is_initialized(m)) {
+    msegmentptr s = &m->seg;
+    sum += m->topsize + TOP_FOOT_SIZE;
+    while (s != 0) {
+      mchunkptr q = align_as_chunk(s->base);
+      mchunkptr lastq = 0;
+      assert(pinuse(q));
+      while (segment_holds(s, q) &&
+	     q != m->top && q->head != FENCEPOST_HEAD) {
+	sum += chunksize(q);
+	if (cinuse(q)) {
+	  assert(!bin_find(m, q));
+	  do_check_inuse_chunk(m, q);
+	}
+	else {
+	  assert(q == m->dv || bin_find(m, q));
+	  assert(lastq == 0 || cinuse(lastq)); /* Not 2 consecutive free */
+	  do_check_free_chunk(m, q);
+	}
+	lastq = q;
+	q = next_chunk(q);
+      }
+      s = s->next;
+    }
+  }
+  return sum;
+}
+
+/* Check all properties of malloc_state. */
+static void do_check_malloc_state(mstate m) {
+  bindex_t i;
+  size_t total;
+  /* check bins */
+  for (i = 0; i < NSMALLBINS; ++i)
+    do_check_smallbin(m, i);
+  for (i = 0; i < NTREEBINS; ++i)
+    do_check_treebin(m, i);
+
+  if (m->dvsize != 0) { /* check dv chunk */
+    do_check_any_chunk(m, m->dv);
+    assert(m->dvsize == chunksize(m->dv));
+    assert(m->dvsize >= MIN_CHUNK_SIZE);
+    assert(bin_find(m, m->dv) == 0);
+  }
+
+  if (m->top != 0) {   /* check top chunk */
+    do_check_top_chunk(m, m->top);
+    /*assert(m->topsize == chunksize(m->top)); redundant */
+    assert(m->topsize > 0);
+    assert(bin_find(m, m->top) == 0);
+  }
+
+  total = traverse_and_check(m);
+  assert(total <= m->footprint);
+  assert(m->footprint <= m->max_footprint);
+}
+#endif /* DEBUG */
+
+/* ----------------------------- statistics ------------------------------ */
+
+#if !NO_MALLINFO
+static struct mallinfo internal_mallinfo(mstate m) {
+  struct mallinfo nm = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+  ensure_initialization();
+  if (!PREACTION(m)) {
+    check_malloc_state(m);
+    if (is_initialized(m)) {
+      size_t nfree = SIZE_T_ONE; /* top always free */
+      size_t mfree = m->topsize + TOP_FOOT_SIZE;
+      size_t sum = mfree;
+      msegmentptr s = &m->seg;
+      while (s != 0) {
+	mchunkptr q = align_as_chunk(s->base);
+	while (segment_holds(s, q) &&
+	       q != m->top && q->head != FENCEPOST_HEAD) {
+	  size_t sz = chunksize(q);
+	  sum += sz;
+	  if (!cinuse(q)) {
+	    mfree += sz;
+	    ++nfree;
+	  }
+	  q = next_chunk(q);
+	}
+	s = s->next;
+      }
+
+      nm.arena    = sum;
+      nm.ordblks  = nfree;
+      nm.hblkhd   = m->footprint - sum;
+      nm.usmblks  = m->max_footprint;
+      nm.uordblks = m->footprint - mfree;
+      nm.fordblks = mfree;
+      nm.keepcost = m->topsize;
+    }
+
+    POSTACTION(m);
+  }
+  return nm;
+}
+#endif /* !NO_MALLINFO */
+
+static void internal_malloc_stats(mstate m) {
+  ensure_initialization();
+  if (!PREACTION(m)) {
+    size_t maxfp = 0;
+    size_t fp = 0;
+    size_t used = 0;
+    check_malloc_state(m);
+    if (is_initialized(m)) {
+      msegmentptr s = &m->seg;
+      maxfp = m->max_footprint;
+      fp = m->footprint;
+      used = fp - (m->topsize + TOP_FOOT_SIZE);
+
+      while (s != 0) {
+	mchunkptr q = align_as_chunk(s->base);
+	while (segment_holds(s, q) &&
+	       q != m->top && q->head != FENCEPOST_HEAD) {
+	  if (!cinuse(q))
+	    used -= chunksize(q);
+	  q = next_chunk(q);
+	}
+	s = s->next;
+      }
+    }
+
+    fprintf(stderr, "max system bytes = %10lu\n", (unsigned long)(maxfp));
+    fprintf(stderr, "system bytes     = %10lu\n", (unsigned long)(fp));
+    fprintf(stderr, "in use bytes     = %10lu\n", (unsigned long)(used));
+
+    POSTACTION(m);
+  }
+}
+
+/* ----------------------- Operations on smallbins ----------------------- */
+
+/*
+  Various forms of linking and unlinking are defined as macros.  Even
+  the ones for trees, which are very long but have very short typical
+  paths.  This is ugly but reduces reliance on inlining support of
+  compilers.
+*/
+
+/* Link a free chunk into a smallbin  */
+#define insert_small_chunk(M, P, S) {\
+  bindex_t I  = small_index(S);\
+  mchunkptr B = smallbin_at(M, I);\
+  mchunkptr F = B;\
+  assert(S >= MIN_CHUNK_SIZE);\
+  if (!smallmap_is_marked(M, I))\
+    mark_smallmap(M, I);\
+  else if (RTCHECK(ok_address(M, B->fd)))\
+    F = B->fd;\
+  else {\
+    CORRUPTION_ERROR_ACTION(M);\
+  }\
+  B->fd = P;\
+  F->bk = P;\
+  P->fd = F;\
+  P->bk = B;\
+}
+
+/* Unlink a chunk from a smallbin  */
+#define unlink_small_chunk(M, P, S) {\
+  mchunkptr F = P->fd;\
+  mchunkptr B = P->bk;\
+  bindex_t I = small_index(S);\
+  assert(P != B);\
+  assert(P != F);\
+  assert(chunksize(P) == small_index2size(I));\
+  if (F == B)\
+    clear_smallmap(M, I);\
+  else if (RTCHECK((F == smallbin_at(M,I) || ok_address(M, F)) &&\
+		   (B == smallbin_at(M,I) || ok_address(M, B)))) {\
+    F->bk = B;\
+    B->fd = F;\
+  }\
+  else {\
+    CORRUPTION_ERROR_ACTION(M);\
+  }\
+}
+
+/* Unlink the first chunk from a smallbin */
+#define unlink_first_small_chunk(M, B, P, I) {\
+  mchunkptr F = P->fd;\
+  assert(P != B);\
+  assert(P != F);\
+  assert(chunksize(P) == small_index2size(I));\
+  if (B == F)\
+    clear_smallmap(M, I);\
+  else if (RTCHECK(ok_address(M, F))) {\
+    B->fd = F;\
+    F->bk = B;\
+  }\
+  else {\
+    CORRUPTION_ERROR_ACTION(M);\
+  }\
+}
+
+
+
+/* Replace dv node, binning the old one */
+/* Used only when dvsize known to be small */
+#define replace_dv(M, P, S) {\
+  size_t DVS = M->dvsize;\
+  if (DVS != 0) {\
+    mchunkptr DV = M->dv;\
+    assert(is_small(DVS));\
+    insert_small_chunk(M, DV, DVS);\
+  }\
+  M->dvsize = S;\
+  M->dv = P;\
+}
+
+/* ------------------------- Operations on trees ------------------------- */
+
+/* Insert chunk into tree */
+#define insert_large_chunk(M, X, S) {\
+  tbinptr* H;\
+  bindex_t I;\
+  compute_tree_index(S, I);\
+  H = treebin_at(M, I);\
+  X->index = I;\
+  X->child[0] = X->child[1] = 0;\
+  if (!treemap_is_marked(M, I)) {\
+    mark_treemap(M, I);\
+    *H = X;\
+    X->parent = (tchunkptr)H;\
+    X->fd = X->bk = X;\
+  }\
+  else {\
+    tchunkptr T = *H;\
+    size_t K = S << leftshift_for_tree_index(I);\
+    for (;;) {\
+      if (chunksize(T) != S) {\
+	tchunkptr* C = &(T->child[(K >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1]);\
+	K <<= 1;\
+	if (*C != 0)\
+	  T = *C;\
+	else if (RTCHECK(ok_address(M, C))) {\
+	  *C = X;\
+	  X->parent = T;\
+	  X->fd = X->bk = X;\
+	  break;\
+	}\
+	else {\
+	  CORRUPTION_ERROR_ACTION(M);\
+	  break;\
+	}\
+      }\
+      else {\
+	tchunkptr F = T->fd;\
+	if (RTCHECK(ok_address(M, T) && ok_address(M, F))) {\
+	  T->fd = F->bk = X;\
+	  X->fd = F;\
+	  X->bk = T;\
+	  X->parent = 0;\
+	  break;\
+	}\
+	else {\
+	  CORRUPTION_ERROR_ACTION(M);\
+	  break;\
+	}\
+      }\
+    }\
+  }\
+}
+
+/*
+  Unlink steps:
+
+  1. If x is a chained node, unlink it from its same-sized fd/bk links
+     and choose its bk node as its replacement.
+  2. If x was the last node of its size, but not a leaf node, it must
+     be replaced with a leaf node (not merely one with an open left or
+     right), to make sure that lefts and rights of descendants
+     correspond properly to bit masks.  We use the rightmost descendant
+     of x.  We could use any other leaf, but this is easy to locate and
+     tends to counteract removal of leftmosts elsewhere, and so keeps
+     paths shorter than minimally guaranteed.  This doesn't loop much
+     because on average a node in a tree is near the bottom.
+  3. If x is the base of a chain (i.e., has parent links) relink
+     x's parent and children to x's replacement (or null if none).
+*/
+
+#define unlink_large_chunk(M, X) {\
+  tchunkptr XP = X->parent;\
+  tchunkptr R;\
+  if (X->bk != X) {\
+    tchunkptr F = X->fd;\
+    R = X->bk;\
+    if (RTCHECK(ok_address(M, F))) {\
+      F->bk = R;\
+      R->fd = F;\
+    }\
+    else {\
+      CORRUPTION_ERROR_ACTION(M);\
+    }\
+  }\
+  else {\
+    tchunkptr* RP;\
+    if (((R = *(RP = &(X->child[1]))) != 0) ||\
+	((R = *(RP = &(X->child[0]))) != 0)) {\
+      tchunkptr* CP;\
+      while ((*(CP = &(R->child[1])) != 0) ||\
+	     (*(CP = &(R->child[0])) != 0)) {\
+	R = *(RP = CP);\
+      }\
+      if (RTCHECK(ok_address(M, RP)))\
+	*RP = 0;\
+      else {\
+	CORRUPTION_ERROR_ACTION(M);\
+      }\
+    }\
+  }\
+  if (XP != 0) {\
+    tbinptr* H = treebin_at(M, X->index);\
+    if (X == *H) {\
+      if ((*H = R) == 0) \
+	clear_treemap(M, X->index);\
+    }\
+    else if (RTCHECK(ok_address(M, XP))) {\
+      if (XP->child[0] == X) \
+	XP->child[0] = R;\
+      else \
+	XP->child[1] = R;\
+    }\
+    else\
+      CORRUPTION_ERROR_ACTION(M);\
+    if (R != 0) {\
+      if (RTCHECK(ok_address(M, R))) {\
+	tchunkptr C0, C1;\
+	R->parent = XP;\
+	if ((C0 = X->child[0]) != 0) {\
+	  if (RTCHECK(ok_address(M, C0))) {\
+	    R->child[0] = C0;\
+	    C0->parent = R;\
+	  }\
+	  else\
+	    CORRUPTION_ERROR_ACTION(M);\
+	}\
+	if ((C1 = X->child[1]) != 0) {\
+	  if (RTCHECK(ok_address(M, C1))) {\
+	    R->child[1] = C1;\
+	    C1->parent = R;\
+	  }\
+	  else\
+	    CORRUPTION_ERROR_ACTION(M);\
+	}\
+      }\
+      else\
+	CORRUPTION_ERROR_ACTION(M);\
+    }\
+  }\
+}
+
+/* Relays to large vs small bin operations */
+
+#define insert_chunk(M, P, S)\
+  if (is_small(S)) insert_small_chunk(M, P, S)\
+  else { tchunkptr TP = (tchunkptr)(P); insert_large_chunk(M, TP, S); }
+
+#define unlink_chunk(M, P, S)\
+  if (is_small(S)) unlink_small_chunk(M, P, S)\
+  else { tchunkptr TP = (tchunkptr)(P); unlink_large_chunk(M, TP); }
+
+
+/* Relays to internal calls to malloc/free from realloc, memalign etc */
+
+#if ONLY_MSPACES
+#define internal_malloc(m, b) mspace_malloc(m, b)
+#define internal_free(m, mem) mspace_free(m,mem);
+#else /* ONLY_MSPACES */
+#if MSPACES
+#define internal_malloc(m, b)\
+   (m == gm)? dlmalloc(b) : mspace_malloc(m, b)
+#define internal_free(m, mem)\
+   if (m == gm) dlfree(mem); else mspace_free(m,mem);
+#else /* MSPACES */
+#define internal_malloc(m, b) dlmalloc(b)
+#define internal_free(m, mem) dlfree(mem)
+#endif /* MSPACES */
+#endif /* ONLY_MSPACES */
+
+/* -----------------------  Direct-mmapping chunks ----------------------- */
+
+/*
+  Directly mmapped chunks are set up with an offset to the start of
+  the mmapped region stored in the prev_foot field of the chunk. This
+  allows reconstruction of the required argument to MUNMAP when freed,
+  and also allows adjustment of the returned chunk to meet alignment
+  requirements (especially in memalign).  There is also enough space
+  allocated to hold a fake next chunk of size SIZE_T_SIZE to maintain
+  the PINUSE bit so frees can be checked.
+*/
+
+/* Malloc using mmap */
+static void* mmap_alloc(mstate m, size_t nb) {
+  size_t mmsize = mmap_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
+  if (mmsize > nb) {     /* Check for wrap around 0 */
+    char* mm = (char*)(CALL_DIRECT_MMAP(mmsize));
+    if (mm != CMFAIL) {
+      size_t offset = align_offset(chunk2mem(mm));
+      size_t psize = mmsize - offset - MMAP_FOOT_PAD;
+      mchunkptr p = (mchunkptr)(mm + offset);
+      p->prev_foot = offset | IS_MMAPPED_BIT;
+      (p)->head = (psize|CINUSE_BIT);
+      mark_inuse_foot(m, p, psize);
+      chunk_plus_offset(p, psize)->head = FENCEPOST_HEAD;
+      chunk_plus_offset(p, psize+SIZE_T_SIZE)->head = 0;
+
+      if (mm < m->least_addr)
+	m->least_addr = mm;
+      if ((m->footprint += mmsize) > m->max_footprint)
+	m->max_footprint = m->footprint;
+      assert(is_aligned(chunk2mem(p)));
+      check_mmapped_chunk(m, p);
+      return chunk2mem(p);
+    }
+  }
+  return 0;
+}
+
+/* Realloc using mmap */
+static mchunkptr mmap_resize(mstate m, mchunkptr oldp, size_t nb) {
+  size_t oldsize = chunksize(oldp);
+  if (is_small(nb)) /* Can't shrink mmap regions below small size */
+    return 0;
+  /* Keep old chunk if big enough but not too big */
+  if (oldsize >= nb + SIZE_T_SIZE &&
+      (oldsize - nb) <= (mparams.granularity << 1))
+    return oldp;
+  else {
+    size_t offset = oldp->prev_foot & ~IS_MMAPPED_BIT;
+    size_t oldmmsize = oldsize + offset + MMAP_FOOT_PAD;
+    size_t newmmsize = mmap_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
+    char* cp = (char*)CALL_MREMAP((char*)oldp - offset,
+				  oldmmsize, newmmsize, 1);
+    if (cp != CMFAIL) {
+      mchunkptr newp = (mchunkptr)(cp + offset);
+      size_t psize = newmmsize - offset - MMAP_FOOT_PAD;
+      newp->head = (psize|CINUSE_BIT);
+      mark_inuse_foot(m, newp, psize);
+      chunk_plus_offset(newp, psize)->head = FENCEPOST_HEAD;
+      chunk_plus_offset(newp, psize+SIZE_T_SIZE)->head = 0;
+
+      if (cp < m->least_addr)
+	m->least_addr = cp;
+      if ((m->footprint += newmmsize - oldmmsize) > m->max_footprint)
+	m->max_footprint = m->footprint;
+      check_mmapped_chunk(m, newp);
+      return newp;
+    }
+  }
+  return 0;
+}
+
+/* -------------------------- mspace management -------------------------- */
+
+/* Initialize top chunk and its size */
+static void init_top(mstate m, mchunkptr p, size_t psize) {
+  /* Ensure alignment */
+  size_t offset = align_offset(chunk2mem(p));
+  p = (mchunkptr)((char*)p + offset);
+  psize -= offset;
+
+  m->top = p;
+  m->topsize = psize;
+  p->head = psize | PINUSE_BIT;
+  /* set size of fake trailing chunk holding overhead space only once */
+  chunk_plus_offset(p, psize)->head = TOP_FOOT_SIZE;
+  m->trim_check = mparams.trim_threshold; /* reset on each update */
+}
+
+/* Initialize bins for a new mstate that is otherwise zeroed out */
+static void init_bins(mstate m) {
+  /* Establish circular links for smallbins */
+  bindex_t i;
+  for (i = 0; i < NSMALLBINS; ++i) {
+    sbinptr bin = smallbin_at(m,i);
+    bin->fd = bin->bk = bin;
+  }
+}
+
+#if PROCEED_ON_ERROR
+
+/* default corruption action */
+static void reset_on_error(mstate m) {
+  int i;
+  ++malloc_corruption_error_count;
+  /* Reinitialize fields to forget about all memory */
+  m->smallbins = m->treebins = 0;
+  m->dvsize = m->topsize = 0;
+  m->seg.base = 0;
+  m->seg.size = 0;
+  m->seg.next = 0;
+  m->top = m->dv = 0;
+  for (i = 0; i < NTREEBINS; ++i)
+    *treebin_at(m, i) = 0;
+  init_bins(m);
+}
+#endif /* PROCEED_ON_ERROR */
+
+/* Allocate chunk and prepend remainder with chunk in successor base. */
+static void* prepend_alloc(mstate m, char* newbase, char* oldbase,
+			   size_t nb) {
+  mchunkptr p = align_as_chunk(newbase);
+  mchunkptr oldfirst = align_as_chunk(oldbase);
+  size_t psize = (char*)oldfirst - (char*)p;
+  mchunkptr q = chunk_plus_offset(p, nb);
+  size_t qsize = psize - nb;
+  set_size_and_pinuse_of_inuse_chunk(m, p, nb);
+
+  assert((char*)oldfirst > (char*)q);
+  assert(pinuse(oldfirst));
+  assert(qsize >= MIN_CHUNK_SIZE);
+
+  /* consolidate remainder with first chunk of old base */
+  if (oldfirst == m->top) {
+    size_t tsize = m->topsize += qsize;
+    m->top = q;
+    q->head = tsize | PINUSE_BIT;
+    check_top_chunk(m, q);
+  }
+  else if (oldfirst == m->dv) {
+    size_t dsize = m->dvsize += qsize;
+    m->dv = q;
+    set_size_and_pinuse_of_free_chunk(q, dsize);
+  }
+  else {
+    if (!cinuse(oldfirst)) {
+      size_t nsize = chunksize(oldfirst);
+      unlink_chunk(m, oldfirst, nsize);
+      oldfirst = chunk_plus_offset(oldfirst, nsize);
+      qsize += nsize;
+    }
+    set_free_with_pinuse(q, qsize, oldfirst);
+    insert_chunk(m, q, qsize);
+    check_free_chunk(m, q);
+  }
+
+  check_malloced_chunk(m, chunk2mem(p), nb);
+  return chunk2mem(p);
+}
+
+/* Add a segment to hold a new noncontiguous region */
+static void add_segment(mstate m, char* tbase, size_t tsize, flag_t mmapped) {
+  /* Determine locations and sizes of segment, fenceposts, old top */
+  char* old_top = (char*)m->top;
+  msegmentptr oldsp = segment_holding(m, old_top);
+  char* old_end = oldsp->base + oldsp->size;
+  size_t ssize = pad_request(sizeof(struct malloc_segment));
+  char* rawsp = old_end - (ssize + FOUR_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
+  size_t offset = align_offset(chunk2mem(rawsp));
+  char* asp = rawsp + offset;
+  char* csp = (asp < (old_top + MIN_CHUNK_SIZE))? old_top : asp;
+  mchunkptr sp = (mchunkptr)csp;
+  msegmentptr ss = (msegmentptr)(chunk2mem(sp));
+  mchunkptr tnext = chunk_plus_offset(sp, ssize);
+  mchunkptr p = tnext;
+  int nfences = 0;
+
+  /* reset top to new space */
+  init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
+
+  /* Set up segment record */
+  assert(is_aligned(ss));
+  set_size_and_pinuse_of_inuse_chunk(m, sp, ssize);
+  *ss = m->seg; /* Push current record */
+  m->seg.base = tbase;
+  m->seg.size = tsize;
+  m->seg.sflags = mmapped;
+  m->seg.next = ss;
+
+  /* Insert trailing fenceposts */
+  for (;;) {
+    mchunkptr nextp = chunk_plus_offset(p, SIZE_T_SIZE);
+    p->head = FENCEPOST_HEAD;
+    ++nfences;
+    if ((char*)(&(nextp->head)) < old_end)
+      p = nextp;
+    else
+      break;
+  }
+  assert(nfences >= 2);
+
+  /* Insert the rest of old top into a bin as an ordinary free chunk */
+  if (csp != old_top) {
+    mchunkptr q = (mchunkptr)old_top;
+    size_t psize = csp - old_top;
+    mchunkptr tn = chunk_plus_offset(q, psize);
+    set_free_with_pinuse(q, psize, tn);
+    insert_chunk(m, q, psize);
+  }
+
+  check_top_chunk(m, m->top);
+}
+
+/* -------------------------- System allocation -------------------------- */
+
+/* Get memory from system using MORECORE or MMAP */
+static void* sys_alloc(mstate m, size_t nb) {
+  char* tbase = CMFAIL;
+  size_t tsize = 0;
+  flag_t mmap_flag = 0;
+
+  ensure_initialization();
+
+  /* Directly map large chunks */
+  if (use_mmap(m) && nb >= mparams.mmap_threshold) {
+    void* mem = mmap_alloc(m, nb);
+    if (mem != 0)
+      return mem;
+  }
+
+  /*
+    Try getting memory in any of three ways (in most-preferred to
+    least-preferred order):
+    1. A call to MORECORE that can normally contiguously extend memory.
+       (disabled if not MORECORE_CONTIGUOUS or not HAVE_MORECORE or
+       main space is mmapped or a previous contiguous call failed)
+    2. A call to MMAP new space (disabled if not HAVE_MMAP).
+       Note that under the default settings, if MORECORE is unable to
+       fulfill a request, and HAVE_MMAP is true, then mmap is
+       used as a noncontiguous system allocator. This is a useful backup
+       strategy for systems with holes in address spaces -- in this case
+       sbrk cannot contiguously expand the heap, but mmap may be able to
+       find space.
+    3. A call to MORECORE that cannot usually contiguously extend memory.
+       (disabled if not HAVE_MORECORE)
+
+   In all cases, we need to request enough bytes from system to ensure
+   we can malloc nb bytes upon success, so pad with enough space for
+   top_foot, plus alignment-pad to make sure we don't lose bytes if
+   not on boundary, and round this up to a granularity unit.
+  */
+
+  if (MORECORE_CONTIGUOUS && !use_noncontiguous(m)) {
+    char* br = CMFAIL;
+    msegmentptr ss = (m->top == 0)? 0 : segment_holding(m, (char*)m->top);
+    size_t asize = 0;
+    ACQUIRE_MALLOC_GLOBAL_LOCK();
+
+    if (ss == 0) {  /* First time through or recovery */
+      char* base = (char*)CALL_MORECORE(0);
+      if (base != CMFAIL) {
+	asize = granularity_align(nb + SYS_ALLOC_PADDING);
+	/* Adjust to end on a page boundary */
+	if (!is_page_aligned(base))
+	  asize += (page_align((size_t)base) - (size_t)base);
+	/* Can't call MORECORE if size is negative when treated as signed */
+	if (asize < HALF_MAX_SIZE_T &&
+	    (br = (char*)(CALL_MORECORE(asize))) == base) {
+	  tbase = base;
+	  tsize = asize;
+	}
+      }
+    }
+    else {
+      /* Subtract out existing available top space from MORECORE request. */
+      asize = granularity_align(nb - m->topsize + SYS_ALLOC_PADDING);
+      /* Use mem here only if it did continuously extend old space */
+      if (asize < HALF_MAX_SIZE_T &&
+	  (br = (char*)(CALL_MORECORE(asize))) == ss->base+ss->size) {
+	tbase = br;
+	tsize = asize;
+      }
+    }
+
+    if (tbase == CMFAIL) {    /* Cope with partial failure */
+      if (br != CMFAIL) {    /* Try to use/extend the space we did get */
+	if (asize < HALF_MAX_SIZE_T &&
+	    asize < nb + SYS_ALLOC_PADDING) {
+	  size_t esize = granularity_align(nb + SYS_ALLOC_PADDING - asize);
+	  if (esize < HALF_MAX_SIZE_T) {
+	    char* end = (char*)CALL_MORECORE(esize);
+	    if (end != CMFAIL)
+	      asize += esize;
+	    else {            /* Can't use; try to release */
+	      (void) CALL_MORECORE(-asize);
+	      br = CMFAIL;
+	    }
+	  }
+	}
+      }
+      if (br != CMFAIL) {    /* Use the space we did get */
+	tbase = br;
+	tsize = asize;
+      }
+      else
+	disable_contiguous(m); /* Don't try contiguous path in the future */
+    }
+
+    RELEASE_MALLOC_GLOBAL_LOCK();
+  }
+
+  if (HAVE_MMAP && tbase == CMFAIL) {  /* Try MMAP */
+    size_t rsize = granularity_align(nb + SYS_ALLOC_PADDING);
+    if (rsize > nb) { /* Fail if wraps around zero */
+      char* mp = (char*)(CALL_MMAP(rsize));
+      if (mp != CMFAIL) {
+	tbase = mp;
+	tsize = rsize;
+	mmap_flag = IS_MMAPPED_BIT;
+      }
+    }
+  }
+
+  if (HAVE_MORECORE && tbase == CMFAIL) { /* Try noncontiguous MORECORE */
+    size_t asize = granularity_align(nb + SYS_ALLOC_PADDING);
+    if (asize < HALF_MAX_SIZE_T) {
+      char* br = CMFAIL;
+      char* end = CMFAIL;
+      ACQUIRE_MALLOC_GLOBAL_LOCK();
+      br = (char*)(CALL_MORECORE(asize));
+      end = (char*)(CALL_MORECORE(0));
+      RELEASE_MALLOC_GLOBAL_LOCK();
+      if (br != CMFAIL && end != CMFAIL && br < end) {
+	size_t ssize = end - br;
+	if (ssize > nb + TOP_FOOT_SIZE) {
+	  tbase = br;
+	  tsize = ssize;
+	}
+      }
+    }
+  }
+
+  if (tbase != CMFAIL) {
+
+    if ((m->footprint += tsize) > m->max_footprint)
+      m->max_footprint = m->footprint;
+
+    if (!is_initialized(m)) { /* first-time initialization */
+      m->seg.base = m->least_addr = tbase;
+      m->seg.size = tsize;
+      m->seg.sflags = mmap_flag;
+      m->magic = mparams.magic;
+      m->release_checks = MAX_RELEASE_CHECK_RATE;
+      init_bins(m);
+#if !ONLY_MSPACES
+      if (is_global(m))
+	init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
+      else
+#endif
+      {
+	/* Offset top by embedded malloc_state */
+	mchunkptr mn = next_chunk(mem2chunk(m));
+	init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) -TOP_FOOT_SIZE);
+      }
+    }
+
+    else {
+      /* Try to merge with an existing segment */
+      msegmentptr sp = &m->seg;
+      /* Only consider most recent segment if traversal suppressed */
+      while (sp != 0 && tbase != sp->base + sp->size)
+	sp = (NO_SEGMENT_TRAVERSAL) ? 0 : sp->next;
+      if (sp != 0 &&
+	  !is_extern_segment(sp) &&
+	  (sp->sflags & IS_MMAPPED_BIT) == mmap_flag &&
+	  segment_holds(sp, m->top)) { /* append */
+	sp->size += tsize;
+	init_top(m, m->top, m->topsize + tsize);
+      }
+      else {
+	if (tbase < m->least_addr)
+	  m->least_addr = tbase;
+	sp = &m->seg;
+	while (sp != 0 && sp->base != tbase + tsize)
+	  sp = (NO_SEGMENT_TRAVERSAL) ? 0 : sp->next;
+	if (sp != 0 &&
+	    !is_extern_segment(sp) &&
+	    (sp->sflags & IS_MMAPPED_BIT) == mmap_flag) {
+	  char* oldbase = sp->base;
+	  sp->base = tbase;
+	  sp->size += tsize;
+	  return prepend_alloc(m, tbase, oldbase, nb);
+	}
+	else
+	  add_segment(m, tbase, tsize, mmap_flag);
+      }
+    }
+
+    if (nb < m->topsize) { /* Allocate from new or extended top space */
+      size_t rsize = m->topsize -= nb;
+      mchunkptr p = m->top;
+      mchunkptr r = m->top = chunk_plus_offset(p, nb);
+      r->head = rsize | PINUSE_BIT;
+      set_size_and_pinuse_of_inuse_chunk(m, p, nb);
+      check_top_chunk(m, m->top);
+      check_malloced_chunk(m, chunk2mem(p), nb);
+      return chunk2mem(p);
+    }
+  }
+
+  MALLOC_FAILURE_ACTION;
+  return 0;
+}
+
+/* -----------------------  system deallocation -------------------------- */
+
+/* Unmap and unlink any mmapped segments that don't contain used chunks */
+static size_t release_unused_segments(mstate m) {
+  size_t released = 0;
+  int nsegs = 0;
+  msegmentptr pred = &m->seg;
+  msegmentptr sp = pred->next;
+  while (sp != 0) {
+    char* base = sp->base;
+    size_t size = sp->size;
+    msegmentptr next = sp->next;
+    ++nsegs;
+    if (is_mmapped_segment(sp) && !is_extern_segment(sp)) {
+      mchunkptr p = align_as_chunk(base);
+      size_t psize = chunksize(p);
+      /* Can unmap if first chunk holds entire segment and not pinned */
+      if (!cinuse(p) && (char*)p + psize >= base + size - TOP_FOOT_SIZE) {
+	tchunkptr tp = (tchunkptr)p;
+	assert(segment_holds(sp, (char*)sp));
+	if (p == m->dv) {
+	  m->dv = 0;
+	  m->dvsize = 0;
+	}
+	else {
+	  unlink_large_chunk(m, tp);
+	}
+	if (CALL_MUNMAP(base, size) == 0) {
+	  released += size;
+	  m->footprint -= size;
+	  /* unlink obsoleted record */
+	  sp = pred;
+	  sp->next = next;
+	}
+	else { /* back out if cannot unmap */
+	  insert_large_chunk(m, tp, psize);
+	}
+      }
+    }
+    if (NO_SEGMENT_TRAVERSAL) /* scan only first segment */
+      break;
+    pred = sp;
+    sp = next;
+  }
+  /* Reset check counter */
+  m->release_checks = ((nsegs > MAX_RELEASE_CHECK_RATE)?
+		       nsegs : MAX_RELEASE_CHECK_RATE);
+  return released;
+}
+
+static int sys_trim(mstate m, size_t pad) {
+  size_t released = 0;
+  ensure_initialization();
+  if (pad < MAX_REQUEST && is_initialized(m)) {
+    pad += TOP_FOOT_SIZE; /* ensure enough room for segment overhead */
+
+    if (m->topsize > pad) {
+      /* Shrink top space in granularity-size units, keeping at least one */
+      size_t unit = mparams.granularity;
+      size_t extra = ((m->topsize - pad + (unit - SIZE_T_ONE)) / unit -
+		      SIZE_T_ONE) * unit;
+      msegmentptr sp = segment_holding(m, (char*)m->top);
+
+      if (!is_extern_segment(sp)) {
+	if (is_mmapped_segment(sp)) {
+	  if (HAVE_MMAP &&
+	      sp->size >= extra &&
+	      !has_segment_link(m, sp)) { /* can't shrink if pinned */
+	    size_t newsize = sp->size - extra;
+	    /* Prefer mremap, fall back to munmap */
+	    if ((CALL_MREMAP(sp->base, sp->size, newsize, 0) != MFAIL) ||
+		(CALL_MUNMAP(sp->base + newsize, extra) == 0)) {
+	      released = extra;
+	    }
+	  }
+	}
+	else if (HAVE_MORECORE) {
+	  if (extra >= HALF_MAX_SIZE_T) /* Avoid wrapping negative */
+	    extra = (HALF_MAX_SIZE_T) + SIZE_T_ONE - unit;
+	  ACQUIRE_MALLOC_GLOBAL_LOCK();
+	  {
+	    /* Make sure end of memory is where we last set it. */
+	    char* old_br = (char*)(CALL_MORECORE(0));
+	    if (old_br == sp->base + sp->size) {
+	      char* rel_br = (char*)(CALL_MORECORE(-extra));
+	      char* new_br = (char*)(CALL_MORECORE(0));
+	      if (rel_br != CMFAIL && new_br < old_br)
+		released = old_br - new_br;
+	    }
+	  }
+	  RELEASE_MALLOC_GLOBAL_LOCK();
+	}
+      }
+
+      if (released != 0) {
+	sp->size -= released;
+	m->footprint -= released;
+	init_top(m, m->top, m->topsize - released);
+	check_top_chunk(m, m->top);
+      }
+    }
+
+    /* Unmap any unused mmapped segments */
+    if (HAVE_MMAP)
+      released += release_unused_segments(m);
+
+    /* On failure, disable autotrim to avoid repeated failed future calls */
+    if (released == 0 && m->topsize > m->trim_check)
+      m->trim_check = MAX_SIZE_T;
+  }
+
+  return (released != 0)? 1 : 0;
+}
+
+
+/* ---------------------------- malloc support --------------------------- */
+
+/* allocate a large request from the best fitting chunk in a treebin */
+static void* tmalloc_large(mstate m, size_t nb) {
+  tchunkptr v = 0;
+  size_t rsize = -nb; /* Unsigned negation */
+  tchunkptr t;
+  bindex_t idx;
+  compute_tree_index(nb, idx);
+  if ((t = *treebin_at(m, idx)) != 0) {
+    /* Traverse tree for this bin looking for node with size == nb */
+    size_t sizebits = nb << leftshift_for_tree_index(idx);
+    tchunkptr rst = 0;  /* The deepest untaken right subtree */
+    for (;;) {
+      tchunkptr rt;
+      size_t trem = chunksize(t) - nb;
+      if (trem < rsize) {
+	v = t;
+	if ((rsize = trem) == 0)
+	  break;
+      }
+      rt = t->child[1];
+      t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
+      if (rt != 0 && rt != t)
+	rst = rt;
+      if (t == 0) {
+	t = rst; /* set t to least subtree holding sizes > nb */
+	break;
+      }
+      sizebits <<= 1;
+    }
+  }
+  if (t == 0 && v == 0) { /* set t to root of next non-empty treebin */
+    binmap_t leftbits = left_bits(idx2bit(idx)) & m->treemap;
+    if (leftbits != 0) {
+      bindex_t i;
+      binmap_t leastbit = least_bit(leftbits);
+      compute_bit2idx(leastbit, i);
+      t = *treebin_at(m, i);
+    }
+  }
+
+  while (t != 0) { /* find smallest of tree or subtree */
+    size_t trem = chunksize(t) - nb;
+    if (trem < rsize) {
+      rsize = trem;
+      v = t;
+    }
+    t = leftmost_child(t);
+  }
+
+  /*  If dv is a better fit, return 0 so malloc will use it */
+  if (v != 0 && rsize < (size_t)(m->dvsize - nb)) {
+    if (RTCHECK(ok_address(m, v))) { /* split */
+      mchunkptr r = chunk_plus_offset(v, nb);
+      assert(chunksize(v) == rsize + nb);
+      if (RTCHECK(ok_next(v, r))) {
+	unlink_large_chunk(m, v);
+	if (rsize < MIN_CHUNK_SIZE)
+	  set_inuse_and_pinuse(m, v, (rsize + nb));
+	else {
+	  set_size_and_pinuse_of_inuse_chunk(m, v, nb);
+	  set_size_and_pinuse_of_free_chunk(r, rsize);
+	  insert_chunk(m, r, rsize);
+	}
+	return chunk2mem(v);
+      }
+    }
+    CORRUPTION_ERROR_ACTION(m);
+  }
+  return 0;
+}
+
+/* allocate a small request from the best fitting chunk in a treebin */
+static void* tmalloc_small(mstate m, size_t nb) {
+  tchunkptr t, v;
+  size_t rsize;
+  bindex_t i;
+  binmap_t leastbit = least_bit(m->treemap);
+  compute_bit2idx(leastbit, i);
+  v = t = *treebin_at(m, i);
+  rsize = chunksize(t) - nb;
+
+  while ((t = leftmost_child(t)) != 0) {
+    size_t trem = chunksize(t) - nb;
+    if (trem < rsize) {
+      rsize = trem;
+      v = t;
+    }
+  }
+
+  if (RTCHECK(ok_address(m, v))) {
+    mchunkptr r = chunk_plus_offset(v, nb);
+    assert(chunksize(v) == rsize + nb);
+    if (RTCHECK(ok_next(v, r))) {
+      unlink_large_chunk(m, v);
+      if (rsize < MIN_CHUNK_SIZE)
+	set_inuse_and_pinuse(m, v, (rsize + nb));
+      else {
+	set_size_and_pinuse_of_inuse_chunk(m, v, nb);
+	set_size_and_pinuse_of_free_chunk(r, rsize);
+	replace_dv(m, r, rsize);
+      }
+      return chunk2mem(v);
+    }
+  }
+
+  CORRUPTION_ERROR_ACTION(m);
+  return 0;
+}
+
+/* --------------------------- realloc support --------------------------- */
+
+static void* internal_realloc(mstate m, void* oldmem, size_t bytes) {
+  if (bytes >= MAX_REQUEST) {
+    MALLOC_FAILURE_ACTION;
+    return 0;
+  }
+  if (!PREACTION(m)) {
+    mchunkptr oldp = mem2chunk(oldmem);
+    size_t oldsize = chunksize(oldp);
+    mchunkptr next = chunk_plus_offset(oldp, oldsize);
+    mchunkptr newp = 0;
+    void* extra = 0;
+
+    /* Try to either shrink or extend into top. Else malloc-copy-free */
+
+    if (RTCHECK(ok_address(m, oldp) && ok_cinuse(oldp) &&
+		ok_next(oldp, next) && ok_pinuse(next))) {
+      size_t nb = request2size(bytes);
+      if (is_mmapped(oldp))
+	newp = mmap_resize(m, oldp, nb);
+      else if (oldsize >= nb) { /* already big enough */
+	size_t rsize = oldsize - nb;
+	newp = oldp;
+	if (rsize >= MIN_CHUNK_SIZE) {
+	  mchunkptr remainder = chunk_plus_offset(newp, nb);
+	  set_inuse(m, newp, nb);
+	  set_inuse(m, remainder, rsize);
+	  extra = chunk2mem(remainder);
+	}
+      }
+      else if (next == m->top && oldsize + m->topsize > nb) {
+	/* Expand into top */
+	size_t newsize = oldsize + m->topsize;
+	size_t newtopsize = newsize - nb;
+	mchunkptr newtop = chunk_plus_offset(oldp, nb);
+	set_inuse(m, oldp, nb);
+	newtop->head = newtopsize |PINUSE_BIT;
+	m->top = newtop;
+	m->topsize = newtopsize;
+	newp = oldp;
+      }
+    }
+    else {
+      USAGE_ERROR_ACTION(m, oldmem);
+      POSTACTION(m);
+      return 0;
+    }
+
+    POSTACTION(m);
+
+    if (newp != 0) {
+      if (extra != 0) {
+	internal_free(m, extra);
+      }
+      check_inuse_chunk(m, newp);
+      return chunk2mem(newp);
+    }
+    else {
+      void* newmem = internal_malloc(m, bytes);
+      if (newmem != 0) {
+	size_t oc = oldsize - overhead_for(oldp);
+	memcpy(newmem, oldmem, (oc < bytes)? oc : bytes);
+	internal_free(m, oldmem);
+      }
+      return newmem;
+    }
+  }
+  return 0;
+}
+
+/* --------------------------- memalign support -------------------------- */
+
+static void* internal_memalign(mstate m, size_t alignment, size_t bytes) {
+  if (alignment <= MALLOC_ALIGNMENT)    /* Can just use malloc */
+    return internal_malloc(m, bytes);
+  if (alignment <  MIN_CHUNK_SIZE) /* must be at least a minimum chunk size */
+    alignment = MIN_CHUNK_SIZE;
+  if ((alignment & (alignment-SIZE_T_ONE)) != 0) {/* Ensure a power of 2 */
+    size_t a = MALLOC_ALIGNMENT << 1;
+    while (a < alignment) a <<= 1;
+    alignment = a;
+  }
+
+  if (bytes >= MAX_REQUEST - alignment) {
+    if (m != 0)  { /* Test isn't needed but avoids compiler warning */
+      MALLOC_FAILURE_ACTION;
+    }
+  }
+  else {
+    size_t nb = request2size(bytes);
+    size_t req = nb + alignment + MIN_CHUNK_SIZE - CHUNK_OVERHEAD;
+    char* mem = (char*)internal_malloc(m, req);
+    if (mem != 0) {
+      void* leader = 0;
+      void* trailer = 0;
+      mchunkptr p = mem2chunk(mem);
+
+      if (PREACTION(m)) return 0;
+      if ((((size_t)(mem)) % alignment) != 0) { /* misaligned */
+	/*
+	  Find an aligned spot inside chunk.  Since we need to give
+	  back leading space in a chunk of at least MIN_CHUNK_SIZE, if
+	  the first calculation places us at a spot with less than
+	  MIN_CHUNK_SIZE leader, we can move to the next aligned spot.
+	  We've allocated enough total room so that this is always
+	  possible.
+	*/
+	char* br = (char*)mem2chunk((size_t)(((size_t)(mem +
+						       alignment -
+						       SIZE_T_ONE)) &
+					     -alignment));
+	char* pos = ((size_t)(br - (char*)(p)) >= MIN_CHUNK_SIZE)?
+	  br : br+alignment;
+	mchunkptr newp = (mchunkptr)pos;
+	size_t leadsize = pos - (char*)(p);
+	size_t newsize = chunksize(p) - leadsize;
+
+	if (is_mmapped(p)) { /* For mmapped chunks, just adjust offset */
+	  newp->prev_foot = p->prev_foot + leadsize;
+	  newp->head = (newsize|CINUSE_BIT);
+	}
+	else { /* Otherwise, give back leader, use the rest */
+	  set_inuse(m, newp, newsize);
+	  set_inuse(m, p, leadsize);
+	  leader = chunk2mem(p);
+	}
+	p = newp;
+      }
+
+      /* Give back spare room at the end */
+      if (!is_mmapped(p)) {
+	size_t size = chunksize(p);
+	if (size > nb + MIN_CHUNK_SIZE) {
+	  size_t remainder_size = size - nb;
+	  mchunkptr remainder = chunk_plus_offset(p, nb);
+	  set_inuse(m, p, nb);
+	  set_inuse(m, remainder, remainder_size);
+	  trailer = chunk2mem(remainder);
+	}
+      }
+
+      assert (chunksize(p) >= nb);
+      assert((((size_t)(chunk2mem(p))) % alignment) == 0);
+      check_inuse_chunk(m, p);
+      POSTACTION(m);
+      if (leader != 0) {
+	internal_free(m, leader);
+      }
+      if (trailer != 0) {
+	internal_free(m, trailer);
+      }
+      return chunk2mem(p);
+    }
+  }
+  return 0;
+}
+
+/* ------------------------ comalloc/coalloc support --------------------- */
+
+static void** ialloc(mstate m,
+		     size_t n_elements,
+		     size_t* sizes,
+		     int opts,
+		     void* chunks[]) {
+  /*
+    This provides common support for independent_X routines, handling
+    all of the combinations that can result.
+
+    The opts arg has:
+    bit 0 set if all elements are same size (using sizes[0])
+    bit 1 set if elements should be zeroed
+  */
+
+  size_t    element_size;   /* chunksize of each element, if all same */
+  size_t    contents_size;  /* total size of elements */
+  size_t    array_size;     /* request size of pointer array */
+  void*     mem;            /* malloced aggregate space */
+  mchunkptr p;              /* corresponding chunk */
+  size_t    remainder_size; /* remaining bytes while splitting */
+  void**    marray;         /* either "chunks" or malloced ptr array */
+  mchunkptr array_chunk;    /* chunk for malloced ptr array */
+  flag_t    was_enabled;    /* to disable mmap */
+  size_t    size;
+  size_t    i;
+
+  ensure_initialization();
+  /* compute array length, if needed */
+  if (chunks != 0) {
+    if (n_elements == 0)
+      return chunks; /* nothing to do */
+    marray = chunks;
+    array_size = 0;
+  }
+  else {
+    /* if empty req, must still return chunk representing empty array */
+    if (n_elements == 0)
+      return (void**)internal_malloc(m, 0);
+    marray = 0;
+    array_size = request2size(n_elements * (sizeof(void*)));
+  }
+
+  /* compute total element size */
+  if (opts & 0x1) { /* all-same-size */
+    element_size = request2size(*sizes);
+    contents_size = n_elements * element_size;
+  }
+  else { /* add up all the sizes */
+    element_size = 0;
+    contents_size = 0;
+    for (i = 0; i != n_elements; ++i)
+      contents_size += request2size(sizes[i]);
+  }
+
+  size = contents_size + array_size;
+
+  /*
+     Allocate the aggregate chunk.  First disable direct-mmapping so
+     malloc won't use it, since we would not be able to later
+     free/realloc space internal to a segregated mmap region.
+  */
+  was_enabled = use_mmap(m);
+  disable_mmap(m);
+  mem = internal_malloc(m, size - CHUNK_OVERHEAD);
+  if (was_enabled)
+    enable_mmap(m);
+  if (mem == 0)
+    return 0;
+
+  if (PREACTION(m)) return 0;
+  p = mem2chunk(mem);
+  remainder_size = chunksize(p);
+
+  assert(!is_mmapped(p));
+
+  if (opts & 0x2) {       /* optionally clear the elements */
+    memset((size_t*)mem, 0, remainder_size - SIZE_T_SIZE - array_size);
+  }
+
+  /* If not provided, allocate the pointer array as final part of chunk */
+  if (marray == 0) {
+    size_t  array_chunk_size;
+    array_chunk = chunk_plus_offset(p, contents_size);
+    array_chunk_size = remainder_size - contents_size;
+    marray = (void**) (chunk2mem(array_chunk));
+    set_size_and_pinuse_of_inuse_chunk(m, array_chunk, array_chunk_size);
+    remainder_size = contents_size;
+  }
+
+  /* split out elements */
+  for (i = 0; ; ++i) {
+    marray[i] = chunk2mem(p);
+    if (i != n_elements-1) {
+      if (element_size != 0)
+	size = element_size;
+      else
+	size = request2size(sizes[i]);
+      remainder_size -= size;
+      set_size_and_pinuse_of_inuse_chunk(m, p, size);
+      p = chunk_plus_offset(p, size);
+    }
+    else { /* the final element absorbs any overallocation slop */
+      set_size_and_pinuse_of_inuse_chunk(m, p, remainder_size);
+      break;
+    }
+  }
+
+#if DEBUG
+  if (marray != chunks) {
+    /* final element must have exactly exhausted chunk */
+    if (element_size != 0) {
+      assert(remainder_size == element_size);
+    }
+    else {
+      assert(remainder_size == request2size(sizes[i]));
+    }
+    check_inuse_chunk(m, mem2chunk(marray));
+  }
+  for (i = 0; i != n_elements; ++i)
+    check_inuse_chunk(m, mem2chunk(marray[i]));
+
+#endif /* DEBUG */
+
+  POSTACTION(m);
+  return marray;
+}
+
+
+/* -------------------------- public routines ---------------------------- */
+
+#if !ONLY_MSPACES
+
+void* dlmalloc(size_t bytes) {
+  /*
+     Basic algorithm:
+     If a small request (< 256 bytes minus per-chunk overhead):
+       1. If one exists, use a remainderless chunk in associated smallbin.
+	  (Remainderless means that there are too few excess bytes to
+	  represent as a chunk.)
+       2. If it is big enough, use the dv chunk, which is normally the
+	  chunk adjacent to the one used for the most recent small request.
+       3. If one exists, split the smallest available chunk in a bin,
+	  saving remainder in dv.
+       4. If it is big enough, use the top chunk.
+       5. If available, get memory from system and use it
+     Otherwise, for a large request:
+       1. Find the smallest available binned chunk that fits, and use it
+	  if it is better fitting than dv chunk, splitting if necessary.
+       2. If better fitting than any binned chunk, use the dv chunk.
+       3. If it is big enough, use the top chunk.
+       4. If request size >= mmap threshold, try to directly mmap this chunk.
+       5. If available, get memory from system and use it
+
+     The ugly goto's here ensure that postaction occurs along all paths.
+  */
+
+#if USE_LOCKS
+  ensure_initialization(); /* initialize in sys_alloc if not using locks */
+#endif
+
+  if (!PREACTION(gm)) {
+    void* mem;
+    size_t nb;
+    if (bytes <= MAX_SMALL_REQUEST) {
+      bindex_t idx;
+      binmap_t smallbits;
+      nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
+      idx = small_index(nb);
+      smallbits = gm->smallmap >> idx;
+
+      if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
+	mchunkptr b, p;
+	idx += ~smallbits & 1;       /* Uses next bin if idx empty */
+	b = smallbin_at(gm, idx);
+	p = b->fd;
+	assert(chunksize(p) == small_index2size(idx));
+	unlink_first_small_chunk(gm, b, p, idx);
+	set_inuse_and_pinuse(gm, p, small_index2size(idx));
+	mem = chunk2mem(p);
+	check_malloced_chunk(gm, mem, nb);
+	goto postaction;
+      }
+
+      else if (nb > gm->dvsize) {
+	if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
+	  mchunkptr b, p, r;
+	  size_t rsize;
+	  bindex_t i;
+	  binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
+	  binmap_t leastbit = least_bit(leftbits);
+	  compute_bit2idx(leastbit, i);
+	  b = smallbin_at(gm, i);
+	  p = b->fd;
+	  assert(chunksize(p) == small_index2size(i));
+	  unlink_first_small_chunk(gm, b, p, i);
+	  rsize = small_index2size(i) - nb;
+	  /* Fit here cannot be remainderless if 4byte sizes */
+	  if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
+	    set_inuse_and_pinuse(gm, p, small_index2size(i));
+	  else {
+	    set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
+	    r = chunk_plus_offset(p, nb);
+	    set_size_and_pinuse_of_free_chunk(r, rsize);
+	    replace_dv(gm, r, rsize);
+	  }
+	  mem = chunk2mem(p);
+	  check_malloced_chunk(gm, mem, nb);
+	  goto postaction;
+	}
+
+	else if (gm->treemap != 0 && (mem = tmalloc_small(gm, nb)) != 0) {
+	  check_malloced_chunk(gm, mem, nb);
+	  goto postaction;
+	}
+      }
+    }
+    else if (bytes >= MAX_REQUEST)
+      nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
+    else {
+      nb = pad_request(bytes);
+      if (gm->treemap != 0 && (mem = tmalloc_large(gm, nb)) != 0) {
+	check_malloced_chunk(gm, mem, nb);
+	goto postaction;
+      }
+    }
+
+    if (nb <= gm->dvsize) {
+      size_t rsize = gm->dvsize - nb;
+      mchunkptr p = gm->dv;
+      if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
+	mchunkptr r = gm->dv = chunk_plus_offset(p, nb);
+	gm->dvsize = rsize;
+	set_size_and_pinuse_of_free_chunk(r, rsize);
+	set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
+      }
+      else { /* exhaust dv */
+	size_t dvs = gm->dvsize;
+	gm->dvsize = 0;
+	gm->dv = 0;
+	set_inuse_and_pinuse(gm, p, dvs);
+      }
+      mem = chunk2mem(p);
+      check_malloced_chunk(gm, mem, nb);
+      goto postaction;
+    }
+
+    else if (nb < gm->topsize) { /* Split top */
+      size_t rsize = gm->topsize -= nb;
+      mchunkptr p = gm->top;
+      mchunkptr r = gm->top = chunk_plus_offset(p, nb);
+      r->head = rsize | PINUSE_BIT;
+      set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
+      mem = chunk2mem(p);
+      check_top_chunk(gm, gm->top);
+      check_malloced_chunk(gm, mem, nb);
+      goto postaction;
+    }
+
+    mem = sys_alloc(gm, nb);
+
+  postaction:
+    POSTACTION(gm);
+    return mem;
+  }
+
+  return 0;
+}
+
+void dlfree(void* mem) {
+  /*
+     Consolidate freed chunks with preceding or succeeding bordering
+     free chunks, if they exist, and then place in a bin.  Intermixed
+     with special cases for top, dv, mmapped chunks, and usage errors.
+  */
+
+  if (mem != 0) {
+    mchunkptr p  = mem2chunk(mem);
+#if FOOTERS
+    mstate fm = get_mstate_for(p);
+    if (!ok_magic(fm)) {
+      USAGE_ERROR_ACTION(fm, p);
+      return;
+    }
+#else /* FOOTERS */
+#define fm gm
+#endif /* FOOTERS */
+    if (!PREACTION(fm)) {
+      check_inuse_chunk(fm, p);
+      if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
+	size_t psize = chunksize(p);
+	mchunkptr next = chunk_plus_offset(p, psize);
+	if (!pinuse(p)) {
+	  size_t prevsize = p->prev_foot;
+	  if ((prevsize & IS_MMAPPED_BIT) != 0) {
+	    prevsize &= ~IS_MMAPPED_BIT;
+	    psize += prevsize + MMAP_FOOT_PAD;
+	    if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
+	      fm->footprint -= psize;
+	    goto postaction;
+	  }
+	  else {
+	    mchunkptr prev = chunk_minus_offset(p, prevsize);
+	    psize += prevsize;
+	    p = prev;
+	    if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
+	      if (p != fm->dv) {
+		unlink_chunk(fm, p, prevsize);
+	      }
+	      else if ((next->head & INUSE_BITS) == INUSE_BITS) {
+		fm->dvsize = psize;
+		set_free_with_pinuse(p, psize, next);
+		goto postaction;
+	      }
+	    }
+	    else
+	      goto erroraction;
+	  }
+	}
+
+	if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
+	  if (!cinuse(next)) {  /* consolidate forward */
+	    if (next == fm->top) {
+	      size_t tsize = fm->topsize += psize;
+	      fm->top = p;
+	      p->head = tsize | PINUSE_BIT;
+	      if (p == fm->dv) {
+		fm->dv = 0;
+		fm->dvsize = 0;
+	      }
+	      if (should_trim(fm, tsize))
+		sys_trim(fm, 0);
+	      goto postaction;
+	    }
+	    else if (next == fm->dv) {
+	      size_t dsize = fm->dvsize += psize;
+	      fm->dv = p;
+	      set_size_and_pinuse_of_free_chunk(p, dsize);
+	      goto postaction;
+	    }
+	    else {
+	      size_t nsize = chunksize(next);
+	      psize += nsize;
+	      unlink_chunk(fm, next, nsize);
+	      set_size_and_pinuse_of_free_chunk(p, psize);
+	      if (p == fm->dv) {
+		fm->dvsize = psize;
+		goto postaction;
+	      }
+	    }
+	  }
+	  else
+	    set_free_with_pinuse(p, psize, next);
+
+	  if (is_small(psize)) {
+	    insert_small_chunk(fm, p, psize);
+	    check_free_chunk(fm, p);
+	  }
+	  else {
+	    tchunkptr tp = (tchunkptr)p;
+	    insert_large_chunk(fm, tp, psize);
+	    check_free_chunk(fm, p);
+	    if (--fm->release_checks == 0)
+	      release_unused_segments(fm);
+	  }
+	  goto postaction;
+	}
+      }
+    erroraction:
+      USAGE_ERROR_ACTION(fm, p);
+    postaction:
+      POSTACTION(fm);
+    }
+  }
+#if !FOOTERS
+#undef fm
+#endif /* FOOTERS */
+}
+
+void* dlcalloc(size_t n_elements, size_t elem_size) {
+  void* mem;
+  size_t req = 0;
+  if (n_elements != 0) {
+    req = n_elements * elem_size;
+    if (((n_elements | elem_size) & ~(size_t)0xffff) &&
+	(req / n_elements != elem_size))
+      req = MAX_SIZE_T; /* force downstream failure on overflow */
+  }
+  mem = dlmalloc(req);
+  if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
+    memset(mem, 0, req);
+  return mem;
+}
+
+void* dlrealloc(void* oldmem, size_t bytes) {
+  if (oldmem == 0)
+    return dlmalloc(bytes);
+#ifdef REALLOC_ZERO_BYTES_FREES
+  if (bytes == 0) {
+    dlfree(oldmem);
+    return 0;
+  }
+#endif /* REALLOC_ZERO_BYTES_FREES */
+  else {
+#if ! FOOTERS
+    mstate m = gm;
+#else /* FOOTERS */
+    mstate m = get_mstate_for(mem2chunk(oldmem));
+    if (!ok_magic(m)) {
+      USAGE_ERROR_ACTION(m, oldmem);
+      return 0;
+    }
+#endif /* FOOTERS */
+    return internal_realloc(m, oldmem, bytes);
+  }
+}
+
+void* dlmemalign(size_t alignment, size_t bytes) {
+  return internal_memalign(gm, alignment, bytes);
+}
+
+void** dlindependent_calloc(size_t n_elements, size_t elem_size,
+				 void* chunks[]) {
+  size_t sz = elem_size; /* serves as 1-element array */
+  return ialloc(gm, n_elements, &sz, 3, chunks);
+}
+
+void** dlindependent_comalloc(size_t n_elements, size_t sizes[],
+				   void* chunks[]) {
+  return ialloc(gm, n_elements, sizes, 0, chunks);
+}
+
+void* dlvalloc(size_t bytes) {
+  size_t pagesz;
+  ensure_initialization();
+  pagesz = mparams.page_size;
+  return dlmemalign(pagesz, bytes);
+}
+
+void* dlpvalloc(size_t bytes) {
+  size_t pagesz;
+  ensure_initialization();
+  pagesz = mparams.page_size;
+  return dlmemalign(pagesz, (bytes + pagesz - SIZE_T_ONE) & ~(pagesz - SIZE_T_ONE));
+}
+
+int dlmalloc_trim(size_t pad) {
+  ensure_initialization();
+  int result = 0;
+  if (!PREACTION(gm)) {
+    result = sys_trim(gm, pad);
+    POSTACTION(gm);
+  }
+  return result;
+}
+
+size_t dlmalloc_footprint(void) {
+  return gm->footprint;
+}
+
+size_t dlmalloc_max_footprint(void) {
+  return gm->max_footprint;
+}
+
+#if !NO_MALLINFO
+struct mallinfo dlmallinfo(void) {
+  return internal_mallinfo(gm);
+}
+#endif /* NO_MALLINFO */
+
+void dlmalloc_stats() {
+  internal_malloc_stats(gm);
+}
+
+int dlmallopt(int param_number, int value) {
+  return change_mparam(param_number, value);
+}
+
+#endif /* !ONLY_MSPACES */
+
+size_t dlmalloc_usable_size(void* mem) {
+  if (mem != 0) {
+    mchunkptr p = mem2chunk(mem);
+    if (cinuse(p))
+      return chunksize(p) - overhead_for(p);
+  }
+  return 0;
+}
+
+/* ----------------------------- user mspaces ---------------------------- */
+
+#if MSPACES
+
+static mstate init_user_mstate(char* tbase, size_t tsize) {
+  size_t msize = pad_request(sizeof(struct malloc_state));
+  mchunkptr mn;
+  mchunkptr msp = align_as_chunk(tbase);
+  mstate m = (mstate)(chunk2mem(msp));
+  memset(m, 0, msize);
+  INITIAL_LOCK(&m->mutex);
+  msp->head = (msize|PINUSE_BIT|CINUSE_BIT);
+  m->seg.base = m->least_addr = tbase;
+  m->seg.size = m->footprint = m->max_footprint = tsize;
+  m->magic = mparams.magic;
+  m->release_checks = MAX_RELEASE_CHECK_RATE;
+  m->mflags = mparams.default_mflags;
+  m->extp = 0;
+  m->exts = 0;
+  disable_contiguous(m);
+  init_bins(m);
+  mn = next_chunk(mem2chunk(m));
+  init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) - TOP_FOOT_SIZE);
+  check_top_chunk(m, m->top);
+  return m;
+}
+
+mspace create_mspace(size_t capacity, int locked) {
+  mstate m = 0;
+  size_t msize;
+  ensure_initialization();
+  msize = pad_request(sizeof(struct malloc_state));
+  if (capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
+    size_t rs = ((capacity == 0)? mparams.granularity :
+		 (capacity + TOP_FOOT_SIZE + msize));
+    size_t tsize = granularity_align(rs);
+    char* tbase = (char*)(CALL_MMAP(tsize));
+    if (tbase != CMFAIL) {
+      m = init_user_mstate(tbase, tsize);
+      m->seg.sflags = IS_MMAPPED_BIT;
+      set_lock(m, locked);
+    }
+  }
+  return (mspace)m;
+}
+
+mspace create_mspace_with_base(void* base, size_t capacity, int locked) {
+  mstate m = 0;
+  size_t msize;
+  ensure_initialization();
+  msize = pad_request(sizeof(struct malloc_state));
+  if (capacity > msize + TOP_FOOT_SIZE &&
+      capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
+    m = init_user_mstate((char*)base, capacity);
+    m->seg.sflags = EXTERN_BIT;
+    set_lock(m, locked);
+  }
+  return (mspace)m;
+}
+
+int mspace_mmap_large_chunks(mspace msp, int enable) {
+  int ret = 0;
+  mstate ms = (mstate)msp;
+  if (!PREACTION(ms)) {
+    if (use_mmap(ms))
+      ret = 1;
+    if (enable)
+      enable_mmap(ms);
+    else
+      disable_mmap(ms);
+    POSTACTION(ms);
+  }
+  return ret;
+}
+
+size_t destroy_mspace(mspace msp) {
+  size_t freed = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    msegmentptr sp = &ms->seg;
+    while (sp != 0) {
+      char* base = sp->base;
+      size_t size = sp->size;
+      flag_t flag = sp->sflags;
+      sp = sp->next;
+      if ((flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) &&
+	  CALL_MUNMAP(base, size) == 0)
+	freed += size;
+    }
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return freed;
+}
+
+/*
+  mspace versions of routines are near-clones of the global
+  versions. This is not so nice but better than the alternatives.
+*/
+
+
+void* mspace_malloc(mspace msp, size_t bytes) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  if (!PREACTION(ms)) {
+    void* mem;
+    size_t nb;
+    if (bytes <= MAX_SMALL_REQUEST) {
+      bindex_t idx;
+      binmap_t smallbits;
+      nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
+      idx = small_index(nb);
+      smallbits = ms->smallmap >> idx;
+
+      if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
+	mchunkptr b, p;
+	idx += ~smallbits & 1;       /* Uses next bin if idx empty */
+	b = smallbin_at(ms, idx);
+	p = b->fd;
+	assert(chunksize(p) == small_index2size(idx));
+	unlink_first_small_chunk(ms, b, p, idx);
+	set_inuse_and_pinuse(ms, p, small_index2size(idx));
+	mem = chunk2mem(p);
+	check_malloced_chunk(ms, mem, nb);
+	goto postaction;
+      }
+
+      else if (nb > ms->dvsize) {
+	if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
+	  mchunkptr b, p, r;
+	  size_t rsize;
+	  bindex_t i;
+	  binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
+	  binmap_t leastbit = least_bit(leftbits);
+	  compute_bit2idx(leastbit, i);
+	  b = smallbin_at(ms, i);
+	  p = b->fd;
+	  assert(chunksize(p) == small_index2size(i));
+	  unlink_first_small_chunk(ms, b, p, i);
+	  rsize = small_index2size(i) - nb;
+	  /* Fit here cannot be remainderless if 4byte sizes */
+	  if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
+	    set_inuse_and_pinuse(ms, p, small_index2size(i));
+	  else {
+	    set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
+	    r = chunk_plus_offset(p, nb);
+	    set_size_and_pinuse_of_free_chunk(r, rsize);
+	    replace_dv(ms, r, rsize);
+	  }
+	  mem = chunk2mem(p);
+	  check_malloced_chunk(ms, mem, nb);
+	  goto postaction;
+	}
+
+	else if (ms->treemap != 0 && (mem = tmalloc_small(ms, nb)) != 0) {
+	  check_malloced_chunk(ms, mem, nb);
+	  goto postaction;
+	}
+      }
+    }
+    else if (bytes >= MAX_REQUEST)
+      nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
+    else {
+      nb = pad_request(bytes);
+      if (ms->treemap != 0 && (mem = tmalloc_large(ms, nb)) != 0) {
+	check_malloced_chunk(ms, mem, nb);
+	goto postaction;
+      }
+    }
+
+    if (nb <= ms->dvsize) {
+      size_t rsize = ms->dvsize - nb;
+      mchunkptr p = ms->dv;
+      if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
+	mchunkptr r = ms->dv = chunk_plus_offset(p, nb);
+	ms->dvsize = rsize;
+	set_size_and_pinuse_of_free_chunk(r, rsize);
+	set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
+      }
+      else { /* exhaust dv */
+	size_t dvs = ms->dvsize;
+	ms->dvsize = 0;
+	ms->dv = 0;
+	set_inuse_and_pinuse(ms, p, dvs);
+      }
+      mem = chunk2mem(p);
+      check_malloced_chunk(ms, mem, nb);
+      goto postaction;
+    }
+
+    else if (nb < ms->topsize) { /* Split top */
+      size_t rsize = ms->topsize -= nb;
+      mchunkptr p = ms->top;
+      mchunkptr r = ms->top = chunk_plus_offset(p, nb);
+      r->head = rsize | PINUSE_BIT;
+      set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
+      mem = chunk2mem(p);
+      check_top_chunk(ms, ms->top);
+      check_malloced_chunk(ms, mem, nb);
+      goto postaction;
+    }
+
+    mem = sys_alloc(ms, nb);
+
+  postaction:
+    POSTACTION(ms);
+    return mem;
+  }
+
+  return 0;
+}
+
+void mspace_free(mspace msp, void* mem) {
+  if (mem != 0) {
+    mchunkptr p  = mem2chunk(mem);
+#if FOOTERS
+    mstate fm = get_mstate_for(p);
+#else /* FOOTERS */
+    mstate fm = (mstate)msp;
+#endif /* FOOTERS */
+    if (!ok_magic(fm)) {
+      USAGE_ERROR_ACTION(fm, p);
+      return;
+    }
+    if (!PREACTION(fm)) {
+      check_inuse_chunk(fm, p);
+      if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
+	size_t psize = chunksize(p);
+	mchunkptr next = chunk_plus_offset(p, psize);
+	if (!pinuse(p)) {
+	  size_t prevsize = p->prev_foot;
+	  if ((prevsize & IS_MMAPPED_BIT) != 0) {
+	    prevsize &= ~IS_MMAPPED_BIT;
+	    psize += prevsize + MMAP_FOOT_PAD;
+	    if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
+	      fm->footprint -= psize;
+	    goto postaction;
+	  }
+	  else {
+	    mchunkptr prev = chunk_minus_offset(p, prevsize);
+	    psize += prevsize;
+	    p = prev;
+	    if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
+	      if (p != fm->dv) {
+		unlink_chunk(fm, p, prevsize);
+	      }
+	      else if ((next->head & INUSE_BITS) == INUSE_BITS) {
+		fm->dvsize = psize;
+		set_free_with_pinuse(p, psize, next);
+		goto postaction;
+	      }
+	    }
+	    else
+	      goto erroraction;
+	  }
+	}
+
+	if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
+	  if (!cinuse(next)) {  /* consolidate forward */
+	    if (next == fm->top) {
+	      size_t tsize = fm->topsize += psize;
+	      fm->top = p;
+	      p->head = tsize | PINUSE_BIT;
+	      if (p == fm->dv) {
+		fm->dv = 0;
+		fm->dvsize = 0;
+	      }
+	      if (should_trim(fm, tsize))
+		sys_trim(fm, 0);
+	      goto postaction;
+	    }
+	    else if (next == fm->dv) {
+	      size_t dsize = fm->dvsize += psize;
+	      fm->dv = p;
+	      set_size_and_pinuse_of_free_chunk(p, dsize);
+	      goto postaction;
+	    }
+	    else {
+	      size_t nsize = chunksize(next);
+	      psize += nsize;
+	      unlink_chunk(fm, next, nsize);
+	      set_size_and_pinuse_of_free_chunk(p, psize);
+	      if (p == fm->dv) {
+		fm->dvsize = psize;
+		goto postaction;
+	      }
+	    }
+	  }
+	  else
+	    set_free_with_pinuse(p, psize, next);
+
+	  if (is_small(psize)) {
+	    insert_small_chunk(fm, p, psize);
+	    check_free_chunk(fm, p);
+	  }
+	  else {
+	    tchunkptr tp = (tchunkptr)p;
+	    insert_large_chunk(fm, tp, psize);
+	    check_free_chunk(fm, p);
+	    if (--fm->release_checks == 0)
+	      release_unused_segments(fm);
+	  }
+	  goto postaction;
+	}
+      }
+    erroraction:
+      USAGE_ERROR_ACTION(fm, p);
+    postaction:
+      POSTACTION(fm);
+    }
+  }
+}
+
+void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size) {
+  void* mem;
+  size_t req = 0;
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  if (n_elements != 0) {
+    req = n_elements * elem_size;
+    if (((n_elements | elem_size) & ~(size_t)0xffff) &&
+	(req / n_elements != elem_size))
+      req = MAX_SIZE_T; /* force downstream failure on overflow */
+  }
+  mem = internal_malloc(ms, req);
+  if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
+    memset(mem, 0, req);
+  return mem;
+}
+
+void* mspace_realloc(mspace msp, void* oldmem, size_t bytes) {
+  if (oldmem == 0)
+    return mspace_malloc(msp, bytes);
+#ifdef REALLOC_ZERO_BYTES_FREES
+  if (bytes == 0) {
+    mspace_free(msp, oldmem);
+    return 0;
+  }
+#endif /* REALLOC_ZERO_BYTES_FREES */
+  else {
+#if FOOTERS
+    mchunkptr p  = mem2chunk(oldmem);
+    mstate ms = get_mstate_for(p);
+#else /* FOOTERS */
+    mstate ms = (mstate)msp;
+#endif /* FOOTERS */
+    if (!ok_magic(ms)) {
+      USAGE_ERROR_ACTION(ms,ms);
+      return 0;
+    }
+    return internal_realloc(ms, oldmem, bytes);
+  }
+}
+
+void* mspace_memalign(mspace msp, size_t alignment, size_t bytes) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  return internal_memalign(ms, alignment, bytes);
+}
+
+void** mspace_independent_calloc(mspace msp, size_t n_elements,
+				 size_t elem_size, void* chunks[]) {
+  size_t sz = elem_size; /* serves as 1-element array */
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  return ialloc(ms, n_elements, &sz, 3, chunks);
+}
+
+void** mspace_independent_comalloc(mspace msp, size_t n_elements,
+				   size_t sizes[], void* chunks[]) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+    return 0;
+  }
+  return ialloc(ms, n_elements, sizes, 0, chunks);
+}
+
+int mspace_trim(mspace msp, size_t pad) {
+  int result = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    if (!PREACTION(ms)) {
+      result = sys_trim(ms, pad);
+      POSTACTION(ms);
+    }
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return result;
+}
+
+void mspace_malloc_stats(mspace msp) {
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    internal_malloc_stats(ms);
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+}
+
+size_t mspace_footprint(mspace msp) {
+  size_t result = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    result = ms->footprint;
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return result;
+}
+
+
+size_t mspace_max_footprint(mspace msp) {
+  size_t result = 0;
+  mstate ms = (mstate)msp;
+  if (ok_magic(ms)) {
+    result = ms->max_footprint;
+  }
+  else {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return result;
+}
+
+
+#if !NO_MALLINFO
+struct mallinfo mspace_mallinfo(mspace msp) {
+  mstate ms = (mstate)msp;
+  if (!ok_magic(ms)) {
+    USAGE_ERROR_ACTION(ms,ms);
+  }
+  return internal_mallinfo(ms);
+}
+#endif /* NO_MALLINFO */
+
+size_t mspace_usable_size(void* mem) {
+  if (mem != 0) {
+    mchunkptr p = mem2chunk(mem);
+    if (cinuse(p))
+      return chunksize(p) - overhead_for(p);
+  }
+  return 0;
+}
+
+int mspace_mallopt(int param_number, int value) {
+  return change_mparam(param_number, value);
+}
+
+#endif /* MSPACES */
+
+/* -------------------- Alternative MORECORE functions ------------------- */
+
+/*
+  Guidelines for creating a custom version of MORECORE:
+
+  * For best performance, MORECORE should allocate in multiples of pagesize.
+  * MORECORE may allocate more memory than requested. (Or even less,
+      but this will usually result in a malloc failure.)
+  * MORECORE must not allocate memory when given argument zero, but
+      instead return one past the end address of memory from previous
+      nonzero call.
+  * For best performance, consecutive calls to MORECORE with positive
+      arguments should return increasing addresses, indicating that
+      space has been contiguously extended.
+  * Even though consecutive calls to MORECORE need not return contiguous
+      addresses, it must be OK for malloc'ed chunks to span multiple
+      regions in those cases where they do happen to be contiguous.
+  * MORECORE need not handle negative arguments -- it may instead
+      just return MFAIL when given negative arguments.
+      Negative arguments are always multiples of pagesize. MORECORE
+      must not misinterpret negative args as large positive unsigned
+      args. You can suppress all such calls from even occurring by defining
+      MORECORE_CANNOT_TRIM,
+
+  As an example alternative MORECORE, here is a custom allocator
+  kindly contributed for pre-OSX macOS.  It uses virtually but not
+  necessarily physically contiguous non-paged memory (locked in,
+  present and won't get swapped out).  You can use it by uncommenting
+  this section, adding some #includes, and setting up the appropriate
+  defines above:
+
+      #define MORECORE osMoreCore
+
+  There is also a shutdown routine that should somehow be called for
+  cleanup upon program exit.
+
+  #define MAX_POOL_ENTRIES 100
+  #define MINIMUM_MORECORE_SIZE  (64 * 1024U)
+  static int next_os_pool;
+  void *our_os_pools[MAX_POOL_ENTRIES];
+
+  void *osMoreCore(int size)
+  {
+    void *ptr = 0;
+    static void *sbrk_top = 0;
+
+    if (size > 0)
+    {
+      if (size < MINIMUM_MORECORE_SIZE)
+	 size = MINIMUM_MORECORE_SIZE;
+      if (CurrentExecutionLevel() == kTaskLevel)
+	 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
+      if (ptr == 0)
+      {
+	return (void *) MFAIL;
+      }
+      // save ptrs so they can be freed during cleanup
+      our_os_pools[next_os_pool] = ptr;
+      next_os_pool++;
+      ptr = (void *) ((((size_t) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
+      sbrk_top = (char *) ptr + size;
+      return ptr;
+    }
+    else if (size < 0)
+    {
+      // we don't currently support shrink behavior
+      return (void *) MFAIL;
+    }
+    else
+    {
+      return sbrk_top;
+    }
+  }
+
+  // cleanup any allocated memory pools
+  // called as last thing before shutting down driver
+
+  void osCleanupMem(void)
+  {
+    void **ptr;
+
+    for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
+      if (*ptr)
+      {
+	 PoolDeallocate(*ptr);
+	 *ptr = 0;
+      }
+  }
+
+*/
+
+
+/* -----------------------------------------------------------------------
+History:
+    V2.8.4 (not yet released)
+      * Add mspace_mmap_large_chunks; thanks to Jean Brouwers
+      * Fix insufficient sys_alloc padding when using 16byte alignment
+      * Fix bad error check in mspace_footprint
+      * Adaptations for ptmalloc, courtesy of Wolfram Gloger.
+      * Reentrant spin locks, courtesy of Earl Chew and others
+      * Win32 improvements, courtesy of Niall Douglas and Earl Chew
+      * Add NO_SEGMENT_TRAVERSAL and MAX_RELEASE_CHECK_RATE options
+      * Extension hook in malloc_state
+      * Various small adjustments to reduce warnings on some compilers
+      * Various configuration extensions/changes for more platforms. Thanks
+	 to all who contributed these.
+
+    V2.8.3 Thu Sep 22 11:16:32 2005  Doug Lea  (dl at gee)
+      * Add max_footprint functions
+      * Ensure all appropriate literals are size_t
+      * Fix conditional compilation problem for some #define settings
+      * Avoid concatenating segments with the one provided
+	in create_mspace_with_base
+      * Rename some variables to avoid compiler shadowing warnings
+      * Use explicit lock initialization.
+      * Better handling of sbrk interference.
+      * Simplify and fix segment insertion, trimming and mspace_destroy
+      * Reinstate REALLOC_ZERO_BYTES_FREES option from 2.7.x
+      * Thanks especially to Dennis Flanagan for help on these.
+
+    V2.8.2 Sun Jun 12 16:01:10 2005  Doug Lea  (dl at gee)
+      * Fix memalign brace error.
+
+    V2.8.1 Wed Jun  8 16:11:46 2005  Doug Lea  (dl at gee)
+      * Fix improper #endif nesting in C++
+      * Add explicit casts needed for C++
+
+    V2.8.0 Mon May 30 14:09:02 2005  Doug Lea  (dl at gee)
+      * Use trees for large bins
+      * Support mspaces
+      * Use segments to unify sbrk-based and mmap-based system allocation,
+	removing need for emulation on most platforms without sbrk.
+      * Default safety checks
+      * Optional footer checks. Thanks to William Robertson for the idea.
+      * Internal code refactoring
+      * Incorporate suggestions and platform-specific changes.
+	Thanks to Dennis Flanagan, Colin Plumb, Niall Douglas,
+	Aaron Bachmann,  Emery Berger, and others.
+      * Speed up non-fastbin processing enough to remove fastbins.
+      * Remove useless cfree() to avoid conflicts with other apps.
+      * Remove internal memcpy, memset. Compilers handle builtins better.
+      * Remove some options that no one ever used and rename others.
+
+    V2.7.2 Sat Aug 17 09:07:30 2002  Doug Lea  (dl at gee)
+      * Fix malloc_state bitmap array misdeclaration
+
+    V2.7.1 Thu Jul 25 10:58:03 2002  Doug Lea  (dl at gee)
+      * Allow tuning of FIRST_SORTED_BIN_SIZE
+      * Use PTR_UINT as type for all ptr->int casts. Thanks to John Belmonte.
+      * Better detection and support for non-contiguousness of MORECORE.
+	Thanks to Andreas Mueller, Conal Walsh, and Wolfram Gloger
+      * Bypass most of malloc if no frees. Thanks To Emery Berger.
+      * Fix freeing of old top non-contiguous chunk im sysmalloc.
+      * Raised default trim and map thresholds to 256K.
+      * Fix mmap-related #defines. Thanks to Lubos Lunak.
+      * Fix copy macros; added LACKS_FCNTL_H. Thanks to Neal Walfield.
+      * Branch-free bin calculation
+      * Default trim and mmap thresholds now 256K.
+
+    V2.7.0 Sun Mar 11 14:14:06 2001  Doug Lea  (dl at gee)
+      * Introduce independent_comalloc and independent_calloc.
+	Thanks to Michael Pachos for motivation and help.
+      * Make optional .h file available
+      * Allow > 2GB requests on 32bit systems.
+      * new WIN32 sbrk, mmap, munmap, lock code from <Walter@GeNeSys-e.de>.
+	Thanks also to Andreas Mueller <a.mueller at paradatec.de>,
+	and Anonymous.
+      * Allow override of MALLOC_ALIGNMENT (Thanks to Ruud Waij for
+	helping test this.)
+      * memalign: check alignment arg
+      * realloc: don't try to shift chunks backwards, since this
+	leads to  more fragmentation in some programs and doesn't
+	seem to help in any others.
+      * Collect all cases in malloc requiring system memory into sysmalloc
+      * Use mmap as backup to sbrk
+      * Place all internal state in malloc_state
+      * Introduce fastbins (although similar to 2.5.1)
+      * Many minor tunings and cosmetic improvements
+      * Introduce USE_PUBLIC_MALLOC_WRAPPERS, USE_MALLOC_LOCK
+      * Introduce MALLOC_FAILURE_ACTION, MORECORE_CONTIGUOUS
+	Thanks to Tony E. Bennett <tbennett@nvidia.com> and others.
+      * Include errno.h to support default failure action.
+
+    V2.6.6 Sun Dec  5 07:42:19 1999  Doug Lea  (dl at gee)
+      * return null for negative arguments
+      * Added Several WIN32 cleanups from Martin C. Fong <mcfong at yahoo.com>
+	 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
+	  (e.g. WIN32 platforms)
+	 * Cleanup header file inclusion for WIN32 platforms
+	 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
+	 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
+	   memory allocation routines
+	 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
+	 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
+	   usage of 'assert' in non-WIN32 code
+	 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
+	   avoid infinite loop
+      * Always call 'fREe()' rather than 'free()'
+
+    V2.6.5 Wed Jun 17 15:57:31 1998  Doug Lea  (dl at gee)
+      * Fixed ordering problem with boundary-stamping
+
+    V2.6.3 Sun May 19 08:17:58 1996  Doug Lea  (dl at gee)
+      * Added pvalloc, as recommended by H.J. Liu
+      * Added 64bit pointer support mainly from Wolfram Gloger
+      * Added anonymously donated WIN32 sbrk emulation
+      * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
+      * malloc_extend_top: fix mask error that caused wastage after
+	foreign sbrks
+      * Add linux mremap support code from HJ Liu
+
+    V2.6.2 Tue Dec  5 06:52:55 1995  Doug Lea  (dl at gee)
+      * Integrated most documentation with the code.
+      * Add support for mmap, with help from
+	Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
+      * Use last_remainder in more cases.
+      * Pack bins using idea from  colin@nyx10.cs.du.edu
+      * Use ordered bins instead of best-fit threshold
+      * Eliminate block-local decls to simplify tracing and debugging.
+      * Support another case of realloc via move into top
+      * Fix error occurring when initial sbrk_base not word-aligned.
+      * Rely on page size for units instead of SBRK_UNIT to
+	avoid surprises about sbrk alignment conventions.
+      * Add mallinfo, mallopt. Thanks to Raymond Nijssen
+	(raymond@es.ele.tue.nl) for the suggestion.
+      * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
+      * More precautions for cases where other routines call sbrk,
+	courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
+      * Added macros etc., allowing use in linux libc from
+	H.J. Lu (hjl@gnu.ai.mit.edu)
+      * Inverted this history list
+
+    V2.6.1 Sat Dec  2 14:10:57 1995  Doug Lea  (dl at gee)
+      * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
+      * Removed all preallocation code since under current scheme
+	the work required to undo bad preallocations exceeds
+	the work saved in good cases for most test programs.
+      * No longer use return list or unconsolidated bins since
+	no scheme using them consistently outperforms those that don't
+	given above changes.
+      * Use best fit for very large chunks to prevent some worst-cases.
+      * Added some support for debugging
+
+    V2.6.0 Sat Nov  4 07:05:23 1995  Doug Lea  (dl at gee)
+      * Removed footers when chunks are in use. Thanks to
+	Paul Wilson (wilson@cs.texas.edu) for the suggestion.
+
+    V2.5.4 Wed Nov  1 07:54:51 1995  Doug Lea  (dl at gee)
+      * Added malloc_trim, with help from Wolfram Gloger
+	(wmglo@Dent.MED.Uni-Muenchen.DE).
+
+    V2.5.3 Tue Apr 26 10:16:01 1994  Doug Lea  (dl at g)
+
+    V2.5.2 Tue Apr  5 16:20:40 1994  Doug Lea  (dl at g)
+      * realloc: try to expand in both directions
+      * malloc: swap order of clean-bin strategy;
+      * realloc: only conditionally expand backwards
+      * Try not to scavenge used bins
+      * Use bin counts as a guide to preallocation
+      * Occasionally bin return list chunks in first scan
+      * Add a few optimizations from colin@nyx10.cs.du.edu
+
+    V2.5.1 Sat Aug 14 15:40:43 1993  Doug Lea  (dl at g)
+      * faster bin computation & slightly different binning
+      * merged all consolidations to one part of malloc proper
+	 (eliminating old malloc_find_space & malloc_clean_bin)
+      * Scan 2 returns chunks (not just 1)
+      * Propagate failure in realloc if malloc returns 0
+      * Add stuff to allow compilation on non-ANSI compilers
+	  from kpv@research.att.com
+
+    V2.5 Sat Aug  7 07:41:59 1993  Doug Lea  (dl at g.oswego.edu)
+      * removed potential for odd address access in prev_chunk
+      * removed dependency on getpagesize.h
+      * misc cosmetics and a bit more internal documentation
+      * anticosmetics: mangled names in macros to evade debugger strangeness
+      * tested on sparc, hp-700, dec-mips, rs6000
+	  with gcc & native cc (hp, dec only) allowing
+	  Detlefs & Zorn comparison study (in SIGPLAN Notices.)
+
+    Trial version Fri Aug 28 13:14:29 1992  Doug Lea  (dl at g.oswego.edu)
+      * Based loosely on libg++-1.2X malloc. (It retains some of the overall
+	 structure of old version,  but most details differ.)
+
+*/
diff --git a/compat/nedmalloc/nedmalloc.c b/compat/nedmalloc/nedmalloc.c
new file mode 100644
index 000000000000..1cc31c350223
--- /dev/null
+++ b/compat/nedmalloc/nedmalloc.c
@@ -0,0 +1,953 @@
+/* Alternative malloc implementation for multiple threads without
+lock contention based on dlmalloc. (C) 2005-2006 Niall Douglas
+
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef _MSC_VER
+/* Enable full aliasing on MSVC */
+/*#pragma optimize("a", on)*/
+#endif
+
+/*#define FULLSANITYCHECKS*/
+
+#include "nedmalloc.h"
+#if defined(WIN32)
+ #include <malloc.h>
+#endif
+#define MSPACES 1
+#define ONLY_MSPACES 1
+#ifndef USE_LOCKS
+ #define USE_LOCKS 1
+#endif
+#define FOOTERS 1           /* Need to enable footers so frees lock the right mspace */
+#undef DEBUG				/* dlmalloc wants DEBUG either 0 or 1 */
+#ifdef _DEBUG
+ #define DEBUG 1
+#else
+ #define DEBUG 0
+#endif
+#ifdef NDEBUG               /* Disable assert checking on release builds */
+ #undef DEBUG
+#endif
+/* The default of 64Kb means we spend too much time kernel-side */
+#ifndef DEFAULT_GRANULARITY
+#define DEFAULT_GRANULARITY (1*1024*1024)
+#endif
+/*#define USE_SPIN_LOCKS 0*/
+
+
+/*#define FORCEINLINE*/
+#include "malloc.c.h"
+#ifdef NDEBUG               /* Disable assert checking on release builds */
+ #undef DEBUG
+#endif
+
+/* The maximum concurrent threads in a pool possible */
+#ifndef MAXTHREADSINPOOL
+#define MAXTHREADSINPOOL 16
+#endif
+/* The maximum number of threadcaches which can be allocated */
+#ifndef THREADCACHEMAXCACHES
+#define THREADCACHEMAXCACHES 256
+#endif
+/* The maximum size to be allocated from the thread cache */
+#ifndef THREADCACHEMAX
+#define THREADCACHEMAX 8192
+#endif
+#if 0
+/* The number of cache entries for finer grained bins. This is (topbitpos(THREADCACHEMAX)-4)*2 */
+#define THREADCACHEMAXBINS ((13-4)*2)
+#else
+/* The number of cache entries. This is (topbitpos(THREADCACHEMAX)-4) */
+#define THREADCACHEMAXBINS (13-4)
+#endif
+/* Point at which the free space in a thread cache is garbage collected */
+#ifndef THREADCACHEMAXFREESPACE
+#define THREADCACHEMAXFREESPACE (512*1024)
+#endif
+
+
+#ifdef WIN32
+ #define TLSVAR			DWORD
+ #define TLSALLOC(k)	(*(k)=TlsAlloc(), TLS_OUT_OF_INDEXES==*(k))
+ #define TLSFREE(k)		(!TlsFree(k))
+ #define TLSGET(k)		TlsGetValue(k)
+ #define TLSSET(k, a)	(!TlsSetValue(k, a))
+ #ifdef DEBUG
+static LPVOID ChkedTlsGetValue(DWORD idx)
+{
+	LPVOID ret=TlsGetValue(idx);
+	assert(S_OK==GetLastError());
+	return ret;
+}
+  #undef TLSGET
+  #define TLSGET(k) ChkedTlsGetValue(k)
+ #endif
+#else
+ #define TLSVAR			pthread_key_t
+ #define TLSALLOC(k)	pthread_key_create(k, 0)
+ #define TLSFREE(k)		pthread_key_delete(k)
+ #define TLSGET(k)		pthread_getspecific(k)
+ #define TLSSET(k, a)	pthread_setspecific(k, a)
+#endif
+
+#if 0
+/* Only enable if testing with valgrind. Causes misoperation */
+#define mspace_malloc(p, s) malloc(s)
+#define mspace_realloc(p, m, s) realloc(m, s)
+#define mspace_calloc(p, n, s) calloc(n, s)
+#define mspace_free(p, m) free(m)
+#endif
+
+
+#if defined(__cplusplus)
+#if !defined(NO_NED_NAMESPACE)
+namespace nedalloc {
+#else
+extern "C" {
+#endif
+#endif
+
+size_t nedblksize(void *mem) THROWSPEC
+{
+#if 0
+	/* Only enable if testing with valgrind. Causes misoperation */
+	return THREADCACHEMAX;
+#else
+	if(mem)
+	{
+		mchunkptr p=mem2chunk(mem);
+		assert(cinuse(p));	/* If this fails, someone tried to free a block twice */
+		if(cinuse(p))
+			return chunksize(p)-overhead_for(p);
+	}
+	return 0;
+#endif
+}
+
+void nedsetvalue(void *v) THROWSPEC					{ nedpsetvalue(0, v); }
+void * nedmalloc(size_t size) THROWSPEC				{ return nedpmalloc(0, size); }
+void * nedcalloc(size_t no, size_t size) THROWSPEC	{ return nedpcalloc(0, no, size); }
+void * nedrealloc(void *mem, size_t size) THROWSPEC	{ return nedprealloc(0, mem, size); }
+void   nedfree(void *mem) THROWSPEC					{ nedpfree(0, mem); }
+void * nedmemalign(size_t alignment, size_t bytes) THROWSPEC { return nedpmemalign(0, alignment, bytes); }
+#if !NO_MALLINFO
+struct mallinfo nedmallinfo(void) THROWSPEC			{ return nedpmallinfo(0); }
+#endif
+int    nedmallopt(int parno, int value) THROWSPEC	{ return nedpmallopt(0, parno, value); }
+int    nedmalloc_trim(size_t pad) THROWSPEC			{ return nedpmalloc_trim(0, pad); }
+void   nedmalloc_stats(void) THROWSPEC					{ nedpmalloc_stats(0); }
+size_t nedmalloc_footprint(void) THROWSPEC				{ return nedpmalloc_footprint(0); }
+void **nedindependent_calloc(size_t elemsno, size_t elemsize, void **chunks) THROWSPEC	{ return nedpindependent_calloc(0, elemsno, elemsize, chunks); }
+void **nedindependent_comalloc(size_t elems, size_t *sizes, void **chunks) THROWSPEC	{ return nedpindependent_comalloc(0, elems, sizes, chunks); }
+
+struct threadcacheblk_t;
+typedef struct threadcacheblk_t threadcacheblk;
+struct threadcacheblk_t
+{	/* Keep less than 16 bytes on 32 bit systems and 32 bytes on 64 bit systems */
+#ifdef FULLSANITYCHECKS
+	unsigned int magic;
+#endif
+	unsigned int lastUsed, size;
+	threadcacheblk *next, *prev;
+};
+typedef struct threadcache_t
+{
+#ifdef FULLSANITYCHECKS
+	unsigned int magic1;
+#endif
+	int mymspace;						/* Last mspace entry this thread used */
+	long threadid;
+	unsigned int mallocs, frees, successes;
+	size_t freeInCache;					/* How much free space is stored in this cache */
+	threadcacheblk *bins[(THREADCACHEMAXBINS+1)*2];
+#ifdef FULLSANITYCHECKS
+	unsigned int magic2;
+#endif
+} threadcache;
+struct nedpool_t
+{
+	MLOCK_T mutex;
+	void *uservalue;
+	int threads;						/* Max entries in m to use */
+	threadcache *caches[THREADCACHEMAXCACHES];
+	TLSVAR mycache;						/* Thread cache for this thread. 0 for unset, negative for use mspace-1 directly, otherwise is cache-1 */
+	mstate m[MAXTHREADSINPOOL+1];		/* mspace entries for this pool */
+};
+static nedpool syspool;
+
+static FORCEINLINE unsigned int size2binidx(size_t _size) THROWSPEC
+{	/* 8=1000	16=10000	20=10100	24=11000	32=100000	48=110000	4096=1000000000000 */
+	unsigned int topbit, size=(unsigned int)(_size>>4);
+	/* 16=1		20=1	24=1	32=10	48=11	64=100	96=110	128=1000	4096=100000000 */
+
+#if defined(__GNUC__)
+	topbit = sizeof(size)*__CHAR_BIT__ - 1 - __builtin_clz(size);
+#elif defined(_MSC_VER) && _MSC_VER>=1300
+	{
+	    unsigned long bsrTopBit;
+
+	    _BitScanReverse(&bsrTopBit, size);
+
+	    topbit = bsrTopBit;
+	}
+#else
+#if 0
+	union {
+		unsigned asInt[2];
+		double asDouble;
+	};
+	int n;
+
+	asDouble = (double)size + 0.5;
+	topbit = (asInt[!FOX_BIGENDIAN] >> 20) - 1023;
+#else
+	{
+		unsigned int x=size;
+		x = x | (x >> 1);
+		x = x | (x >> 2);
+		x = x | (x >> 4);
+		x = x | (x >> 8);
+		x = x | (x >>16);
+		x = ~x;
+		x = x - ((x >> 1) & 0x55555555);
+		x = (x & 0x33333333) + ((x >> 2) & 0x33333333);
+		x = (x + (x >> 4)) & 0x0F0F0F0F;
+		x = x + (x << 8);
+		x = x + (x << 16);
+		topbit=31 - (x >> 24);
+	}
+#endif
+#endif
+	return topbit;
+}
+
+
+#ifdef FULLSANITYCHECKS
+static void tcsanitycheck(threadcacheblk **ptr) THROWSPEC
+{
+	assert((ptr[0] && ptr[1]) || (!ptr[0] && !ptr[1]));
+	if(ptr[0] && ptr[1])
+	{
+		assert(nedblksize(ptr[0])>=sizeof(threadcacheblk));
+		assert(nedblksize(ptr[1])>=sizeof(threadcacheblk));
+		assert(*(unsigned int *) "NEDN"==ptr[0]->magic);
+		assert(*(unsigned int *) "NEDN"==ptr[1]->magic);
+		assert(!ptr[0]->prev);
+		assert(!ptr[1]->next);
+		if(ptr[0]==ptr[1])
+		{
+			assert(!ptr[0]->next);
+			assert(!ptr[1]->prev);
+		}
+	}
+}
+static void tcfullsanitycheck(threadcache *tc) THROWSPEC
+{
+	threadcacheblk **tcbptr=tc->bins;
+	int n;
+	for(n=0; n<=THREADCACHEMAXBINS; n++, tcbptr+=2)
+	{
+		threadcacheblk *b, *ob=0;
+		tcsanitycheck(tcbptr);
+		for(b=tcbptr[0]; b; ob=b, b=b->next)
+		{
+			assert(*(unsigned int *) "NEDN"==b->magic);
+			assert(!ob || ob->next==b);
+			assert(!ob || b->prev==ob);
+		}
+	}
+}
+#endif
+
+static NOINLINE void RemoveCacheEntries(nedpool *p, threadcache *tc, unsigned int age) THROWSPEC
+{
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	if(tc->freeInCache)
+	{
+		threadcacheblk **tcbptr=tc->bins;
+		int n;
+		for(n=0; n<=THREADCACHEMAXBINS; n++, tcbptr+=2)
+		{
+			threadcacheblk **tcb=tcbptr+1;		/* come from oldest end of list */
+			/*tcsanitycheck(tcbptr);*/
+			for(; *tcb && tc->frees-(*tcb)->lastUsed>=age; )
+			{
+				threadcacheblk *f=*tcb;
+				size_t blksize=f->size; /*nedblksize(f);*/
+				assert(blksize<=nedblksize(f));
+				assert(blksize);
+#ifdef FULLSANITYCHECKS
+				assert(*(unsigned int *) "NEDN"==(*tcb)->magic);
+#endif
+				*tcb=(*tcb)->prev;
+				if(*tcb)
+					(*tcb)->next=0;
+				else
+					*tcbptr=0;
+				tc->freeInCache-=blksize;
+				assert((long) tc->freeInCache>=0);
+				mspace_free(0, f);
+				/*tcsanitycheck(tcbptr);*/
+			}
+		}
+	}
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+}
+static void DestroyCaches(nedpool *p) THROWSPEC
+{
+	if(p->caches)
+	{
+		threadcache *tc;
+		int n;
+		for(n=0; n<THREADCACHEMAXCACHES; n++)
+		{
+			if((tc=p->caches[n]))
+			{
+				tc->frees++;
+				RemoveCacheEntries(p, tc, 0);
+				assert(!tc->freeInCache);
+				tc->mymspace=-1;
+				tc->threadid=0;
+				mspace_free(0, tc);
+				p->caches[n]=0;
+			}
+		}
+	}
+}
+
+static NOINLINE threadcache *AllocCache(nedpool *p) THROWSPEC
+{
+	threadcache *tc=0;
+	int n, end;
+	ACQUIRE_LOCK(&p->mutex);
+	for(n=0; n<THREADCACHEMAXCACHES && p->caches[n]; n++);
+	if(THREADCACHEMAXCACHES==n)
+	{	/* List exhausted, so disable for this thread */
+		RELEASE_LOCK(&p->mutex);
+		return 0;
+	}
+	tc=p->caches[n]=(threadcache *) mspace_calloc(p->m[0], 1, sizeof(threadcache));
+	if(!tc)
+	{
+		RELEASE_LOCK(&p->mutex);
+		return 0;
+	}
+#ifdef FULLSANITYCHECKS
+	tc->magic1=*(unsigned int *)"NEDMALC1";
+	tc->magic2=*(unsigned int *)"NEDMALC2";
+#endif
+	tc->threadid=(long)(size_t)CURRENT_THREAD;
+	for(end=0; p->m[end]; end++);
+	tc->mymspace=tc->threadid % end;
+	RELEASE_LOCK(&p->mutex);
+	if(TLSSET(p->mycache, (void *)(size_t)(n+1))) abort();
+	return tc;
+}
+
+static void *threadcache_malloc(nedpool *p, threadcache *tc, size_t *size) THROWSPEC
+{
+	void *ret=0;
+	unsigned int bestsize;
+	unsigned int idx=size2binidx(*size);
+	size_t blksize=0;
+	threadcacheblk *blk, **binsptr;
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	/* Calculate best fit bin size */
+	bestsize=1<<(idx+4);
+#if 0
+	/* Finer grained bin fit */
+	idx<<=1;
+	if(*size>bestsize)
+	{
+		idx++;
+		bestsize+=bestsize>>1;
+	}
+	if(*size>bestsize)
+	{
+		idx++;
+		bestsize=1<<(4+(idx>>1));
+	}
+#else
+	if(*size>bestsize)
+	{
+		idx++;
+		bestsize<<=1;
+	}
+#endif
+	assert(bestsize>=*size);
+	if(*size<bestsize) *size=bestsize;
+	assert(*size<=THREADCACHEMAX);
+	assert(idx<=THREADCACHEMAXBINS);
+	binsptr=&tc->bins[idx*2];
+	/* Try to match close, but move up a bin if necessary */
+	blk=*binsptr;
+	if(!blk || blk->size<*size)
+	{	/* Bump it up a bin */
+		if(idx<THREADCACHEMAXBINS)
+		{
+			idx++;
+			binsptr+=2;
+			blk=*binsptr;
+		}
+	}
+	if(blk)
+	{
+		blksize=blk->size; /*nedblksize(blk);*/
+		assert(nedblksize(blk)>=blksize);
+		assert(blksize>=*size);
+		if(blk->next)
+			blk->next->prev=0;
+		*binsptr=blk->next;
+		if(!*binsptr)
+			binsptr[1]=0;
+#ifdef FULLSANITYCHECKS
+		blk->magic=0;
+#endif
+		assert(binsptr[0]!=blk && binsptr[1]!=blk);
+		assert(nedblksize(blk)>=sizeof(threadcacheblk) && nedblksize(blk)<=THREADCACHEMAX+CHUNK_OVERHEAD);
+		/*printf("malloc: %p, %p, %p, %lu\n", p, tc, blk, (long) size);*/
+		ret=(void *) blk;
+	}
+	++tc->mallocs;
+	if(ret)
+	{
+		assert(blksize>=*size);
+		++tc->successes;
+		tc->freeInCache-=blksize;
+		assert((long) tc->freeInCache>=0);
+	}
+#if defined(DEBUG) && 0
+	if(!(tc->mallocs & 0xfff))
+	{
+		printf("*** threadcache=%u, mallocs=%u (%f), free=%u (%f), freeInCache=%u\n", (unsigned int) tc->threadid, tc->mallocs,
+			(float) tc->successes/tc->mallocs, tc->frees, (float) tc->successes/tc->frees, (unsigned int) tc->freeInCache);
+	}
+#endif
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	return ret;
+}
+static NOINLINE void ReleaseFreeInCache(nedpool *p, threadcache *tc, int mymspace) THROWSPEC
+{
+	unsigned int age=THREADCACHEMAXFREESPACE/8192;
+	/*ACQUIRE_LOCK(&p->m[mymspace]->mutex);*/
+	while(age && tc->freeInCache>=THREADCACHEMAXFREESPACE)
+	{
+		RemoveCacheEntries(p, tc, age);
+		/*printf("*** Removing cache entries older than %u (%u)\n", age, (unsigned int) tc->freeInCache);*/
+		age>>=1;
+	}
+	/*RELEASE_LOCK(&p->m[mymspace]->mutex);*/
+}
+static void threadcache_free(nedpool *p, threadcache *tc, int mymspace, void *mem, size_t size) THROWSPEC
+{
+	unsigned int bestsize;
+	unsigned int idx=size2binidx(size);
+	threadcacheblk **binsptr, *tck=(threadcacheblk *) mem;
+	assert(size>=sizeof(threadcacheblk) && size<=THREADCACHEMAX+CHUNK_OVERHEAD);
+#ifdef DEBUG
+	{	/* Make sure this is a valid memory block */
+	    mchunkptr p  = mem2chunk(mem);
+	    mstate fm = get_mstate_for(p);
+	    if (!ok_magic(fm)) {
+	      USAGE_ERROR_ACTION(fm, p);
+	      return;
+	    }
+	}
+#endif
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+	/* Calculate best fit bin size */
+	bestsize=1<<(idx+4);
+#if 0
+	/* Finer grained bin fit */
+	idx<<=1;
+	if(size>bestsize)
+	{
+		unsigned int biggerbestsize=bestsize+bestsize<<1;
+		if(size>=biggerbestsize)
+		{
+			idx++;
+			bestsize=biggerbestsize;
+		}
+	}
+#endif
+	if(bestsize!=size)	/* dlmalloc can round up, so we round down to preserve indexing */
+		size=bestsize;
+	binsptr=&tc->bins[idx*2];
+	assert(idx<=THREADCACHEMAXBINS);
+	if(tck==*binsptr)
+	{
+		fprintf(stderr, "Attempt to free already freed memory block %p - aborting!\n", tck);
+		abort();
+	}
+#ifdef FULLSANITYCHECKS
+	tck->magic=*(unsigned int *) "NEDN";
+#endif
+	tck->lastUsed=++tc->frees;
+	tck->size=(unsigned int) size;
+	tck->next=*binsptr;
+	tck->prev=0;
+	if(tck->next)
+		tck->next->prev=tck;
+	else
+		binsptr[1]=tck;
+	assert(!*binsptr || (*binsptr)->size==tck->size);
+	*binsptr=tck;
+	assert(tck==tc->bins[idx*2]);
+	assert(tc->bins[idx*2+1]==tck || binsptr[0]->next->prev==tck);
+	/*printf("free: %p, %p, %p, %lu\n", p, tc, mem, (long) size);*/
+	tc->freeInCache+=size;
+#ifdef FULLSANITYCHECKS
+	tcfullsanitycheck(tc);
+#endif
+#if 1
+	if(tc->freeInCache>=THREADCACHEMAXFREESPACE)
+		ReleaseFreeInCache(p, tc, mymspace);
+#endif
+}
+
+
+
+
+static NOINLINE int InitPool(nedpool *p, size_t capacity, int threads) THROWSPEC
+{	/* threads is -1 for system pool */
+	ensure_initialization();
+	ACQUIRE_MALLOC_GLOBAL_LOCK();
+	if(p->threads) goto done;
+	if(INITIAL_LOCK(&p->mutex)) goto err;
+	if(TLSALLOC(&p->mycache)) goto err;
+	if(!(p->m[0]=(mstate) create_mspace(capacity, 1))) goto err;
+	p->m[0]->extp=p;
+	p->threads=(threads<1 || threads>MAXTHREADSINPOOL) ? MAXTHREADSINPOOL : threads;
+done:
+	RELEASE_MALLOC_GLOBAL_LOCK();
+	return 1;
+err:
+	if(threads<0)
+		abort();			/* If you can't allocate for system pool, we're screwed */
+	DestroyCaches(p);
+	if(p->m[0])
+	{
+		destroy_mspace(p->m[0]);
+		p->m[0]=0;
+	}
+	if(p->mycache)
+	{
+		if(TLSFREE(p->mycache)) abort();
+		p->mycache=0;
+	}
+	RELEASE_MALLOC_GLOBAL_LOCK();
+	return 0;
+}
+static NOINLINE mstate FindMSpace(nedpool *p, threadcache *tc, int *lastUsed, size_t size) THROWSPEC
+{	/* Gets called when thread's last used mspace is in use. The strategy
+	is to run through the list of all available mspaces looking for an
+	unlocked one and if we fail, we create a new one so long as we don't
+	exceed p->threads */
+	int n, end;
+	for(n=end=*lastUsed+1; p->m[n]; end=++n)
+	{
+		if(TRY_LOCK(&p->m[n]->mutex)) goto found;
+	}
+	for(n=0; n<*lastUsed && p->m[n]; n++)
+	{
+		if(TRY_LOCK(&p->m[n]->mutex)) goto found;
+	}
+	if(end<p->threads)
+	{
+		mstate temp;
+		if(!(temp=(mstate) create_mspace(size, 1)))
+			goto badexit;
+		/* Now we're ready to modify the lists, we lock */
+		ACQUIRE_LOCK(&p->mutex);
+		while(p->m[end] && end<p->threads)
+			end++;
+		if(end>=p->threads)
+		{	/* Drat, must destroy it now */
+			RELEASE_LOCK(&p->mutex);
+			destroy_mspace((mspace) temp);
+			goto badexit;
+		}
+		/* We really want to make sure this goes into memory now but we
+		have to be careful of breaking aliasing rules, so write it twice */
+		{
+			volatile struct malloc_state **_m=(volatile struct malloc_state **) &p->m[end];
+			*_m=(p->m[end]=temp);
+		}
+		ACQUIRE_LOCK(&p->m[end]->mutex);
+		/*printf("Created mspace idx %d\n", end);*/
+		RELEASE_LOCK(&p->mutex);
+		n=end;
+		goto found;
+	}
+	/* Let it lock on the last one it used */
+badexit:
+	ACQUIRE_LOCK(&p->m[*lastUsed]->mutex);
+	return p->m[*lastUsed];
+found:
+	*lastUsed=n;
+	if(tc)
+		tc->mymspace=n;
+	else
+	{
+		if(TLSSET(p->mycache, (void *)(size_t)(-(n+1)))) abort();
+	}
+	return p->m[n];
+}
+
+nedpool *nedcreatepool(size_t capacity, int threads) THROWSPEC
+{
+	nedpool *ret;
+	if(!(ret=(nedpool *) nedpcalloc(0, 1, sizeof(nedpool)))) return 0;
+	if(!InitPool(ret, capacity, threads))
+	{
+		nedpfree(0, ret);
+		return 0;
+	}
+	return ret;
+}
+void neddestroypool(nedpool *p) THROWSPEC
+{
+	int n;
+	ACQUIRE_LOCK(&p->mutex);
+	DestroyCaches(p);
+	for(n=0; p->m[n]; n++)
+	{
+		destroy_mspace(p->m[n]);
+		p->m[n]=0;
+	}
+	RELEASE_LOCK(&p->mutex);
+	if(TLSFREE(p->mycache)) abort();
+	nedpfree(0, p);
+}
+
+void nedpsetvalue(nedpool *p, void *v) THROWSPEC
+{
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	p->uservalue=v;
+}
+void *nedgetvalue(nedpool **p, void *mem) THROWSPEC
+{
+	nedpool *np=0;
+	mchunkptr mcp=mem2chunk(mem);
+	mstate fm;
+	if(!(is_aligned(chunk2mem(mcp))) && mcp->head != FENCEPOST_HEAD) return 0;
+	if(!cinuse(mcp)) return 0;
+	if(!next_pinuse(mcp)) return 0;
+	if(!is_mmapped(mcp) && !pinuse(mcp))
+	{
+		if(next_chunk(prev_chunk(mcp))!=mcp) return 0;
+	}
+	fm=get_mstate_for(mcp);
+	if(!ok_magic(fm)) return 0;
+	if(!ok_address(fm, mcp)) return 0;
+	if(!fm->extp) return 0;
+	np=(nedpool *) fm->extp;
+	if(p) *p=np;
+	return np->uservalue;
+}
+
+void neddisablethreadcache(nedpool *p) THROWSPEC
+{
+	int mycache;
+	if(!p)
+	{
+		p=&syspool;
+		if(!syspool.threads) InitPool(&syspool, 0, -1);
+	}
+	mycache=(int)(size_t) TLSGET(p->mycache);
+	if(!mycache)
+	{	/* Set to mspace 0 */
+		if(TLSSET(p->mycache, (void *)-1)) abort();
+	}
+	else if(mycache>0)
+	{	/* Set to last used mspace */
+		threadcache *tc=p->caches[mycache-1];
+#if defined(DEBUG)
+		printf("Threadcache utilisation: %lf%% in cache with %lf%% lost to other threads\n",
+			100.0*tc->successes/tc->mallocs, 100.0*((double) tc->mallocs-tc->frees)/tc->mallocs);
+#endif
+		if(TLSSET(p->mycache, (void *)(size_t)(-tc->mymspace))) abort();
+		tc->frees++;
+		RemoveCacheEntries(p, tc, 0);
+		assert(!tc->freeInCache);
+		tc->mymspace=-1;
+		tc->threadid=0;
+		mspace_free(0, p->caches[mycache-1]);
+		p->caches[mycache-1]=0;
+	}
+}
+
+#define GETMSPACE(m,p,tc,ms,s,action)           \
+  do                                            \
+  {                                             \
+    mstate m = GetMSpace((p),(tc),(ms),(s));    \
+    action;                                     \
+    RELEASE_LOCK(&m->mutex);                    \
+  } while (0)
+
+static FORCEINLINE mstate GetMSpace(nedpool *p, threadcache *tc, int mymspace, size_t size) THROWSPEC
+{	/* Returns a locked and ready for use mspace */
+	mstate m=p->m[mymspace];
+	assert(m);
+	if(!TRY_LOCK(&p->m[mymspace]->mutex)) m=FindMSpace(p, tc, &mymspace, size);\
+	/*assert(IS_LOCKED(&p->m[mymspace]->mutex));*/
+	return m;
+}
+static FORCEINLINE void GetThreadCache(nedpool **p, threadcache **tc, int *mymspace, size_t *size) THROWSPEC
+{
+	int mycache;
+	if(size && *size<sizeof(threadcacheblk)) *size=sizeof(threadcacheblk);
+	if(!*p)
+	{
+		*p=&syspool;
+		if(!syspool.threads) InitPool(&syspool, 0, -1);
+	}
+	mycache=(int)(size_t) TLSGET((*p)->mycache);
+	if(mycache>0)
+	{
+		*tc=(*p)->caches[mycache-1];
+		*mymspace=(*tc)->mymspace;
+	}
+	else if(!mycache)
+	{
+		*tc=AllocCache(*p);
+		if(!*tc)
+		{	/* Disable */
+			if(TLSSET((*p)->mycache, (void *)-1)) abort();
+			*mymspace=0;
+		}
+		else
+			*mymspace=(*tc)->mymspace;
+	}
+	else
+	{
+		*tc=0;
+		*mymspace=-mycache-1;
+	}
+	assert(*mymspace>=0);
+	assert((long)(size_t)CURRENT_THREAD==(*tc)->threadid);
+#ifdef FULLSANITYCHECKS
+	if(*tc)
+	{
+		if(*(unsigned int *)"NEDMALC1"!=(*tc)->magic1 || *(unsigned int *)"NEDMALC2"!=(*tc)->magic2)
+		{
+			abort();
+		}
+	}
+#endif
+}
+
+void * nedpmalloc(nedpool *p, size_t size) THROWSPEC
+{
+	void *ret=0;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &size);
+#if THREADCACHEMAX
+	if(tc && size<=THREADCACHEMAX)
+	{	/* Use the thread cache */
+		ret=threadcache_malloc(p, tc, &size);
+	}
+#endif
+	if(!ret)
+	{	/* Use this thread's mspace */
+	GETMSPACE(m, p, tc, mymspace, size,
+		  ret=mspace_malloc(m, size));
+	}
+	return ret;
+}
+void * nedpcalloc(nedpool *p, size_t no, size_t size) THROWSPEC
+{
+	size_t rsize=size*no;
+	void *ret=0;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &rsize);
+#if THREADCACHEMAX
+	if(tc && rsize<=THREADCACHEMAX)
+	{	/* Use the thread cache */
+		if((ret=threadcache_malloc(p, tc, &rsize)))
+			memset(ret, 0, rsize);
+	}
+#endif
+	if(!ret)
+	{	/* Use this thread's mspace */
+	GETMSPACE(m, p, tc, mymspace, rsize,
+		  ret=mspace_calloc(m, 1, rsize));
+	}
+	return ret;
+}
+void * nedprealloc(nedpool *p, void *mem, size_t size) THROWSPEC
+{
+	void *ret=0;
+	threadcache *tc;
+	int mymspace;
+	if(!mem) return nedpmalloc(p, size);
+	GetThreadCache(&p, &tc, &mymspace, &size);
+#if THREADCACHEMAX
+	if(tc && size && size<=THREADCACHEMAX)
+	{	/* Use the thread cache */
+		size_t memsize=nedblksize(mem);
+		assert(memsize);
+		if((ret=threadcache_malloc(p, tc, &size)))
+		{
+			memcpy(ret, mem, memsize<size ? memsize : size);
+			if(memsize<=THREADCACHEMAX)
+				threadcache_free(p, tc, mymspace, mem, memsize);
+			else
+				mspace_free(0, mem);
+		}
+	}
+#endif
+	if(!ret)
+	{	/* Reallocs always happen in the mspace they happened in, so skip
+		locking the preferred mspace for this thread */
+		ret=mspace_realloc(0, mem, size);
+	}
+	return ret;
+}
+void   nedpfree(nedpool *p, void *mem) THROWSPEC
+{	/* Frees always happen in the mspace they happened in, so skip
+	locking the preferred mspace for this thread */
+	threadcache *tc;
+	int mymspace;
+	size_t memsize;
+	assert(mem);
+	GetThreadCache(&p, &tc, &mymspace, 0);
+#if THREADCACHEMAX
+	memsize=nedblksize(mem);
+	assert(memsize);
+	if(mem && tc && memsize<=(THREADCACHEMAX+CHUNK_OVERHEAD))
+		threadcache_free(p, tc, mymspace, mem, memsize);
+	else
+#endif
+		mspace_free(0, mem);
+}
+void * nedpmemalign(nedpool *p, size_t alignment, size_t bytes) THROWSPEC
+{
+	void *ret;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &bytes);
+	{	/* Use this thread's mspace */
+	GETMSPACE(m, p, tc, mymspace, bytes,
+		  ret=mspace_memalign(m, alignment, bytes));
+	}
+	return ret;
+}
+#if !NO_MALLINFO
+struct mallinfo nedpmallinfo(nedpool *p) THROWSPEC
+{
+	int n;
+	struct mallinfo ret={0};
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		struct mallinfo t=mspace_mallinfo(p->m[n]);
+		ret.arena+=t.arena;
+		ret.ordblks+=t.ordblks;
+		ret.hblkhd+=t.hblkhd;
+		ret.usmblks+=t.usmblks;
+		ret.uordblks+=t.uordblks;
+		ret.fordblks+=t.fordblks;
+		ret.keepcost+=t.keepcost;
+	}
+	return ret;
+}
+#endif
+int    nedpmallopt(nedpool *p, int parno, int value) THROWSPEC
+{
+	return mspace_mallopt(parno, value);
+}
+int    nedpmalloc_trim(nedpool *p, size_t pad) THROWSPEC
+{
+	int n, ret=0;
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		ret+=mspace_trim(p->m[n], pad);
+	}
+	return ret;
+}
+void   nedpmalloc_stats(nedpool *p) THROWSPEC
+{
+	int n;
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		mspace_malloc_stats(p->m[n]);
+	}
+}
+size_t nedpmalloc_footprint(nedpool *p) THROWSPEC
+{
+	size_t ret=0;
+	int n;
+	if(!p) { p=&syspool; if(!syspool.threads) InitPool(&syspool, 0, -1); }
+	for(n=0; p->m[n]; n++)
+	{
+		ret+=mspace_footprint(p->m[n]);
+	}
+	return ret;
+}
+void **nedpindependent_calloc(nedpool *p, size_t elemsno, size_t elemsize, void **chunks) THROWSPEC
+{
+	void **ret;
+	threadcache *tc;
+	int mymspace;
+	GetThreadCache(&p, &tc, &mymspace, &elemsize);
+    GETMSPACE(m, p, tc, mymspace, elemsno*elemsize,
+	      ret=mspace_independent_calloc(m, elemsno, elemsize, chunks));
+	return ret;
+}
+void **nedpindependent_comalloc(nedpool *p, size_t elems, size_t *sizes, void **chunks) THROWSPEC
+{
+	void **ret;
+	threadcache *tc;
+	int mymspace;
+	size_t i, *adjustedsizes=(size_t *) alloca(elems*sizeof(size_t));
+	if(!adjustedsizes) return 0;
+	for(i=0; i<elems; i++)
+		adjustedsizes[i]=sizes[i]<sizeof(threadcacheblk) ? sizeof(threadcacheblk) : sizes[i];
+	GetThreadCache(&p, &tc, &mymspace, 0);
+	GETMSPACE(m, p, tc, mymspace, 0,
+	      ret=mspace_independent_comalloc(m, elems, adjustedsizes, chunks));
+	return ret;
+}
+
+#if defined(__cplusplus)
+}
+#endif
diff --git a/compat/nedmalloc/nedmalloc.h b/compat/nedmalloc/nedmalloc.h
new file mode 100644
index 000000000000..f960e66063a7
--- /dev/null
+++ b/compat/nedmalloc/nedmalloc.h
@@ -0,0 +1,180 @@
+/* nedalloc, an alternative malloc implementation for multiple threads without
+lock contention based on dlmalloc v2.8.3. (C) 2005 Niall Douglas
+
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef NEDMALLOC_H
+#define NEDMALLOC_H
+
+
+/* See malloc.c.h for what each function does.
+
+REPLACE_SYSTEM_ALLOCATOR causes nedalloc's functions to be called malloc,
+free etc. instead of nedmalloc, nedfree etc. You may or may not want this.
+
+NO_NED_NAMESPACE prevents the functions from being defined in the nedalloc
+namespace when in C++ (uses the global namespace instead).
+
+EXTSPEC can be defined to be __declspec(dllexport) or
+__attribute__ ((visibility("default"))) or whatever you like. It defaults
+to extern.
+
+USE_LOCKS can be 2 if you want to define your own MLOCK_T, INITIAL_LOCK,
+ACQUIRE_LOCK, RELEASE_LOCK, TRY_LOCK, IS_LOCKED and NULL_LOCK_INITIALIZER.
+
+*/
+
+#include <stddef.h>   /* for size_t */
+
+#ifndef EXTSPEC
+ #define EXTSPEC extern
+#endif
+
+#if defined(_MSC_VER) && _MSC_VER>=1400
+ #define MALLOCATTR __declspec(restrict)
+#endif
+#ifdef __GNUC__
+ #define MALLOCATTR __attribute__ ((malloc))
+#endif
+#ifndef MALLOCATTR
+ #define MALLOCATTR
+#endif
+
+#ifdef REPLACE_SYSTEM_ALLOCATOR
+ #define nedmalloc               malloc
+ #define nedcalloc               calloc
+ #define nedrealloc              realloc
+ #define nedfree                 free
+ #define nedmemalign             memalign
+ #define nedmallinfo             mallinfo
+ #define nedmallopt              mallopt
+ #define nedmalloc_trim          malloc_trim
+ #define nedmalloc_stats         malloc_stats
+ #define nedmalloc_footprint     malloc_footprint
+ #define nedindependent_calloc   independent_calloc
+ #define nedindependent_comalloc independent_comalloc
+ #ifdef _MSC_VER
+  #define nedblksize              _msize
+ #endif
+#endif
+
+#ifndef NO_MALLINFO
+#define NO_MALLINFO 0
+#endif
+
+#if !NO_MALLINFO
+struct mallinfo;
+#endif
+
+#if defined(__cplusplus)
+ #if !defined(NO_NED_NAMESPACE)
+namespace nedalloc {
+ #else
+extern "C" {
+ #endif
+ #define THROWSPEC throw()
+#else
+ #define THROWSPEC
+#endif
+
+/* These are the global functions */
+
+/* Gets the usable size of an allocated block. Note this will always be bigger than what was
+asked for due to rounding etc.
+*/
+EXTSPEC size_t nedblksize(void *mem) THROWSPEC;
+
+EXTSPEC void nedsetvalue(void *v) THROWSPEC;
+
+EXTSPEC MALLOCATTR void * nedmalloc(size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedcalloc(size_t no, size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedrealloc(void *mem, size_t size) THROWSPEC;
+EXTSPEC void   nedfree(void *mem) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedmemalign(size_t alignment, size_t bytes) THROWSPEC;
+#if !NO_MALLINFO
+EXTSPEC struct mallinfo nedmallinfo(void) THROWSPEC;
+#endif
+EXTSPEC int    nedmallopt(int parno, int value) THROWSPEC;
+EXTSPEC int    nedmalloc_trim(size_t pad) THROWSPEC;
+EXTSPEC void   nedmalloc_stats(void) THROWSPEC;
+EXTSPEC size_t nedmalloc_footprint(void) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedindependent_calloc(size_t elemsno, size_t elemsize, void **chunks) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedindependent_comalloc(size_t elems, size_t *sizes, void **chunks) THROWSPEC;
+
+/* These are the pool functions */
+struct nedpool_t;
+typedef struct nedpool_t nedpool;
+
+/* Creates a memory pool for use with the nedp* functions below.
+Capacity is how much to allocate immediately (if you know you'll be allocating a lot
+of memory very soon) which you can leave at zero. Threads specifies how many threads
+will *normally* be accessing the pool concurrently. Setting this to zero means it
+extends on demand, but be careful of this as it can rapidly consume system resources
+where bursts of concurrent threads use a pool at once.
+*/
+EXTSPEC MALLOCATTR nedpool *nedcreatepool(size_t capacity, int threads) THROWSPEC;
+
+/* Destroys a memory pool previously created by nedcreatepool().
+*/
+EXTSPEC void neddestroypool(nedpool *p) THROWSPEC;
+
+/* Sets a value to be associated with a pool. You can retrieve this value by passing
+any memory block allocated from that pool.
+*/
+EXTSPEC void nedpsetvalue(nedpool *p, void *v) THROWSPEC;
+/* Gets a previously set value using nedpsetvalue() or zero if memory is unknown.
+Optionally can also retrieve pool.
+*/
+EXTSPEC void *nedgetvalue(nedpool **p, void *mem) THROWSPEC;
+
+/* Disables the thread cache for the calling thread, returning any existing cache
+data to the central pool.
+*/
+EXTSPEC void neddisablethreadcache(nedpool *p) THROWSPEC;
+
+EXTSPEC MALLOCATTR void * nedpmalloc(nedpool *p, size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedpcalloc(nedpool *p, size_t no, size_t size) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedprealloc(nedpool *p, void *mem, size_t size) THROWSPEC;
+EXTSPEC void   nedpfree(nedpool *p, void *mem) THROWSPEC;
+EXTSPEC MALLOCATTR void * nedpmemalign(nedpool *p, size_t alignment, size_t bytes) THROWSPEC;
+#if !NO_MALLINFO
+EXTSPEC struct mallinfo nedpmallinfo(nedpool *p) THROWSPEC;
+#endif
+EXTSPEC int    nedpmallopt(nedpool *p, int parno, int value) THROWSPEC;
+EXTSPEC int    nedpmalloc_trim(nedpool *p, size_t pad) THROWSPEC;
+EXTSPEC void   nedpmalloc_stats(nedpool *p) THROWSPEC;
+EXTSPEC size_t nedpmalloc_footprint(nedpool *p) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedpindependent_calloc(nedpool *p, size_t elemsno, size_t elemsize, void **chunks) THROWSPEC;
+EXTSPEC MALLOCATTR void **nedpindependent_comalloc(nedpool *p, size_t elems, size_t *sizes, void **chunks) THROWSPEC;
+
+#if defined(__cplusplus)
+}
+#endif
+
+#undef MALLOCATTR
+#undef EXTSPEC
+
+#endif
diff --git a/compat/obstack.c b/compat/obstack.c
new file mode 100644
index 000000000000..27cd5c1ea1f9
--- /dev/null
+++ b/compat/obstack.c
@@ -0,0 +1,413 @@
+/* obstack.c - subroutines used implicitly by object stack macros
+   Copyright (C) 1988, 1989, 1990, 1991, 1992, 1993, 1994, 1996, 1997, 1998,
+   1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include "git-compat-util.h"
+#include <gettext.h>
+#include "obstack.h"
+
+/* NOTE BEFORE MODIFYING THIS FILE: This version number must be
+   incremented whenever callers compiled using an old obstack.h can no
+   longer properly call the functions in this obstack.c.  */
+#define OBSTACK_INTERFACE_VERSION 1
+
+/* Comment out all this code if we are using the GNU C Library, and are not
+   actually compiling the library itself, and the installed library
+   supports the same library interface we do.  This code is part of the GNU
+   C Library, but also included in many other GNU distributions.  Compiling
+   and linking in this code is a waste when using the GNU C library
+   (especially if it is a shared library).  Rather than having every GNU
+   program understand `configure --with-gnu-libc' and omit the object
+   files, it is simpler to just do this in the source for each such file.  */
+
+#include <stdio.h>		/* Random thing to get __GNU_LIBRARY__.  */
+#if !defined _LIBC && defined __GNU_LIBRARY__ && __GNU_LIBRARY__ > 1
+# include <gnu-versions.h>
+# if _GNU_OBSTACK_INTERFACE_VERSION == OBSTACK_INTERFACE_VERSION
+#  define ELIDE_CODE
+# endif
+#endif
+
+#include <stddef.h>
+
+#ifndef ELIDE_CODE
+
+
+# if HAVE_INTTYPES_H
+#  include <inttypes.h>
+# endif
+# if HAVE_STDINT_H || defined _LIBC
+#  include <stdint.h>
+# endif
+
+/* Determine default alignment.  */
+union fooround
+{
+  uintmax_t i;
+  long double d;
+  void *p;
+};
+struct fooalign
+{
+  char c;
+  union fooround u;
+};
+/* If malloc were really smart, it would round addresses to DEFAULT_ALIGNMENT.
+   But in fact it might be less smart and round addresses to as much as
+   DEFAULT_ROUNDING.  So we prepare for it to do that.  */
+enum
+  {
+    DEFAULT_ALIGNMENT = offsetof (struct fooalign, u),
+    DEFAULT_ROUNDING = sizeof (union fooround)
+  };
+
+/* When we copy a long block of data, this is the unit to do it with.
+   On some machines, copying successive ints does not work;
+   in such a case, redefine COPYING_UNIT to `long' (if that works)
+   or `char' as a last resort.  */
+# ifndef COPYING_UNIT
+#  define COPYING_UNIT int
+# endif
+
+
+/* The functions allocating more room by calling `obstack_chunk_alloc'
+   jump to the handler pointed to by `obstack_alloc_failed_handler'.
+   This can be set to a user defined function which should either
+   abort gracefully or use longjump - but shouldn't return.  This
+   variable by default points to the internal function
+   `print_and_abort'.  */
+static void print_and_abort (void);
+void (*obstack_alloc_failed_handler) (void) = print_and_abort;
+
+# ifdef _LIBC
+#  if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
+/* A looong time ago (before 1994, anyway; we're not sure) this global variable
+   was used by non-GNU-C macros to avoid multiple evaluation.  The GNU C
+   library still exports it because somebody might use it.  */
+struct obstack *_obstack_compat;
+compat_symbol (libc, _obstack_compat, _obstack, GLIBC_2_0);
+#  endif
+# endif
+
+/* Define a macro that either calls functions with the traditional malloc/free
+   calling interface, or calls functions with the mmalloc/mfree interface
+   (that adds an extra first argument), based on the state of use_extra_arg.
+   For free, do not use ?:, since some compilers, like the MIPS compilers,
+   do not allow (expr) ? void : void.  */
+
+# define CALL_CHUNKFUN(h, size) \
+  (((h) -> use_extra_arg) \
+   ? (*(h)->chunkfun.extra) ((h)->extra_arg, (size)) \
+   : (*(h)->chunkfun.plain) ((size)))
+
+# define CALL_FREEFUN(h, old_chunk) \
+  do { \
+    if ((h) -> use_extra_arg) \
+      (*(h)->freefun.extra) ((h)->extra_arg, (old_chunk)); \
+    else \
+      (*(h)->freefun.plain) ((old_chunk)); \
+  } while (0)
+
+
+/* Initialize an obstack H for use.  Specify chunk size SIZE (0 means default).
+   Objects start on multiples of ALIGNMENT (0 means use default).
+   CHUNKFUN is the function to use to allocate chunks,
+   and FREEFUN the function to free them.
+
+   Return nonzero if successful, calls obstack_alloc_failed_handler if
+   allocation fails.  */
+
+int
+_obstack_begin (struct obstack *h,
+		int size, int alignment,
+		void *(*chunkfun) (long),
+		void (*freefun) (void *))
+{
+  register struct _obstack_chunk *chunk; /* points to new chunk */
+
+  if (alignment == 0)
+    alignment = DEFAULT_ALIGNMENT;
+  if (size == 0)
+    /* Default size is what GNU malloc can fit in a 4096-byte block.  */
+    {
+      /* 12 is sizeof (mhead) and 4 is EXTRA from GNU malloc.
+	 Use the values for range checking, because if range checking is off,
+	 the extra bytes won't be missed terribly, but if range checking is on
+	 and we used a larger request, a whole extra 4096 bytes would be
+	 allocated.
+
+	 These number are irrelevant to the new GNU malloc.  I suspect it is
+	 less sensitive to the size of the request.  */
+      int extra = ((((12 + DEFAULT_ROUNDING - 1) & ~(DEFAULT_ROUNDING - 1))
+		    + 4 + DEFAULT_ROUNDING - 1)
+		   & ~(DEFAULT_ROUNDING - 1));
+      size = 4096 - extra;
+    }
+
+  h->chunkfun.plain = chunkfun;
+  h->freefun.plain = freefun;
+  h->chunk_size = size;
+  h->alignment_mask = alignment - 1;
+  h->use_extra_arg = 0;
+
+  chunk = h->chunk = CALL_CHUNKFUN (h, h -> chunk_size);
+  if (!chunk)
+    (*obstack_alloc_failed_handler) ();
+  h->next_free = h->object_base = __PTR_ALIGN ((char *) chunk, chunk->contents,
+					       alignment - 1);
+  h->chunk_limit = chunk->limit
+    = (char *) chunk + h->chunk_size;
+  chunk->prev = NULL;
+  /* The initial chunk now contains no empty object.  */
+  h->maybe_empty_object = 0;
+  h->alloc_failed = 0;
+  return 1;
+}
+
+int
+_obstack_begin_1 (struct obstack *h, int size, int alignment,
+		  void *(*chunkfun) (void *, long),
+		  void (*freefun) (void *, void *),
+		  void *arg)
+{
+  register struct _obstack_chunk *chunk; /* points to new chunk */
+
+  if (alignment == 0)
+    alignment = DEFAULT_ALIGNMENT;
+  if (size == 0)
+    /* Default size is what GNU malloc can fit in a 4096-byte block.  */
+    {
+      /* 12 is sizeof (mhead) and 4 is EXTRA from GNU malloc.
+	 Use the values for range checking, because if range checking is off,
+	 the extra bytes won't be missed terribly, but if range checking is on
+	 and we used a larger request, a whole extra 4096 bytes would be
+	 allocated.
+
+	 These number are irrelevant to the new GNU malloc.  I suspect it is
+	 less sensitive to the size of the request.  */
+      int extra = ((((12 + DEFAULT_ROUNDING - 1) & ~(DEFAULT_ROUNDING - 1))
+		    + 4 + DEFAULT_ROUNDING - 1)
+		   & ~(DEFAULT_ROUNDING - 1));
+      size = 4096 - extra;
+    }
+
+  h->chunkfun.extra = (struct _obstack_chunk * (*)(void *,long)) chunkfun;
+  h->freefun.extra = (void (*) (void *, struct _obstack_chunk *)) freefun;
+
+  h->chunk_size = size;
+  h->alignment_mask = alignment - 1;
+  h->extra_arg = arg;
+  h->use_extra_arg = 1;
+
+  chunk = h->chunk = CALL_CHUNKFUN (h, h -> chunk_size);
+  if (!chunk)
+    (*obstack_alloc_failed_handler) ();
+  h->next_free = h->object_base = __PTR_ALIGN ((char *) chunk, chunk->contents,
+					       alignment - 1);
+  h->chunk_limit = chunk->limit
+    = (char *) chunk + h->chunk_size;
+  chunk->prev = NULL;
+  /* The initial chunk now contains no empty object.  */
+  h->maybe_empty_object = 0;
+  h->alloc_failed = 0;
+  return 1;
+}
+
+/* Allocate a new current chunk for the obstack *H
+   on the assumption that LENGTH bytes need to be added
+   to the current object, or a new object of length LENGTH allocated.
+   Copies any partial object from the end of the old chunk
+   to the beginning of the new one.  */
+
+void
+_obstack_newchunk (struct obstack *h, int length)
+{
+  register struct _obstack_chunk *old_chunk = h->chunk;
+  register struct _obstack_chunk *new_chunk;
+  register long	new_size;
+  register long obj_size = h->next_free - h->object_base;
+  register long i;
+  long already;
+  char *object_base;
+
+  /* Compute size for new chunk.  */
+  new_size = (obj_size + length) + (obj_size >> 3) + h->alignment_mask + 100;
+  if (new_size < h->chunk_size)
+    new_size = h->chunk_size;
+
+  /* Allocate and initialize the new chunk.  */
+  new_chunk = CALL_CHUNKFUN (h, new_size);
+  if (!new_chunk)
+    (*obstack_alloc_failed_handler) ();
+  h->chunk = new_chunk;
+  new_chunk->prev = old_chunk;
+  new_chunk->limit = h->chunk_limit = (char *) new_chunk + new_size;
+
+  /* Compute an aligned object_base in the new chunk */
+  object_base =
+    __PTR_ALIGN ((char *) new_chunk, new_chunk->contents, h->alignment_mask);
+
+  /* Move the existing object to the new chunk.
+     Word at a time is fast and is safe if the object
+     is sufficiently aligned.  */
+  if (h->alignment_mask + 1 >= DEFAULT_ALIGNMENT)
+    {
+      for (i = obj_size / sizeof (COPYING_UNIT) - 1;
+	   i >= 0; i--)
+	((COPYING_UNIT *)object_base)[i]
+	  = ((COPYING_UNIT *)h->object_base)[i];
+      /* We used to copy the odd few remaining bytes as one extra COPYING_UNIT,
+	 but that can cross a page boundary on a machine
+	 which does not do strict alignment for COPYING_UNITS.  */
+      already = obj_size / sizeof (COPYING_UNIT) * sizeof (COPYING_UNIT);
+    }
+  else
+    already = 0;
+  /* Copy remaining bytes one by one.  */
+  for (i = already; i < obj_size; i++)
+    object_base[i] = h->object_base[i];
+
+  /* If the object just copied was the only data in OLD_CHUNK,
+     free that chunk and remove it from the chain.
+     But not if that chunk might contain an empty object.  */
+  if (! h->maybe_empty_object
+      && (h->object_base
+	  == __PTR_ALIGN ((char *) old_chunk, old_chunk->contents,
+			  h->alignment_mask)))
+    {
+      new_chunk->prev = old_chunk->prev;
+      CALL_FREEFUN (h, old_chunk);
+    }
+
+  h->object_base = object_base;
+  h->next_free = h->object_base + obj_size;
+  /* The new chunk certainly contains no empty object yet.  */
+  h->maybe_empty_object = 0;
+}
+# ifdef _LIBC
+libc_hidden_def (_obstack_newchunk)
+# endif
+
+/* Return nonzero if object OBJ has been allocated from obstack H.
+   This is here for debugging.
+   If you use it in a program, you are probably losing.  */
+
+/* Suppress -Wmissing-prototypes warning.  We don't want to declare this in
+   obstack.h because it is just for debugging.  */
+int _obstack_allocated_p (struct obstack *h, void *obj);
+
+int
+_obstack_allocated_p (struct obstack *h, void *obj)
+{
+  register struct _obstack_chunk *lp;	/* below addr of any objects in this chunk */
+  register struct _obstack_chunk *plp;	/* point to previous chunk if any */
+
+  lp = (h)->chunk;
+  /* We use >= rather than > since the object cannot be exactly at
+     the beginning of the chunk but might be an empty object exactly
+     at the end of an adjacent chunk.  */
+  while (lp != NULL && ((void *) lp >= obj || (void *) (lp)->limit < obj))
+    {
+      plp = lp->prev;
+      lp = plp;
+    }
+  return lp != NULL;
+}
+
+/* Free objects in obstack H, including OBJ and everything allocate
+   more recently than OBJ.  If OBJ is zero, free everything in H.  */
+
+# undef obstack_free
+
+void
+obstack_free (struct obstack *h, void *obj)
+{
+  register struct _obstack_chunk *lp;	/* below addr of any objects in this chunk */
+  register struct _obstack_chunk *plp;	/* point to previous chunk if any */
+
+  lp = h->chunk;
+  /* We use >= because there cannot be an object at the beginning of a chunk.
+     But there can be an empty object at that address
+     at the end of another chunk.  */
+  while (lp != NULL && ((void *) lp >= obj || (void *) (lp)->limit < obj))
+    {
+      plp = lp->prev;
+      CALL_FREEFUN (h, lp);
+      lp = plp;
+      /* If we switch chunks, we can't tell whether the new current
+	 chunk contains an empty object, so assume that it may.  */
+      h->maybe_empty_object = 1;
+    }
+  if (lp)
+    {
+      h->object_base = h->next_free = (char *) (obj);
+      h->chunk_limit = lp->limit;
+      h->chunk = lp;
+    }
+  else if (obj != NULL)
+    /* obj is not in any of the chunks! */
+    abort ();
+}
+
+# ifdef _LIBC
+/* Older versions of libc used a function _obstack_free intended to be
+   called by non-GCC compilers.  */
+strong_alias (obstack_free, _obstack_free)
+# endif
+
+int
+_obstack_memory_used (struct obstack *h)
+{
+  register struct _obstack_chunk* lp;
+  register int nbytes = 0;
+
+  for (lp = h->chunk; lp != NULL; lp = lp->prev)
+    {
+      nbytes += lp->limit - (char *) lp;
+    }
+  return nbytes;
+}
+
+# ifdef _LIBC
+#  include <libio/iolibio.h>
+# endif
+
+# ifndef __attribute__
+/* This feature is available in gcc versions 2.5 and later.  */
+#  if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 5)
+#   define __attribute__(Spec) /* empty */
+#  endif
+# endif
+
+static void
+print_and_abort (void)
+{
+  /* Don't change any of these strings.  Yes, it would be possible to add
+     the newline to the string and use fputs or so.  But this must not
+     happen because the "memory exhausted" message appears in other places
+     like this and the translation should be reused instead of creating
+     a very similar string which requires a separate translation.  */
+# ifdef _LIBC
+  (void) __fxprintf (NULL, "%s\n", _("memory exhausted"));
+# else
+  fprintf (stderr, "%s\n", _("memory exhausted"));
+# endif
+  exit (1);
+}
+
+#endif	/* !ELIDE_CODE */
diff --git a/compat/obstack.h b/compat/obstack.h
new file mode 100644
index 000000000000..ae36ed6a669d
--- /dev/null
+++ b/compat/obstack.h
@@ -0,0 +1,509 @@
+/* obstack.h - object stack macros
+   Copyright (C) 1988-1994,1996-1999,2003,2004,2005,2009
+	Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+/* Summary:
+
+All the apparent functions defined here are macros. The idea
+is that you would use these pre-tested macros to solve a
+very specific set of problems, and they would run fast.
+Caution: no side-effects in arguments please!! They may be
+evaluated MANY times!!
+
+These macros operate a stack of objects.  Each object starts life
+small, and may grow to maturity.  (Consider building a word syllable
+by syllable.)  An object can move while it is growing.  Once it has
+been "finished" it never changes address again.  So the "top of the
+stack" is typically an immature growing object, while the rest of the
+stack is of mature, fixed size and fixed address objects.
+
+These routines grab large chunks of memory, using a function you
+supply, called `obstack_chunk_alloc'.  On occasion, they free chunks,
+by calling `obstack_chunk_free'.  You must define them and declare
+them before using any obstack macros.
+
+Each independent stack is represented by a `struct obstack'.
+Each of the obstack macros expects a pointer to such a structure
+as the first argument.
+
+One motivation for this package is the problem of growing char strings
+in symbol tables.  Unless you are "fascist pig with a read-only mind"
+--Gosper's immortal quote from HAKMEM item 154, out of context--you
+would not like to put any arbitrary upper limit on the length of your
+symbols.
+
+In practice this often means you will build many short symbols and a
+few long symbols.  At the time you are reading a symbol you don't know
+how long it is.  One traditional method is to read a symbol into a
+buffer, realloc()ating the buffer every time you try to read a symbol
+that is longer than the buffer.  This is beaut, but you still will
+want to copy the symbol from the buffer to a more permanent
+symbol-table entry say about half the time.
+
+With obstacks, you can work differently.  Use one obstack for all symbol
+names.  As you read a symbol, grow the name in the obstack gradually.
+When the name is complete, finalize it.  Then, if the symbol exists already,
+free the newly read name.
+
+The way we do this is to take a large chunk, allocating memory from
+low addresses.  When you want to build a symbol in the chunk you just
+add chars above the current "high water mark" in the chunk.  When you
+have finished adding chars, because you got to the end of the symbol,
+you know how long the chars are, and you can create a new object.
+Mostly the chars will not burst over the highest address of the chunk,
+because you would typically expect a chunk to be (say) 100 times as
+long as an average object.
+
+In case that isn't clear, when we have enough chars to make up
+the object, THEY ARE ALREADY CONTIGUOUS IN THE CHUNK (guaranteed)
+so we just point to it where it lies.  No moving of chars is
+needed and this is the second win: potentially long strings need
+never be explicitly shuffled. Once an object is formed, it does not
+change its address during its lifetime.
+
+When the chars burst over a chunk boundary, we allocate a larger
+chunk, and then copy the partly formed object from the end of the old
+chunk to the beginning of the new larger chunk.  We then carry on
+accreting characters to the end of the object as we normally would.
+
+A special macro is provided to add a single char at a time to a
+growing object.  This allows the use of register variables, which
+break the ordinary 'growth' macro.
+
+Summary:
+	We allocate large chunks.
+	We carve out one object at a time from the current chunk.
+	Once carved, an object never moves.
+	We are free to append data of any size to the currently
+	  growing object.
+	Exactly one object is growing in an obstack at any one time.
+	You can run one obstack per control block.
+	You may have as many control blocks as you dare.
+	Because of the way we do it, you can `unwind' an obstack
+	  back to a previous state. (You may remove objects much
+	  as you would with a stack.)
+*/
+
+
+/* Don't do the contents of this file more than once.  */
+
+#ifndef _OBSTACK_H
+#define _OBSTACK_H 1
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* We need the type of a pointer subtraction.  If __PTRDIFF_TYPE__ is
+   defined, as with GNU C, use that; that way we don't pollute the
+   namespace with <stddef.h>'s symbols.  Otherwise, include <stddef.h>
+   and use ptrdiff_t.  */
+
+#ifdef __PTRDIFF_TYPE__
+# define PTR_INT_TYPE __PTRDIFF_TYPE__
+#else
+# include <stddef.h>
+# define PTR_INT_TYPE ptrdiff_t
+#endif
+
+/* If B is the base of an object addressed by P, return the result of
+   aligning P to the next multiple of A + 1.  B and P must be of type
+   char *.  A + 1 must be a power of 2.  */
+
+#define __BPTR_ALIGN(B, P, A) ((B) + (((P) - (B) + (A)) & ~(A)))
+
+/* Similar to _BPTR_ALIGN (B, P, A), except optimize the common case
+   where pointers can be converted to integers, aligned as integers,
+   and converted back again.  If PTR_INT_TYPE is narrower than a
+   pointer (e.g., the AS/400), play it safe and compute the alignment
+   relative to B.  Otherwise, use the faster strategy of computing the
+   alignment relative to 0.  */
+
+#define __PTR_ALIGN(B, P, A)						    \
+  __BPTR_ALIGN (sizeof (PTR_INT_TYPE) < sizeof (void *) ? (B) : (char *) 0, \
+		P, A)
+
+#include <string.h>
+
+struct _obstack_chunk		/* Lives at front of each chunk. */
+{
+  char  *limit;			/* 1 past end of this chunk */
+  struct _obstack_chunk *prev;	/* address of prior chunk or NULL */
+  char	contents[4];		/* objects begin here */
+};
+
+struct obstack		/* control current object in current chunk */
+{
+  long	chunk_size;		/* preferred size to allocate chunks in */
+  struct _obstack_chunk *chunk;	/* address of current struct obstack_chunk */
+  char	*object_base;		/* address of object we are building */
+  char	*next_free;		/* where to add next char to current object */
+  char	*chunk_limit;		/* address of char after current chunk */
+  union
+  {
+    PTR_INT_TYPE tempint;
+    void *tempptr;
+  } temp;			/* Temporary for some macros.  */
+  int   alignment_mask;		/* Mask of alignment for each object. */
+  /* These prototypes vary based on `use_extra_arg'. */
+  union {
+    void *(*plain) (long);
+    struct _obstack_chunk *(*extra) (void *, long);
+  } chunkfun;
+  union {
+    void (*plain) (void *);
+    void (*extra) (void *, struct _obstack_chunk *);
+  } freefun;
+  void *extra_arg;		/* first arg for chunk alloc/dealloc funcs */
+  unsigned use_extra_arg:1;	/* chunk alloc/dealloc funcs take extra arg */
+  unsigned maybe_empty_object:1;/* There is a possibility that the current
+				   chunk contains a zero-length object.  This
+				   prevents freeing the chunk if we allocate
+				   a bigger chunk to replace it. */
+  unsigned alloc_failed:1;	/* No longer used, as we now call the failed
+				   handler on error, but retained for binary
+				   compatibility.  */
+};
+
+/* Declare the external functions we use; they are in obstack.c.  */
+
+extern void _obstack_newchunk (struct obstack *, int);
+extern int _obstack_begin (struct obstack *, int, int,
+			    void *(*) (long), void (*) (void *));
+extern int _obstack_begin_1 (struct obstack *, int, int,
+			     void *(*) (void *, long),
+			     void (*) (void *, void *), void *);
+extern int _obstack_memory_used (struct obstack *);
+
+void obstack_free (struct obstack *, void *);
+
+
+/* Error handler called when `obstack_chunk_alloc' failed to allocate
+   more memory.  This can be set to a user defined function which
+   should either abort gracefully or use longjump - but shouldn't
+   return.  The default action is to print a message and abort.  */
+extern void (*obstack_alloc_failed_handler) (void);
+
+/* Pointer to beginning of object being allocated or to be allocated next.
+   Note that this might not be the final address of the object
+   because a new chunk might be needed to hold the final size.  */
+
+#define obstack_base(h) ((void *) (h)->object_base)
+
+/* Size for allocating ordinary chunks.  */
+
+#define obstack_chunk_size(h) ((h)->chunk_size)
+
+/* Pointer to next byte not yet allocated in current chunk.  */
+
+#define obstack_next_free(h)	((h)->next_free)
+
+/* Mask specifying low bits that should be clear in address of an object.  */
+
+#define obstack_alignment_mask(h) ((h)->alignment_mask)
+
+/* To prevent prototype warnings provide complete argument list.  */
+#define obstack_init(h)						\
+  _obstack_begin ((h), 0, 0,					\
+		  (void *(*) (long)) obstack_chunk_alloc,	\
+		  (void (*) (void *)) obstack_chunk_free)
+
+#define obstack_begin(h, size)					\
+  _obstack_begin ((h), (size), 0,				\
+		  (void *(*) (long)) obstack_chunk_alloc,	\
+		  (void (*) (void *)) obstack_chunk_free)
+
+#define obstack_specify_allocation(h, size, alignment, chunkfun, freefun)  \
+  _obstack_begin ((h), (size), (alignment),				   \
+		  (void *(*) (long)) (chunkfun),			   \
+		  (void (*) (void *)) (freefun))
+
+#define obstack_specify_allocation_with_arg(h, size, alignment, chunkfun, freefun, arg) \
+  _obstack_begin_1 ((h), (size), (alignment),				\
+		    (void *(*) (void *, long)) (chunkfun),		\
+		    (void (*) (void *, void *)) (freefun), (arg))
+
+#define obstack_chunkfun(h, newchunkfun) \
+  ((h)->chunkfun.extra = (struct _obstack_chunk *(*)(void *, long)) (newchunkfun))
+
+#define obstack_freefun(h, newfreefun) \
+  ((h)->freefun.extra = (void (*)(void *, struct _obstack_chunk *)) (newfreefun))
+
+#define obstack_1grow_fast(h,achar) (*((h)->next_free)++ = (achar))
+
+#define obstack_blank_fast(h,n) ((h)->next_free += (n))
+
+#define obstack_memory_used(h) _obstack_memory_used (h)
+
+#if defined __GNUC__ && defined __STDC__ && __STDC__
+/* NextStep 2.0 cc is really gcc 1.93 but it defines __GNUC__ = 2 and
+   does not implement __extension__.  But that compiler doesn't define
+   __GNUC_MINOR__.  */
+# if __GNUC__ < 2 || (__NeXT__ && !__GNUC_MINOR__)
+#  define __extension__
+# endif
+
+/* For GNU C, if not -traditional,
+   we can define these macros to compute all args only once
+   without using a global variable.
+   Also, we can avoid using the `temp' slot, to make faster code.  */
+
+# define obstack_object_size(OBSTACK)					\
+  __extension__								\
+  ({ struct obstack const *__o = (OBSTACK);				\
+     (unsigned) (__o->next_free - __o->object_base); })
+
+# define obstack_room(OBSTACK)						\
+  __extension__								\
+  ({ struct obstack const *__o = (OBSTACK);				\
+     (unsigned) (__o->chunk_limit - __o->next_free); })
+
+# define obstack_make_room(OBSTACK,length)				\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   int __len = (length);						\
+   if (__o->chunk_limit - __o->next_free < __len)			\
+     _obstack_newchunk (__o, __len);					\
+   (void) 0; })
+
+# define obstack_empty_p(OBSTACK)					\
+  __extension__								\
+  ({ struct obstack const *__o = (OBSTACK);				\
+     (__o->chunk->prev == 0						\
+      && __o->next_free == __PTR_ALIGN ((char *) __o->chunk,		\
+					__o->chunk->contents,		\
+					__o->alignment_mask)); })
+
+# define obstack_grow(OBSTACK,where,length)				\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   int __len = (length);						\
+   if (__o->next_free + __len > __o->chunk_limit)			\
+     _obstack_newchunk (__o, __len);					\
+   memcpy (__o->next_free, where, __len);				\
+   __o->next_free += __len;						\
+   (void) 0; })
+
+# define obstack_grow0(OBSTACK,where,length)				\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   int __len = (length);						\
+   if (__o->next_free + __len + 1 > __o->chunk_limit)			\
+     _obstack_newchunk (__o, __len + 1);				\
+   memcpy (__o->next_free, where, __len);				\
+   __o->next_free += __len;						\
+   *(__o->next_free)++ = 0;						\
+   (void) 0; })
+
+# define obstack_1grow(OBSTACK,datum)					\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   if (__o->next_free + 1 > __o->chunk_limit)				\
+     _obstack_newchunk (__o, 1);					\
+   obstack_1grow_fast (__o, datum);					\
+   (void) 0; })
+
+/* These assume that the obstack alignment is good enough for pointers
+   or ints, and that the data added so far to the current object
+   shares that much alignment.  */
+
+# define obstack_ptr_grow(OBSTACK,datum)				\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   if (__o->next_free + sizeof (void *) > __o->chunk_limit)		\
+     _obstack_newchunk (__o, sizeof (void *));				\
+   obstack_ptr_grow_fast (__o, datum); })				\
+
+# define obstack_int_grow(OBSTACK,datum)				\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   if (__o->next_free + sizeof (int) > __o->chunk_limit)		\
+     _obstack_newchunk (__o, sizeof (int));				\
+   obstack_int_grow_fast (__o, datum); })
+
+# define obstack_ptr_grow_fast(OBSTACK,aptr)				\
+__extension__								\
+({ struct obstack *__o1 = (OBSTACK);					\
+   *(const void **) __o1->next_free = (aptr);				\
+   __o1->next_free += sizeof (const void *);				\
+   (void) 0; })
+
+# define obstack_int_grow_fast(OBSTACK,aint)				\
+__extension__								\
+({ struct obstack *__o1 = (OBSTACK);					\
+   *(int *) __o1->next_free = (aint);					\
+   __o1->next_free += sizeof (int);					\
+   (void) 0; })
+
+# define obstack_blank(OBSTACK,length)					\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   int __len = (length);						\
+   if (__o->chunk_limit - __o->next_free < __len)			\
+     _obstack_newchunk (__o, __len);					\
+   obstack_blank_fast (__o, __len);					\
+   (void) 0; })
+
+# define obstack_alloc(OBSTACK,length)					\
+__extension__								\
+({ struct obstack *__h = (OBSTACK);					\
+   obstack_blank (__h, (length));					\
+   obstack_finish (__h); })
+
+# define obstack_copy(OBSTACK,where,length)				\
+__extension__								\
+({ struct obstack *__h = (OBSTACK);					\
+   obstack_grow (__h, (where), (length));				\
+   obstack_finish (__h); })
+
+# define obstack_copy0(OBSTACK,where,length)				\
+__extension__								\
+({ struct obstack *__h = (OBSTACK);					\
+   obstack_grow0 (__h, (where), (length));				\
+   obstack_finish (__h); })
+
+/* The local variable is named __o1 to avoid a name conflict
+   when obstack_blank is called.  */
+# define obstack_finish(OBSTACK)					\
+__extension__								\
+({ struct obstack *__o1 = (OBSTACK);					\
+   void *__value = (void *) __o1->object_base;				\
+   if (__o1->next_free == __value)					\
+     __o1->maybe_empty_object = 1;					\
+   __o1->next_free							\
+     = __PTR_ALIGN (__o1->object_base, __o1->next_free,			\
+		    __o1->alignment_mask);				\
+   if (__o1->next_free - (char *)__o1->chunk				\
+       > __o1->chunk_limit - (char *)__o1->chunk)			\
+     __o1->next_free = __o1->chunk_limit;				\
+   __o1->object_base = __o1->next_free;					\
+   __value; })
+
+# define obstack_free(OBSTACK, OBJ)					\
+__extension__								\
+({ struct obstack *__o = (OBSTACK);					\
+   void *__obj = (OBJ);							\
+   if (__obj > (void *)__o->chunk && __obj < (void *)__o->chunk_limit)  \
+     __o->next_free = __o->object_base = (char *)__obj;			\
+   else (obstack_free) (__o, __obj); })
+
+#else /* not __GNUC__ or not __STDC__ */
+
+# define obstack_object_size(h) \
+ (unsigned) ((h)->next_free - (h)->object_base)
+
+# define obstack_room(h)		\
+ (unsigned) ((h)->chunk_limit - (h)->next_free)
+
+# define obstack_empty_p(h) \
+ ((h)->chunk->prev == 0							\
+  && (h)->next_free == __PTR_ALIGN ((char *) (h)->chunk,		\
+				    (h)->chunk->contents,		\
+				    (h)->alignment_mask))
+
+/* Note that the call to _obstack_newchunk is enclosed in (..., 0)
+   so that we can avoid having void expressions
+   in the arms of the conditional expression.
+   Casting the third operand to void was tried before,
+   but some compilers won't accept it.  */
+
+# define obstack_make_room(h,length)					\
+( (h)->temp.tempint = (length),						\
+  (((h)->next_free + (h)->temp.tempint > (h)->chunk_limit)		\
+   ? (_obstack_newchunk ((h), (h)->temp.tempint), 0) : 0))
+
+# define obstack_grow(h,where,length)					\
+( (h)->temp.tempint = (length),						\
+  (((h)->next_free + (h)->temp.tempint > (h)->chunk_limit)		\
+   ? (_obstack_newchunk ((h), (h)->temp.tempint), 0) : 0),		\
+  memcpy ((h)->next_free, where, (h)->temp.tempint),			\
+  (h)->next_free += (h)->temp.tempint)
+
+# define obstack_grow0(h,where,length)					\
+( (h)->temp.tempint = (length),						\
+  (((h)->next_free + (h)->temp.tempint + 1 > (h)->chunk_limit)		\
+   ? (_obstack_newchunk ((h), (h)->temp.tempint + 1), 0) : 0),		\
+  memcpy ((h)->next_free, where, (h)->temp.tempint),			\
+  (h)->next_free += (h)->temp.tempint,					\
+  *((h)->next_free)++ = 0)
+
+# define obstack_1grow(h,datum)						\
+( (((h)->next_free + 1 > (h)->chunk_limit)				\
+   ? (_obstack_newchunk ((h), 1), 0) : 0),				\
+  obstack_1grow_fast (h, datum))
+
+# define obstack_ptr_grow(h,datum)					\
+( (((h)->next_free + sizeof (char *) > (h)->chunk_limit)		\
+   ? (_obstack_newchunk ((h), sizeof (char *)), 0) : 0),		\
+  obstack_ptr_grow_fast (h, datum))
+
+# define obstack_int_grow(h,datum)					\
+( (((h)->next_free + sizeof (int) > (h)->chunk_limit)			\
+   ? (_obstack_newchunk ((h), sizeof (int)), 0) : 0),			\
+  obstack_int_grow_fast (h, datum))
+
+# define obstack_ptr_grow_fast(h,aptr)					\
+  (((const void **) ((h)->next_free += sizeof (void *)))[-1] = (aptr))
+
+# define obstack_int_grow_fast(h,aint)					\
+  (((int *) ((h)->next_free += sizeof (int)))[-1] = (aint))
+
+# define obstack_blank(h,length)					\
+( (h)->temp.tempint = (length),						\
+  (((h)->chunk_limit - (h)->next_free < (h)->temp.tempint)		\
+   ? (_obstack_newchunk ((h), (h)->temp.tempint), 0) : 0),		\
+  obstack_blank_fast (h, (h)->temp.tempint))
+
+# define obstack_alloc(h,length)					\
+ (obstack_blank ((h), (length)), obstack_finish ((h)))
+
+# define obstack_copy(h,where,length)					\
+ (obstack_grow ((h), (where), (length)), obstack_finish ((h)))
+
+# define obstack_copy0(h,where,length)					\
+ (obstack_grow0 ((h), (where), (length)), obstack_finish ((h)))
+
+# define obstack_finish(h)						\
+( ((h)->next_free == (h)->object_base					\
+   ? (((h)->maybe_empty_object = 1), 0)					\
+   : 0),								\
+  (h)->temp.tempptr = (h)->object_base,					\
+  (h)->next_free							\
+    = __PTR_ALIGN ((h)->object_base, (h)->next_free,			\
+		   (h)->alignment_mask),				\
+  (((h)->next_free - (char *) (h)->chunk				\
+    > (h)->chunk_limit - (char *) (h)->chunk)				\
+   ? ((h)->next_free = (h)->chunk_limit) : 0),				\
+  (h)->object_base = (h)->next_free,					\
+  (h)->temp.tempptr)
+
+# define obstack_free(h,obj)						\
+( (h)->temp.tempint = (char *) (obj) - (char *) (h)->chunk,		\
+  ((((h)->temp.tempint > 0						\
+    && (h)->temp.tempint < (h)->chunk_limit - (char *) (h)->chunk))	\
+   ? (ptrdiff_t) ((h)->next_free = (h)->object_base				\
+	    = (h)->temp.tempint + (char *) (h)->chunk)			\
+   : (((obstack_free) ((h), (h)->temp.tempint + (char *) (h)->chunk), 0), 0)))
+
+#endif /* not __GNUC__ or not __STDC__ */
+
+#ifdef __cplusplus
+}	/* C++ */
+#endif
+
+#endif /* obstack.h */
diff --git a/compat/poll/poll.c b/compat/poll/poll.c
new file mode 100644
index 000000000000..0e95dd493c94
--- /dev/null
+++ b/compat/poll/poll.c
@@ -0,0 +1,633 @@
+/* Emulation for poll(2)
+   Contributed by Paolo Bonzini.
+
+   Copyright 2001-2003, 2006-2011 Free Software Foundation, Inc.
+
+   This file is part of gnulib.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2, or (at your option)
+   any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License along
+   with this program; if not, see <http://www.gnu.org/licenses/>.  */
+
+/* To bump the minimum Windows version to Windows Vista */
+#include "git-compat-util.h"
+
+/* Tell gcc not to warn about the (nfd < 0) tests, below.  */
+#if (__GNUC__ == 4 && 3 <= __GNUC_MINOR__) || 4 < __GNUC__
+# pragma GCC diagnostic ignored "-Wtype-limits"
+#endif
+
+#if defined(WIN32)
+# include <malloc.h>
+#endif
+
+#include <sys/types.h>
+
+#include <errno.h>
+#include <limits.h>
+#include <assert.h>
+
+#if (defined _WIN32 || defined __WIN32__) && ! defined __CYGWIN__
+# define WIN32_NATIVE
+# if defined (_MSC_VER) && !defined(_WIN32_WINNT)
+#  define _WIN32_WINNT 0x0502
+# endif
+# include <winsock2.h>
+# include <windows.h>
+# include <io.h>
+# include <stdio.h>
+# include <conio.h>
+#else
+# include <sys/time.h>
+# include <sys/socket.h>
+# ifndef NO_SYS_SELECT_H
+#  include <sys/select.h>
+# endif
+# include <unistd.h>
+#endif
+
+/* Specification.  */
+#include "poll.h"
+
+#ifdef HAVE_SYS_IOCTL_H
+# include <sys/ioctl.h>
+#endif
+#ifdef HAVE_SYS_FILIO_H
+# include <sys/filio.h>
+#endif
+
+#include <time.h>
+
+#ifndef INFTIM
+# define INFTIM (-1)
+#endif
+
+/* BeOS does not have MSG_PEEK.  */
+#ifndef MSG_PEEK
+# define MSG_PEEK 0
+#endif
+
+#ifdef WIN32_NATIVE
+
+#define IsConsoleHandle(h) (((long) (intptr_t) (h) & 3) == 3)
+
+static BOOL
+IsSocketHandle (HANDLE h)
+{
+  WSANETWORKEVENTS ev;
+
+  if (IsConsoleHandle (h))
+    return FALSE;
+
+  /* Under Wine, it seems that getsockopt returns 0 for pipes too.
+     WSAEnumNetworkEvents instead distinguishes the two correctly.  */
+  ev.lNetworkEvents = 0xDEADBEEF;
+  WSAEnumNetworkEvents ((SOCKET) h, NULL, &ev);
+  return ev.lNetworkEvents != 0xDEADBEEF;
+}
+
+/* Declare data structures for ntdll functions.  */
+typedef struct _FILE_PIPE_LOCAL_INFORMATION {
+  ULONG NamedPipeType;
+  ULONG NamedPipeConfiguration;
+  ULONG MaximumInstances;
+  ULONG CurrentInstances;
+  ULONG InboundQuota;
+  ULONG ReadDataAvailable;
+  ULONG OutboundQuota;
+  ULONG WriteQuotaAvailable;
+  ULONG NamedPipeState;
+  ULONG NamedPipeEnd;
+} FILE_PIPE_LOCAL_INFORMATION, *PFILE_PIPE_LOCAL_INFORMATION;
+
+typedef struct _IO_STATUS_BLOCK
+{
+  union {
+    DWORD Status;
+    PVOID Pointer;
+  } u;
+  ULONG_PTR Information;
+} IO_STATUS_BLOCK, *PIO_STATUS_BLOCK;
+
+typedef enum _FILE_INFORMATION_CLASS {
+  FilePipeLocalInformation = 24
+} FILE_INFORMATION_CLASS, *PFILE_INFORMATION_CLASS;
+
+typedef DWORD (WINAPI *PNtQueryInformationFile)
+	 (HANDLE, IO_STATUS_BLOCK *, VOID *, ULONG, FILE_INFORMATION_CLASS);
+
+# ifndef PIPE_BUF
+#  define PIPE_BUF      512
+# endif
+
+/* Compute revents values for file handle H.  If some events cannot happen
+   for the handle, eliminate them from *P_SOUGHT.  */
+
+static int
+win32_compute_revents (HANDLE h, int *p_sought)
+{
+  int i, ret, happened;
+  INPUT_RECORD *irbuffer;
+  DWORD avail, nbuffer;
+  BOOL bRet;
+  IO_STATUS_BLOCK iosb;
+  FILE_PIPE_LOCAL_INFORMATION fpli;
+  static PNtQueryInformationFile NtQueryInformationFile;
+  static BOOL once_only;
+
+  switch (GetFileType (h))
+    {
+    case FILE_TYPE_PIPE:
+      if (!once_only)
+	{
+	  NtQueryInformationFile = (PNtQueryInformationFile)(void (*)(void))
+	    GetProcAddress (GetModuleHandleW (L"ntdll.dll"),
+			    "NtQueryInformationFile");
+	  once_only = TRUE;
+	}
+
+      happened = 0;
+      if (PeekNamedPipe (h, NULL, 0, NULL, &avail, NULL) != 0)
+	{
+	  if (avail)
+	    happened |= *p_sought & (POLLIN | POLLRDNORM);
+	}
+      else if (GetLastError () == ERROR_BROKEN_PIPE)
+	happened |= POLLHUP;
+
+      else
+	{
+	  /* It was the write-end of the pipe.  Check if it is writable.
+	     If NtQueryInformationFile fails, optimistically assume the pipe is
+	     writable.  This could happen on Win9x, where NtQueryInformationFile
+	     is not available, or if we inherit a pipe that doesn't permit
+	     FILE_READ_ATTRIBUTES access on the write end (I think this should
+	     not happen since WinXP SP2; WINE seems fine too).  Otherwise,
+	     ensure that enough space is available for atomic writes.  */
+	  memset (&iosb, 0, sizeof (iosb));
+	  memset (&fpli, 0, sizeof (fpli));
+
+	  if (!NtQueryInformationFile
+	      || NtQueryInformationFile (h, &iosb, &fpli, sizeof (fpli),
+					 FilePipeLocalInformation)
+	      || fpli.WriteQuotaAvailable >= PIPE_BUF
+	      || (fpli.OutboundQuota < PIPE_BUF &&
+		  fpli.WriteQuotaAvailable == fpli.OutboundQuota))
+	    happened |= *p_sought & (POLLOUT | POLLWRNORM | POLLWRBAND);
+	}
+      return happened;
+
+    case FILE_TYPE_CHAR:
+      ret = WaitForSingleObject (h, 0);
+      if (!IsConsoleHandle (h))
+	return ret == WAIT_OBJECT_0 ? *p_sought & ~(POLLPRI | POLLRDBAND) : 0;
+
+      nbuffer = avail = 0;
+      bRet = GetNumberOfConsoleInputEvents (h, &nbuffer);
+      if (bRet)
+	{
+	  /* Input buffer.  */
+	  *p_sought &= POLLIN | POLLRDNORM;
+	  if (nbuffer == 0)
+	    return POLLHUP;
+	  if (!*p_sought)
+	    return 0;
+
+	  irbuffer = (INPUT_RECORD *) alloca (nbuffer * sizeof (INPUT_RECORD));
+	  bRet = PeekConsoleInput (h, irbuffer, nbuffer, &avail);
+	  if (!bRet || avail == 0)
+	    return POLLHUP;
+
+	  for (i = 0; i < avail; i++)
+	    if (irbuffer[i].EventType == KEY_EVENT)
+	      return *p_sought;
+	  return 0;
+	}
+      else
+	{
+	  /* Screen buffer.  */
+	  *p_sought &= POLLOUT | POLLWRNORM | POLLWRBAND;
+	  return *p_sought;
+	}
+
+    default:
+      ret = WaitForSingleObject (h, 0);
+      if (ret == WAIT_OBJECT_0)
+	return *p_sought & ~(POLLPRI | POLLRDBAND);
+
+      return *p_sought & (POLLOUT | POLLWRNORM | POLLWRBAND);
+    }
+}
+
+/* Convert fd_sets returned by select into revents values.  */
+
+static int
+win32_compute_revents_socket (SOCKET h, int sought, long lNetworkEvents)
+{
+  int happened = 0;
+
+  if ((lNetworkEvents & (FD_READ | FD_ACCEPT | FD_CLOSE)) == FD_ACCEPT)
+    happened |= (POLLIN | POLLRDNORM) & sought;
+
+  else if (lNetworkEvents & (FD_READ | FD_ACCEPT | FD_CLOSE))
+    {
+      int r, error;
+
+      char data[64];
+      WSASetLastError (0);
+      r = recv (h, data, sizeof (data), MSG_PEEK);
+      error = WSAGetLastError ();
+      WSASetLastError (0);
+
+      if (r > 0 || error == WSAENOTCONN)
+	happened |= (POLLIN | POLLRDNORM) & sought;
+
+      /* Distinguish hung-up sockets from other errors.  */
+      else if (r == 0 || error == WSAESHUTDOWN || error == WSAECONNRESET
+	       || error == WSAECONNABORTED || error == WSAENETRESET)
+	happened |= POLLHUP;
+
+      else
+	happened |= POLLERR;
+    }
+
+  if (lNetworkEvents & (FD_WRITE | FD_CONNECT))
+    happened |= (POLLOUT | POLLWRNORM | POLLWRBAND) & sought;
+
+  if (lNetworkEvents & FD_OOB)
+    happened |= (POLLPRI | POLLRDBAND) & sought;
+
+  return happened;
+}
+
+#else /* !MinGW */
+
+/* Convert select(2) returned fd_sets into poll(2) revents values.  */
+static int
+compute_revents (int fd, int sought, fd_set *rfds, fd_set *wfds, fd_set *efds)
+{
+  int happened = 0;
+  if (FD_ISSET (fd, rfds))
+    {
+      int r;
+      int socket_errno;
+
+# if defined __MACH__ && defined __APPLE__
+      /* There is a bug in Mac OS X that causes it to ignore MSG_PEEK
+	 for some kinds of descriptors.  Detect if this descriptor is a
+	 connected socket, a server socket, or something else using a
+	 0-byte recv, and use ioctl(2) to detect POLLHUP.  */
+      r = recv (fd, NULL, 0, MSG_PEEK);
+      socket_errno = (r < 0) ? errno : 0;
+      if (r == 0 || socket_errno == ENOTSOCK)
+	ioctl (fd, FIONREAD, &r);
+# else
+      char data[64];
+      r = recv (fd, data, sizeof (data), MSG_PEEK);
+      socket_errno = (r < 0) ? errno : 0;
+# endif
+      if (r == 0)
+	happened |= POLLHUP;
+
+      /* If the event happened on an unconnected server socket,
+	 that's fine. */
+      else if (r > 0 || ( /* (r == -1) && */ socket_errno == ENOTCONN))
+	happened |= (POLLIN | POLLRDNORM) & sought;
+
+      /* Distinguish hung-up sockets from other errors.  */
+      else if (socket_errno == ESHUTDOWN || socket_errno == ECONNRESET
+	       || socket_errno == ECONNABORTED || socket_errno == ENETRESET)
+	happened |= POLLHUP;
+
+      /* some systems can't use recv() on non-socket, including HP NonStop */
+      else if (/* (r == -1) && */ socket_errno == ENOTSOCK)
+	happened |= (POLLIN | POLLRDNORM) & sought;
+
+      else
+	happened |= POLLERR;
+    }
+
+  if (FD_ISSET (fd, wfds))
+    happened |= (POLLOUT | POLLWRNORM | POLLWRBAND) & sought;
+
+  if (FD_ISSET (fd, efds))
+    happened |= (POLLPRI | POLLRDBAND) & sought;
+
+  return happened;
+}
+#endif /* !MinGW */
+
+int
+poll (struct pollfd *pfd, nfds_t nfd, int timeout)
+{
+#ifndef WIN32_NATIVE
+  fd_set rfds, wfds, efds;
+  struct timeval tv;
+  struct timeval *ptv;
+  int maxfd, rc;
+  nfds_t i;
+
+# ifdef _SC_OPEN_MAX
+  static int sc_open_max = -1;
+
+  if (nfd < 0
+      || (nfd > sc_open_max
+	  && (sc_open_max != -1
+	      || nfd > (sc_open_max = sysconf (_SC_OPEN_MAX)))))
+    {
+      errno = EINVAL;
+      return -1;
+    }
+# else /* !_SC_OPEN_MAX */
+#  ifdef OPEN_MAX
+  if (nfd < 0 || nfd > OPEN_MAX)
+    {
+      errno = EINVAL;
+      return -1;
+    }
+#  endif /* OPEN_MAX -- else, no check is needed */
+# endif /* !_SC_OPEN_MAX */
+
+  /* EFAULT is not necessary to implement, but let's do it in the
+     simplest case. */
+  if (!pfd && nfd)
+    {
+      errno = EFAULT;
+      return -1;
+    }
+
+  /* convert timeout number into a timeval structure */
+  if (timeout == 0)
+    {
+      ptv = &tv;
+      ptv->tv_sec = 0;
+      ptv->tv_usec = 0;
+    }
+  else if (timeout > 0)
+    {
+      ptv = &tv;
+      ptv->tv_sec = timeout / 1000;
+      ptv->tv_usec = (timeout % 1000) * 1000;
+    }
+  else if (timeout == INFTIM)
+    /* wait forever */
+    ptv = NULL;
+  else
+    {
+      errno = EINVAL;
+      return -1;
+    }
+
+  /* create fd sets and determine max fd */
+  maxfd = -1;
+  FD_ZERO (&rfds);
+  FD_ZERO (&wfds);
+  FD_ZERO (&efds);
+  for (i = 0; i < nfd; i++)
+    {
+      if (pfd[i].fd < 0)
+	continue;
+
+      if (pfd[i].events & (POLLIN | POLLRDNORM))
+	FD_SET (pfd[i].fd, &rfds);
+
+      /* see select(2): "the only exceptional condition detectable
+	 is out-of-band data received on a socket", hence we push
+	 POLLWRBAND events onto wfds instead of efds. */
+      if (pfd[i].events & (POLLOUT | POLLWRNORM | POLLWRBAND))
+	FD_SET (pfd[i].fd, &wfds);
+      if (pfd[i].events & (POLLPRI | POLLRDBAND))
+	FD_SET (pfd[i].fd, &efds);
+      if (pfd[i].fd >= maxfd
+	  && (pfd[i].events & (POLLIN | POLLOUT | POLLPRI
+			       | POLLRDNORM | POLLRDBAND
+			       | POLLWRNORM | POLLWRBAND)))
+	{
+	  maxfd = pfd[i].fd;
+	  if (maxfd > FD_SETSIZE)
+	    {
+	      errno = EOVERFLOW;
+	      return -1;
+	    }
+	}
+    }
+
+  /* examine fd sets */
+  rc = select (maxfd + 1, &rfds, &wfds, &efds, ptv);
+  if (rc < 0)
+    return rc;
+
+  /* establish results */
+  rc = 0;
+  for (i = 0; i < nfd; i++)
+    if (pfd[i].fd < 0)
+      pfd[i].revents = 0;
+    else
+      {
+	int happened = compute_revents (pfd[i].fd, pfd[i].events,
+					&rfds, &wfds, &efds);
+	if (happened)
+	  {
+	    pfd[i].revents = happened;
+	    rc++;
+	  }
+	else
+	  {
+	    pfd[i].revents = 0;
+	  }
+      }
+
+  return rc;
+#else
+  static struct timeval tv0;
+  static HANDLE hEvent;
+  WSANETWORKEVENTS ev;
+  HANDLE h, handle_array[FD_SETSIZE + 2];
+  DWORD ret, wait_timeout, nhandles, orig_timeout = 0;
+  ULONGLONG start = 0;
+  fd_set rfds, wfds, xfds;
+  BOOL poll_again;
+  MSG msg;
+  int rc = 0;
+  nfds_t i;
+
+  if (nfd < 0 || timeout < -1)
+    {
+      errno = EINVAL;
+      return -1;
+    }
+
+  if (timeout != INFTIM)
+    {
+      orig_timeout = timeout;
+      start = GetTickCount64();
+    }
+
+  if (!hEvent)
+    hEvent = CreateEvent (NULL, FALSE, FALSE, NULL);
+
+restart:
+  handle_array[0] = hEvent;
+  nhandles = 1;
+  FD_ZERO (&rfds);
+  FD_ZERO (&wfds);
+  FD_ZERO (&xfds);
+
+  /* Classify socket handles and create fd sets. */
+  for (i = 0; i < nfd; i++)
+    {
+      int sought = pfd[i].events;
+      pfd[i].revents = 0;
+      if (pfd[i].fd < 0)
+	continue;
+      if (!(sought & (POLLIN | POLLRDNORM | POLLOUT | POLLWRNORM | POLLWRBAND
+		      | POLLPRI | POLLRDBAND)))
+	continue;
+
+      h = (HANDLE) _get_osfhandle (pfd[i].fd);
+      assert (h != NULL);
+      if (IsSocketHandle (h))
+	{
+	  int requested = FD_CLOSE;
+
+	  /* see above; socket handles are mapped onto select.  */
+	  if (sought & (POLLIN | POLLRDNORM))
+	    {
+	      requested |= FD_READ | FD_ACCEPT;
+	      FD_SET ((SOCKET) h, &rfds);
+	    }
+	  if (sought & (POLLOUT | POLLWRNORM | POLLWRBAND))
+	    {
+	      requested |= FD_WRITE | FD_CONNECT;
+	      FD_SET ((SOCKET) h, &wfds);
+	    }
+	  if (sought & (POLLPRI | POLLRDBAND))
+	    {
+	      requested |= FD_OOB;
+	      FD_SET ((SOCKET) h, &xfds);
+	    }
+
+	  if (requested)
+	    WSAEventSelect ((SOCKET) h, hEvent, requested);
+	}
+      else
+	{
+	  /* Poll now.  If we get an event, do not poll again.  Also,
+	     screen buffer handles are waitable, and they'll block until
+	     a character is available.  win32_compute_revents eliminates
+	     bits for the "wrong" direction. */
+	  pfd[i].revents = win32_compute_revents (h, &sought);
+	  if (sought)
+	    handle_array[nhandles++] = h;
+	  if (pfd[i].revents)
+	    timeout = 0;
+	}
+    }
+
+  if (select (0, &rfds, &wfds, &xfds, &tv0) > 0)
+    {
+      /* Do MsgWaitForMultipleObjects anyway to dispatch messages, but
+	 no need to call select again.  */
+      poll_again = FALSE;
+      wait_timeout = 0;
+    }
+  else
+    {
+      poll_again = TRUE;
+      if (timeout == INFTIM)
+	wait_timeout = INFINITE;
+      else
+	wait_timeout = timeout;
+    }
+
+  for (;;)
+    {
+      ret = MsgWaitForMultipleObjects (nhandles, handle_array, FALSE,
+				       wait_timeout, QS_ALLINPUT);
+
+      if (ret == WAIT_OBJECT_0 + nhandles)
+	{
+	  /* new input of some other kind */
+	  BOOL bRet;
+	  while ((bRet = PeekMessage (&msg, NULL, 0, 0, PM_REMOVE)) != 0)
+	    {
+	      TranslateMessage (&msg);
+	      DispatchMessage (&msg);
+	    }
+	}
+      else
+	break;
+    }
+
+  if (poll_again)
+    select (0, &rfds, &wfds, &xfds, &tv0);
+
+  /* Place a sentinel at the end of the array.  */
+  handle_array[nhandles] = NULL;
+  nhandles = 1;
+  for (i = 0; i < nfd; i++)
+    {
+      int happened;
+
+      if (pfd[i].fd < 0)
+	continue;
+      if (!(pfd[i].events & (POLLIN | POLLRDNORM |
+			     POLLOUT | POLLWRNORM | POLLWRBAND)))
+	continue;
+
+      h = (HANDLE) _get_osfhandle (pfd[i].fd);
+      if (h != handle_array[nhandles])
+	{
+	  /* It's a socket.  */
+	  WSAEnumNetworkEvents ((SOCKET) h, NULL, &ev);
+	  WSAEventSelect ((SOCKET) h, NULL, 0);
+
+	  /* If we're lucky, WSAEnumNetworkEvents already provided a way
+	     to distinguish FD_READ and FD_ACCEPT; this saves a recv later.  */
+	  if (FD_ISSET ((SOCKET) h, &rfds)
+	      && !(ev.lNetworkEvents & (FD_READ | FD_ACCEPT)))
+	    ev.lNetworkEvents |= FD_READ | FD_ACCEPT;
+	  if (FD_ISSET ((SOCKET) h, &wfds))
+	    ev.lNetworkEvents |= FD_WRITE | FD_CONNECT;
+	  if (FD_ISSET ((SOCKET) h, &xfds))
+	    ev.lNetworkEvents |= FD_OOB;
+
+	  happened = win32_compute_revents_socket ((SOCKET) h, pfd[i].events,
+						   ev.lNetworkEvents);
+	}
+      else
+	{
+	  /* Not a socket.  */
+	  int sought = pfd[i].events;
+	  happened = win32_compute_revents (h, &sought);
+	  nhandles++;
+	}
+
+       if ((pfd[i].revents |= happened) != 0)
+	rc++;
+    }
+
+  if (!rc && orig_timeout && timeout != INFTIM)
+    {
+      ULONGLONG elapsed = GetTickCount64() - start;
+      timeout = elapsed >= orig_timeout ? 0 : (int)(orig_timeout - elapsed);
+    }
+
+  if (!rc && timeout)
+    {
+      SleepEx (1, TRUE);
+      goto restart;
+    }
+
+  return rc;
+#endif
+}
diff --git a/compat/poll/poll.h b/compat/poll/poll.h
new file mode 100644
index 000000000000..1e1597360f44
--- /dev/null
+++ b/compat/poll/poll.h
@@ -0,0 +1,67 @@
+/* Header for poll(2) emulation
+   Contributed by Paolo Bonzini.
+
+   Copyright 2001, 2002, 2003, 2007, 2009, 2010 Free Software Foundation, Inc.
+
+   This file is part of gnulib.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2, or (at your option)
+   any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License along
+   with this program; if not, see <http://www.gnu.org/licenses/>.  */
+
+#ifndef _GL_POLL_H
+#define _GL_POLL_H
+
+#if defined(_WIN32_WINNT) && _WIN32_WINNT >= 0x600
+/* Vista has its own, socket-only poll() */
+#undef POLLIN
+#undef POLLPRI
+#undef POLLOUT
+#undef POLLERR
+#undef POLLHUP
+#undef POLLNVAL
+#undef POLLRDNORM
+#undef POLLRDBAND
+#undef POLLWRNORM
+#undef POLLWRBAND
+#define pollfd compat_pollfd
+#endif
+
+/* fake a poll(2) environment */
+#define POLLIN      0x0001      /* any readable data available   */
+#define POLLPRI     0x0002      /* OOB/Urgent readable data      */
+#define POLLOUT     0x0004      /* file descriptor is writeable  */
+#define POLLERR     0x0008      /* some poll error occurred      */
+#define POLLHUP     0x0010      /* file descriptor was "hung up" */
+#define POLLNVAL    0x0020      /* requested events "invalid"    */
+#define POLLRDNORM  0x0040
+#define POLLRDBAND  0x0080
+#define POLLWRNORM  0x0100
+#define POLLWRBAND  0x0200
+
+struct pollfd
+{
+  int fd;                       /* which file descriptor to poll */
+  short events;                 /* events we are interested in   */
+  short revents;                /* events found on return        */
+};
+
+typedef unsigned long nfds_t;
+
+extern int poll (struct pollfd *pfd, nfds_t nfd, int timeout);
+
+/* Define INFTIM only if doing so conforms to POSIX.  */
+#if !defined (_POSIX_C_SOURCE) && !defined (_XOPEN_SOURCE)
+#define INFTIM (-1)
+#endif
+
+#endif /* _GL_POLL_H */
diff --git a/compat/pread.c b/compat/pread.c
new file mode 100644
index 000000000000..978cac4ec91e
--- /dev/null
+++ b/compat/pread.c
@@ -0,0 +1,18 @@
+#include "../git-compat-util.h"
+
+ssize_t git_pread(int fd, void *buf, size_t count, off_t offset)
+{
+        off_t current_offset;
+        ssize_t rc;
+
+        current_offset = lseek(fd, 0, SEEK_CUR);
+
+        if (lseek(fd, offset, SEEK_SET) < 0)
+                return -1;
+
+        rc = read_in_full(fd, buf, count);
+
+        if (current_offset != lseek(fd, current_offset, SEEK_SET))
+                return -1;
+        return rc;
+}
diff --git a/compat/precompose_utf8.c b/compat/precompose_utf8.c
new file mode 100644
index 000000000000..136250fbf6c4
--- /dev/null
+++ b/compat/precompose_utf8.c
@@ -0,0 +1,184 @@
+/*
+ * Converts filenames from decomposed unicode into precomposed unicode.
+ * Used on MacOS X.
+ */
+
+#define PRECOMPOSE_UNICODE_C
+
+#include "cache.h"
+#include "config.h"
+#include "utf8.h"
+#include "precompose_utf8.h"
+
+typedef char *iconv_ibp;
+static const char *repo_encoding = "UTF-8";
+static const char *path_encoding = "UTF-8-MAC";
+
+static size_t has_non_ascii(const char *s, size_t maxlen, size_t *strlen_c)
+{
+	const uint8_t *ptr = (const uint8_t *)s;
+	size_t strlen_chars = 0;
+	size_t ret = 0;
+
+	if (!ptr || !*ptr)
+		return 0;
+
+	while (*ptr && maxlen) {
+		if (*ptr & 0x80)
+			ret++;
+		strlen_chars++;
+		ptr++;
+		maxlen--;
+	}
+	if (strlen_c)
+		*strlen_c = strlen_chars;
+
+	return ret;
+}
+
+
+void probe_utf8_pathname_composition(void)
+{
+	struct strbuf path = STRBUF_INIT;
+	static const char *auml_nfc = "\xc3\xa4";
+	static const char *auml_nfd = "\x61\xcc\x88";
+	int output_fd;
+	if (precomposed_unicode != -1)
+		return; /* We found it defined in the global config, respect it */
+	git_path_buf(&path, "%s", auml_nfc);
+	output_fd = open(path.buf, O_CREAT|O_EXCL|O_RDWR, 0600);
+	if (output_fd >= 0) {
+		close(output_fd);
+		git_path_buf(&path, "%s", auml_nfd);
+		precomposed_unicode = access(path.buf, R_OK) ? 0 : 1;
+		git_config_set("core.precomposeunicode",
+			       precomposed_unicode ? "true" : "false");
+		git_path_buf(&path, "%s", auml_nfc);
+		if (unlink(path.buf))
+			die_errno(_("failed to unlink '%s'"), path.buf);
+	}
+	strbuf_release(&path);
+}
+
+
+void precompose_argv(int argc, const char **argv)
+{
+	int i = 0;
+	const char *oldarg;
+	char *newarg;
+	iconv_t ic_precompose;
+
+	if (precomposed_unicode != 1)
+		return;
+
+	ic_precompose = iconv_open(repo_encoding, path_encoding);
+	if (ic_precompose == (iconv_t) -1)
+		return;
+
+	while (i < argc) {
+		size_t namelen;
+		oldarg = argv[i];
+		if (has_non_ascii(oldarg, (size_t)-1, &namelen)) {
+			newarg = reencode_string_iconv(oldarg, namelen, ic_precompose, 0, NULL);
+			if (newarg)
+				argv[i] = newarg;
+		}
+		i++;
+	}
+	iconv_close(ic_precompose);
+}
+
+
+PREC_DIR *precompose_utf8_opendir(const char *dirname)
+{
+	PREC_DIR *prec_dir = xmalloc(sizeof(PREC_DIR));
+	prec_dir->dirent_nfc = xmalloc(sizeof(dirent_prec_psx));
+	prec_dir->dirent_nfc->max_name_len = sizeof(prec_dir->dirent_nfc->d_name);
+
+	prec_dir->dirp = opendir(dirname);
+	if (!prec_dir->dirp) {
+		free(prec_dir->dirent_nfc);
+		free(prec_dir);
+		return NULL;
+	} else {
+		int ret_errno = errno;
+		prec_dir->ic_precompose = iconv_open(repo_encoding, path_encoding);
+		/* if iconv_open() fails, die() in readdir() if needed */
+		errno = ret_errno;
+	}
+
+	return prec_dir;
+}
+
+struct dirent_prec_psx *precompose_utf8_readdir(PREC_DIR *prec_dir)
+{
+	struct dirent *res;
+	res = readdir(prec_dir->dirp);
+	if (res) {
+		size_t namelenz = strlen(res->d_name) + 1; /* \0 */
+		size_t new_maxlen = namelenz;
+
+		int ret_errno = errno;
+
+		if (new_maxlen > prec_dir->dirent_nfc->max_name_len) {
+			size_t new_len = sizeof(dirent_prec_psx) + new_maxlen -
+				sizeof(prec_dir->dirent_nfc->d_name);
+
+			prec_dir->dirent_nfc = xrealloc(prec_dir->dirent_nfc, new_len);
+			prec_dir->dirent_nfc->max_name_len = new_maxlen;
+		}
+
+		prec_dir->dirent_nfc->d_ino  = res->d_ino;
+		prec_dir->dirent_nfc->d_type = res->d_type;
+
+		if ((precomposed_unicode == 1) && has_non_ascii(res->d_name, (size_t)-1, NULL)) {
+			if (prec_dir->ic_precompose == (iconv_t)-1) {
+				die("iconv_open(%s,%s) failed, but needed:\n"
+						"    precomposed unicode is not supported.\n"
+						"    If you want to use decomposed unicode, run\n"
+						"    \"git config core.precomposeunicode false\"\n",
+						repo_encoding, path_encoding);
+			} else {
+				iconv_ibp	cp = (iconv_ibp)res->d_name;
+				size_t inleft = namelenz;
+				char *outpos = &prec_dir->dirent_nfc->d_name[0];
+				size_t outsz = prec_dir->dirent_nfc->max_name_len;
+				errno = 0;
+				iconv(prec_dir->ic_precompose, &cp, &inleft, &outpos, &outsz);
+				if (errno || inleft) {
+					/*
+					 * iconv() failed and errno could be E2BIG, EILSEQ, EINVAL, EBADF
+					 * MacOS X avoids illegal byte sequences.
+					 * If they occur on a mounted drive (e.g. NFS) it is not worth to
+					 * die() for that, but rather let the user see the original name
+					*/
+					namelenz = 0; /* trigger strlcpy */
+				}
+			}
+		} else
+			namelenz = 0;
+
+		if (!namelenz)
+			strlcpy(prec_dir->dirent_nfc->d_name, res->d_name,
+							prec_dir->dirent_nfc->max_name_len);
+
+		errno = ret_errno;
+		return prec_dir->dirent_nfc;
+	}
+	return NULL;
+}
+
+
+int precompose_utf8_closedir(PREC_DIR *prec_dir)
+{
+	int ret_value;
+	int ret_errno;
+	ret_value = closedir(prec_dir->dirp);
+	ret_errno = errno;
+	if (prec_dir->ic_precompose != (iconv_t)-1)
+		iconv_close(prec_dir->ic_precompose);
+	free(prec_dir->dirent_nfc);
+	free(prec_dir);
+	errno = ret_errno;
+	return ret_value;
+}
diff --git a/compat/precompose_utf8.h b/compat/precompose_utf8.h
new file mode 100644
index 000000000000..6f843d3e1a12
--- /dev/null
+++ b/compat/precompose_utf8.h
@@ -0,0 +1,46 @@
+#ifndef PRECOMPOSE_UNICODE_H
+#define PRECOMPOSE_UNICODE_H
+
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <dirent.h>
+#include <iconv.h>
+
+
+typedef struct dirent_prec_psx {
+	ino_t d_ino;            /* Posix */
+	size_t max_name_len;    /* See below */
+	unsigned char d_type;   /* available on all systems git runs on */
+
+	/*
+	 * See http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/dirent.h.html
+	 * NAME_MAX + 1 should be enough, but some systems have
+	 * NAME_MAX=255 and strlen(d_name) may return 508 or 510
+	 * Solution: allocate more when needed, see precompose_utf8_readdir()
+	 */
+	char   d_name[NAME_MAX+1];
+} dirent_prec_psx;
+
+
+typedef struct {
+	iconv_t ic_precompose;
+	DIR *dirp;
+	struct dirent_prec_psx *dirent_nfc;
+} PREC_DIR;
+
+void precompose_argv(int argc, const char **argv);
+void probe_utf8_pathname_composition(void);
+
+PREC_DIR *precompose_utf8_opendir(const char *dirname);
+struct dirent_prec_psx *precompose_utf8_readdir(PREC_DIR *dirp);
+int precompose_utf8_closedir(PREC_DIR *dirp);
+
+#ifndef PRECOMPOSE_UNICODE_C
+#define dirent dirent_prec_psx
+#define opendir(n) precompose_utf8_opendir(n)
+#define readdir(d) precompose_utf8_readdir(d)
+#define closedir(d) precompose_utf8_closedir(d)
+#define DIR PREC_DIR
+#endif /* PRECOMPOSE_UNICODE_C */
+
+#endif /* PRECOMPOSE_UNICODE_H */
diff --git a/compat/qsort.c b/compat/qsort.c
new file mode 100644
index 000000000000..7d071afb7053
--- /dev/null
+++ b/compat/qsort.c
@@ -0,0 +1,62 @@
+#include "../git-compat-util.h"
+
+/*
+ * A merge sort implementation, simplified from the qsort implementation
+ * by Mike Haertel, which is a part of the GNU C Library.
+ */
+
+static void msort_with_tmp(void *b, size_t n, size_t s,
+			   int (*cmp)(const void *, const void *),
+			   char *t)
+{
+	char *tmp;
+	char *b1, *b2;
+	size_t n1, n2;
+
+	if (n <= 1)
+		return;
+
+	n1 = n / 2;
+	n2 = n - n1;
+	b1 = b;
+	b2 = (char *)b + (n1 * s);
+
+	msort_with_tmp(b1, n1, s, cmp, t);
+	msort_with_tmp(b2, n2, s, cmp, t);
+
+	tmp = t;
+
+	while (n1 > 0 && n2 > 0) {
+		if (cmp(b1, b2) <= 0) {
+			memcpy(tmp, b1, s);
+			tmp += s;
+			b1 += s;
+			--n1;
+		} else {
+			memcpy(tmp, b2, s);
+			tmp += s;
+			b2 += s;
+			--n2;
+		}
+	}
+	if (n1 > 0)
+		memcpy(tmp, b1, n1 * s);
+	memcpy(b, t, (n - n2) * s);
+}
+
+void git_qsort(void *b, size_t n, size_t s,
+	       int (*cmp)(const void *, const void *))
+{
+	const size_t size = st_mult(n, s);
+	char buf[1024];
+
+	if (size < sizeof(buf)) {
+		/* The temporary array fits on the small on-stack buffer. */
+		msort_with_tmp(b, n, s, cmp, buf);
+	} else {
+		/* It's somewhat large, so malloc it.  */
+		char *tmp = xmalloc(size);
+		msort_with_tmp(b, n, s, cmp, tmp);
+		free(tmp);
+	}
+}
diff --git a/compat/qsort_s.c b/compat/qsort_s.c
new file mode 100644
index 000000000000..52d1f0a73dd8
--- /dev/null
+++ b/compat/qsort_s.c
@@ -0,0 +1,69 @@
+#include "../git-compat-util.h"
+
+/*
+ * A merge sort implementation, simplified from the qsort implementation
+ * by Mike Haertel, which is a part of the GNU C Library.
+ * Added context pointer, safety checks and return value.
+ */
+
+static void msort_with_tmp(void *b, size_t n, size_t s,
+			   int (*cmp)(const void *, const void *, void *),
+			   char *t, void *ctx)
+{
+	char *tmp;
+	char *b1, *b2;
+	size_t n1, n2;
+
+	if (n <= 1)
+		return;
+
+	n1 = n / 2;
+	n2 = n - n1;
+	b1 = b;
+	b2 = (char *)b + (n1 * s);
+
+	msort_with_tmp(b1, n1, s, cmp, t, ctx);
+	msort_with_tmp(b2, n2, s, cmp, t, ctx);
+
+	tmp = t;
+
+	while (n1 > 0 && n2 > 0) {
+		if (cmp(b1, b2, ctx) <= 0) {
+			memcpy(tmp, b1, s);
+			tmp += s;
+			b1 += s;
+			--n1;
+		} else {
+			memcpy(tmp, b2, s);
+			tmp += s;
+			b2 += s;
+			--n2;
+		}
+	}
+	if (n1 > 0)
+		memcpy(tmp, b1, n1 * s);
+	memcpy(b, t, (n - n2) * s);
+}
+
+int git_qsort_s(void *b, size_t n, size_t s,
+		int (*cmp)(const void *, const void *, void *), void *ctx)
+{
+	const size_t size = st_mult(n, s);
+	char buf[1024];
+
+	if (!n)
+		return 0;
+	if (!b || !cmp)
+		return -1;
+
+	if (size < sizeof(buf)) {
+		/* The temporary array fits on the small on-stack buffer. */
+		msort_with_tmp(b, n, s, cmp, buf, ctx);
+	} else {
+		/* It's somewhat large, so malloc it.  */
+		char *tmp = xmalloc(size);
+		msort_with_tmp(b, n, s, cmp, tmp, ctx);
+		free(tmp);
+	}
+	return 0;
+}
diff --git a/compat/regex/regcomp.c b/compat/regex/regcomp.c
new file mode 100644
index 000000000000..c0d838834ad8
--- /dev/null
+++ b/compat/regex/regcomp.c
@@ -0,0 +1,3891 @@
+/* Extended regular expression matching and search library.
+   Copyright (C) 2002-2007,2009,2010 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#if defined __TANDEM
+ /* This is currently duplicated from git-compat-utils.h */
+# ifdef NO_INTPTR_T
+ typedef long intptr_t;
+ typedef unsigned long uintptr_t;
+# endif
+#endif
+
+static reg_errcode_t re_compile_internal (regex_t *preg, const char * pattern,
+					  size_t length, reg_syntax_t syntax);
+static void re_compile_fastmap_iter (regex_t *bufp,
+				     const re_dfastate_t *init_state,
+				     char *fastmap);
+static reg_errcode_t init_dfa (re_dfa_t *dfa, size_t pat_len);
+#ifdef RE_ENABLE_I18N
+static void free_charset (re_charset_t *cset);
+#endif /* RE_ENABLE_I18N */
+static void free_workarea_compile (regex_t *preg);
+static reg_errcode_t create_initial_state (re_dfa_t *dfa);
+#ifdef RE_ENABLE_I18N
+static void optimize_utf8 (re_dfa_t *dfa);
+#endif
+static reg_errcode_t analyze (regex_t *preg);
+static reg_errcode_t preorder (bin_tree_t *root,
+			       reg_errcode_t (fn (void *, bin_tree_t *)),
+			       void *extra);
+static reg_errcode_t postorder (bin_tree_t *root,
+				reg_errcode_t (fn (void *, bin_tree_t *)),
+				void *extra);
+static reg_errcode_t optimize_subexps (void *extra, bin_tree_t *node);
+static reg_errcode_t lower_subexps (void *extra, bin_tree_t *node);
+static bin_tree_t *lower_subexp (reg_errcode_t *err, regex_t *preg,
+				 bin_tree_t *node);
+static reg_errcode_t calc_first (void *extra, bin_tree_t *node);
+static reg_errcode_t calc_next (void *extra, bin_tree_t *node);
+static reg_errcode_t link_nfa_nodes (void *extra, bin_tree_t *node);
+static int duplicate_node (re_dfa_t *dfa, int org_idx, unsigned int constraint);
+static int search_duplicated_node (const re_dfa_t *dfa, int org_node,
+				   unsigned int constraint);
+static reg_errcode_t calc_eclosure (re_dfa_t *dfa);
+static reg_errcode_t calc_eclosure_iter (re_node_set *new_set, re_dfa_t *dfa,
+					 int node, int root);
+static reg_errcode_t calc_inveclosure (re_dfa_t *dfa);
+static int fetch_number (re_string_t *input, re_token_t *token,
+			 reg_syntax_t syntax);
+static int peek_token (re_token_t *token, re_string_t *input,
+			reg_syntax_t syntax) internal_function;
+static bin_tree_t *parse (re_string_t *regexp, regex_t *preg,
+			  reg_syntax_t syntax, reg_errcode_t *err);
+static bin_tree_t *parse_reg_exp (re_string_t *regexp, regex_t *preg,
+				  re_token_t *token, reg_syntax_t syntax,
+				  int nest, reg_errcode_t *err);
+static bin_tree_t *parse_branch (re_string_t *regexp, regex_t *preg,
+				 re_token_t *token, reg_syntax_t syntax,
+				 int nest, reg_errcode_t *err);
+static bin_tree_t *parse_expression (re_string_t *regexp, regex_t *preg,
+				     re_token_t *token, reg_syntax_t syntax,
+				     int nest, reg_errcode_t *err);
+static bin_tree_t *parse_sub_exp (re_string_t *regexp, regex_t *preg,
+				  re_token_t *token, reg_syntax_t syntax,
+				  int nest, reg_errcode_t *err);
+static bin_tree_t *parse_dup_op (bin_tree_t *dup_elem, re_string_t *regexp,
+				 re_dfa_t *dfa, re_token_t *token,
+				 reg_syntax_t syntax, reg_errcode_t *err);
+static bin_tree_t *parse_bracket_exp (re_string_t *regexp, re_dfa_t *dfa,
+				      re_token_t *token, reg_syntax_t syntax,
+				      reg_errcode_t *err);
+static reg_errcode_t parse_bracket_element (bracket_elem_t *elem,
+					    re_string_t *regexp,
+					    re_token_t *token, int token_len,
+					    re_dfa_t *dfa,
+					    reg_syntax_t syntax,
+					    int accept_hyphen);
+static reg_errcode_t parse_bracket_symbol (bracket_elem_t *elem,
+					  re_string_t *regexp,
+					  re_token_t *token);
+#ifdef RE_ENABLE_I18N
+static reg_errcode_t build_equiv_class (bitset_t sbcset,
+					re_charset_t *mbcset,
+					int *equiv_class_alloc,
+					const unsigned char *name);
+static reg_errcode_t build_charclass (RE_TRANSLATE_TYPE trans,
+				      bitset_t sbcset,
+				      re_charset_t *mbcset,
+				      int *char_class_alloc,
+				      const char *class_name,
+				      reg_syntax_t syntax);
+#else  /* not RE_ENABLE_I18N */
+static reg_errcode_t build_equiv_class (bitset_t sbcset,
+					const unsigned char *name);
+static reg_errcode_t build_charclass (RE_TRANSLATE_TYPE trans,
+				      bitset_t sbcset,
+				      const char *class_name,
+				      reg_syntax_t syntax);
+#endif /* not RE_ENABLE_I18N */
+static bin_tree_t *build_charclass_op (re_dfa_t *dfa,
+				       RE_TRANSLATE_TYPE trans,
+				       const char *class_name,
+				       const char *extra,
+				       int non_match, reg_errcode_t *err);
+static bin_tree_t *create_tree (re_dfa_t *dfa,
+				bin_tree_t *left, bin_tree_t *right,
+				re_token_type_t type);
+static bin_tree_t *create_token_tree (re_dfa_t *dfa,
+				      bin_tree_t *left, bin_tree_t *right,
+				      const re_token_t *token);
+static bin_tree_t *duplicate_tree (const bin_tree_t *src, re_dfa_t *dfa);
+static void free_token (re_token_t *node);
+static reg_errcode_t free_tree (void *extra, bin_tree_t *node);
+static reg_errcode_t mark_opt_subexp (void *extra, bin_tree_t *node);
+
+/* This table gives an error message for each of the error codes listed
+   in regex.h.  Obviously the order here has to be same as there.
+   POSIX doesn't require that we do anything for REG_NOERROR,
+   but why not be nice?  */
+
+const char __re_error_msgid[] attribute_hidden =
+  {
+#define REG_NOERROR_IDX	0
+    gettext_noop ("Success")	/* REG_NOERROR */
+    "\0"
+#define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
+    gettext_noop ("No match")	/* REG_NOMATCH */
+    "\0"
+#define REG_BADPAT_IDX	(REG_NOMATCH_IDX + sizeof "No match")
+    gettext_noop ("Invalid regular expression") /* REG_BADPAT */
+    "\0"
+#define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
+    gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
+    "\0"
+#define REG_ECTYPE_IDX	(REG_ECOLLATE_IDX + sizeof "Invalid collation character")
+    gettext_noop ("Invalid character class name") /* REG_ECTYPE */
+    "\0"
+#define REG_EESCAPE_IDX	(REG_ECTYPE_IDX + sizeof "Invalid character class name")
+    gettext_noop ("Trailing backslash") /* REG_EESCAPE */
+    "\0"
+#define REG_ESUBREG_IDX	(REG_EESCAPE_IDX + sizeof "Trailing backslash")
+    gettext_noop ("Invalid back reference") /* REG_ESUBREG */
+    "\0"
+#define REG_EBRACK_IDX	(REG_ESUBREG_IDX + sizeof "Invalid back reference")
+    gettext_noop ("Unmatched [ or [^")	/* REG_EBRACK */
+    "\0"
+#define REG_EPAREN_IDX	(REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
+    gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
+    "\0"
+#define REG_EBRACE_IDX	(REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
+    gettext_noop ("Unmatched \\{") /* REG_EBRACE */
+    "\0"
+#define REG_BADBR_IDX	(REG_EBRACE_IDX + sizeof "Unmatched \\{")
+    gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
+    "\0"
+#define REG_ERANGE_IDX	(REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
+    gettext_noop ("Invalid range end")	/* REG_ERANGE */
+    "\0"
+#define REG_ESPACE_IDX	(REG_ERANGE_IDX + sizeof "Invalid range end")
+    gettext_noop ("Memory exhausted") /* REG_ESPACE */
+    "\0"
+#define REG_BADRPT_IDX	(REG_ESPACE_IDX + sizeof "Memory exhausted")
+    gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
+    "\0"
+#define REG_EEND_IDX	(REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
+    gettext_noop ("Premature end of regular expression") /* REG_EEND */
+    "\0"
+#define REG_ESIZE_IDX	(REG_EEND_IDX + sizeof "Premature end of regular expression")
+    gettext_noop ("Regular expression too big") /* REG_ESIZE */
+    "\0"
+#define REG_ERPAREN_IDX	(REG_ESIZE_IDX + sizeof "Regular expression too big")
+    gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
+  };
+
+const size_t __re_error_msgid_idx[] attribute_hidden =
+  {
+    REG_NOERROR_IDX,
+    REG_NOMATCH_IDX,
+    REG_BADPAT_IDX,
+    REG_ECOLLATE_IDX,
+    REG_ECTYPE_IDX,
+    REG_EESCAPE_IDX,
+    REG_ESUBREG_IDX,
+    REG_EBRACK_IDX,
+    REG_EPAREN_IDX,
+    REG_EBRACE_IDX,
+    REG_BADBR_IDX,
+    REG_ERANGE_IDX,
+    REG_ESPACE_IDX,
+    REG_BADRPT_IDX,
+    REG_EEND_IDX,
+    REG_ESIZE_IDX,
+    REG_ERPAREN_IDX
+  };
+
+/* Entry points for GNU code.  */
+
+
+#ifdef ZOS_USS
+
+/* For ZOS USS we must define btowc */
+
+wchar_t 
+btowc (int c)
+{
+   wchar_t wtmp[2];
+   char tmp[2];
+
+   tmp[0] = c;
+   tmp[1] = 0;
+
+   mbtowc (wtmp, tmp, 1);
+   return wtmp[0];
+}
+#endif
+
+/* re_compile_pattern is the GNU regular expression compiler: it
+   compiles PATTERN (of length LENGTH) and puts the result in BUFP.
+   Returns 0 if the pattern was valid, otherwise an error string.
+
+   Assumes the `allocated' (and perhaps `buffer') and `translate' fields
+   are set in BUFP on entry.  */
+
+const char *
+re_compile_pattern (const char *pattern,
+		    size_t length,
+		    struct re_pattern_buffer *bufp)
+{
+  reg_errcode_t ret;
+
+  /* And GNU code determines whether or not to get register information
+     by passing null for the REGS argument to re_match, etc., not by
+     setting no_sub, unless RE_NO_SUB is set.  */
+  bufp->no_sub = !!(re_syntax_options & RE_NO_SUB);
+
+  /* Match anchors at newline.  */
+  bufp->newline_anchor = 1;
+
+  ret = re_compile_internal (bufp, pattern, length, re_syntax_options);
+
+  if (!ret)
+    return NULL;
+  return gettext (__re_error_msgid + __re_error_msgid_idx[(int) ret]);
+}
+#ifdef _LIBC
+weak_alias (__re_compile_pattern, re_compile_pattern)
+#endif
+
+/* Set by `re_set_syntax' to the current regexp syntax to recognize.  Can
+   also be assigned to arbitrarily: each pattern buffer stores its own
+   syntax, so it can be changed between regex compilations.  */
+/* This has no initializer because initialized variables in Emacs
+   become read-only after dumping.  */
+reg_syntax_t re_syntax_options;
+
+
+/* Specify the precise syntax of regexps for compilation.  This provides
+   for compatibility for various utilities which historically have
+   different, incompatible syntaxes.
+
+   The argument SYNTAX is a bit mask comprised of the various bits
+   defined in regex.h.  We return the old syntax.  */
+
+reg_syntax_t
+re_set_syntax (reg_syntax_t syntax)
+{
+  reg_syntax_t ret = re_syntax_options;
+
+  re_syntax_options = syntax;
+  return ret;
+}
+#ifdef _LIBC
+weak_alias (__re_set_syntax, re_set_syntax)
+#endif
+
+int
+re_compile_fastmap (struct re_pattern_buffer *bufp)
+{
+  re_dfa_t *dfa = (re_dfa_t *) bufp->buffer;
+  char *fastmap = bufp->fastmap;
+
+  memset (fastmap, '\0', sizeof (char) * SBC_MAX);
+  re_compile_fastmap_iter (bufp, dfa->init_state, fastmap);
+  if (dfa->init_state != dfa->init_state_word)
+    re_compile_fastmap_iter (bufp, dfa->init_state_word, fastmap);
+  if (dfa->init_state != dfa->init_state_nl)
+    re_compile_fastmap_iter (bufp, dfa->init_state_nl, fastmap);
+  if (dfa->init_state != dfa->init_state_begbuf)
+    re_compile_fastmap_iter (bufp, dfa->init_state_begbuf, fastmap);
+  bufp->fastmap_accurate = 1;
+  return 0;
+}
+#ifdef _LIBC
+weak_alias (__re_compile_fastmap, re_compile_fastmap)
+#endif
+
+static inline void
+__attribute ((always_inline))
+re_set_fastmap (char *fastmap, int icase, int ch)
+{
+  fastmap[ch] = 1;
+  if (icase)
+    fastmap[tolower (ch)] = 1;
+}
+
+/* Helper function for re_compile_fastmap.
+   Compile fastmap for the initial_state INIT_STATE.  */
+
+static void
+re_compile_fastmap_iter (regex_t *bufp, const re_dfastate_t *init_state,
+			 char *fastmap)
+{
+  volatile re_dfa_t *dfa = (re_dfa_t *) bufp->buffer;
+  int node_cnt;
+  int icase = (dfa->mb_cur_max == 1 && (bufp->syntax & RE_ICASE));
+  for (node_cnt = 0; node_cnt < init_state->nodes.nelem; ++node_cnt)
+    {
+      int node = init_state->nodes.elems[node_cnt];
+      re_token_type_t type = dfa->nodes[node].type;
+
+      if (type == CHARACTER)
+	{
+	  re_set_fastmap (fastmap, icase, dfa->nodes[node].opr.c);
+#ifdef RE_ENABLE_I18N
+	  if ((bufp->syntax & RE_ICASE) && dfa->mb_cur_max > 1)
+	    {
+	      unsigned char *buf = re_malloc (unsigned char, dfa->mb_cur_max), *p;
+	      wchar_t wc;
+	      mbstate_t state;
+
+	      p = buf;
+	      *p++ = dfa->nodes[node].opr.c;
+	      while (++node < dfa->nodes_len
+		     && dfa->nodes[node].type == CHARACTER
+		     && dfa->nodes[node].mb_partial)
+		*p++ = dfa->nodes[node].opr.c;
+	      memset (&state, '\0', sizeof (state));
+	      if (__mbrtowc (&wc, (const char *) buf, p - buf,
+			     &state) == p - buf
+		  && (__wcrtomb ((char *) buf, towlower (wc), &state)
+		      != (size_t) -1))
+		re_set_fastmap (fastmap, 0, buf[0]);
+	      re_free (buf);
+	    }
+#endif
+	}
+      else if (type == SIMPLE_BRACKET)
+	{
+	  int i, ch;
+	  for (i = 0, ch = 0; i < BITSET_WORDS; ++i)
+	    {
+	      int j;
+	      bitset_word_t w = dfa->nodes[node].opr.sbcset[i];
+	      for (j = 0; j < BITSET_WORD_BITS; ++j, ++ch)
+		if (w & ((bitset_word_t) 1 << j))
+		  re_set_fastmap (fastmap, icase, ch);
+	    }
+	}
+#ifdef RE_ENABLE_I18N
+      else if (type == COMPLEX_BRACKET)
+	{
+	  re_charset_t *cset = dfa->nodes[node].opr.mbcset;
+	  int i;
+
+# ifdef _LIBC
+	  /* See if we have to try all bytes which start multiple collation
+	     elements.
+	     e.g. In da_DK, we want to catch 'a' since "aa" is a valid
+		  collation element, and don't catch 'b' since 'b' is
+		  the only collation element which starts from 'b' (and
+		  it is caught by SIMPLE_BRACKET).  */
+	      if (_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES) != 0
+		  && (cset->ncoll_syms || cset->nranges))
+		{
+		  const int32_t *table = (const int32_t *)
+		    _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
+		  for (i = 0; i < SBC_MAX; ++i)
+		    if (table[i] < 0)
+		      re_set_fastmap (fastmap, icase, i);
+		}
+# endif /* _LIBC */
+
+	  /* See if we have to start the match at all multibyte characters,
+	     i.e. where we would not find an invalid sequence.  This only
+	     applies to multibyte character sets; for single byte character
+	     sets, the SIMPLE_BRACKET again suffices.  */
+	  if (dfa->mb_cur_max > 1
+	      && (cset->nchar_classes || cset->non_match || cset->nranges
+# ifdef _LIBC
+		  || cset->nequiv_classes
+# endif /* _LIBC */
+		 ))
+	    {
+	      unsigned char c = 0;
+	      do
+		{
+		  mbstate_t mbs;
+		  memset (&mbs, 0, sizeof (mbs));
+		  if (__mbrtowc (NULL, (char *) &c, 1, &mbs) == (size_t) -2)
+		    re_set_fastmap (fastmap, false, (int) c);
+		}
+	      while (++c != 0);
+	    }
+
+	  else
+	    {
+	      /* ... Else catch all bytes which can start the mbchars.  */
+	      for (i = 0; i < cset->nmbchars; ++i)
+		{
+		  char buf[256];
+		  mbstate_t state;
+		  memset (&state, '\0', sizeof (state));
+		  if (__wcrtomb (buf, cset->mbchars[i], &state) != (size_t) -1)
+		    re_set_fastmap (fastmap, icase, *(unsigned char *) buf);
+		  if ((bufp->syntax & RE_ICASE) && dfa->mb_cur_max > 1)
+		    {
+		      if (__wcrtomb (buf, towlower (cset->mbchars[i]), &state)
+			  != (size_t) -1)
+			re_set_fastmap (fastmap, false, *(unsigned char *) buf);
+		    }
+		}
+	    }
+	}
+#endif /* RE_ENABLE_I18N */
+      else if (type == OP_PERIOD
+#ifdef RE_ENABLE_I18N
+	       || type == OP_UTF8_PERIOD
+#endif /* RE_ENABLE_I18N */
+	       || type == END_OF_RE)
+	{
+	  memset (fastmap, '\1', sizeof (char) * SBC_MAX);
+	  if (type == END_OF_RE)
+	    bufp->can_be_null = 1;
+	  return;
+	}
+    }
+}
+
+/* Entry point for POSIX code.  */
+/* regcomp takes a regular expression as a string and compiles it.
+
+   PREG is a regex_t *.  We do not expect any fields to be initialized,
+   since POSIX says we shouldn't.  Thus, we set
+
+     `buffer' to the compiled pattern;
+     `used' to the length of the compiled pattern;
+     `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
+       REG_EXTENDED bit in CFLAGS is set; otherwise, to
+       RE_SYNTAX_POSIX_BASIC;
+     `newline_anchor' to REG_NEWLINE being set in CFLAGS;
+     `fastmap' to an allocated space for the fastmap;
+     `fastmap_accurate' to zero;
+     `re_nsub' to the number of subexpressions in PATTERN.
+
+   PATTERN is the address of the pattern string.
+
+   CFLAGS is a series of bits which affect compilation.
+
+     If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
+     use POSIX basic syntax.
+
+     If REG_NEWLINE is set, then . and [^...] don't match newline.
+     Also, regexec will try a match beginning after every newline.
+
+     If REG_ICASE is set, then we considers upper- and lowercase
+     versions of letters to be equivalent when matching.
+
+     If REG_NOSUB is set, then when PREG is passed to regexec, that
+     routine will report only success or failure, and nothing about the
+     registers.
+
+   It returns 0 if it succeeds, nonzero if it doesn't.  (See regex.h for
+   the return codes and their meanings.)  */
+
+int
+regcomp (regex_t *__restrict preg,
+	 const char *__restrict pattern,
+	 int cflags)
+{
+  reg_errcode_t ret;
+  reg_syntax_t syntax = ((cflags & REG_EXTENDED) ? RE_SYNTAX_POSIX_EXTENDED
+			 : RE_SYNTAX_POSIX_BASIC);
+
+  preg->buffer = NULL;
+  preg->allocated = 0;
+  preg->used = 0;
+
+  /* Try to allocate space for the fastmap.  */
+  preg->fastmap = re_malloc (char, SBC_MAX);
+  if (BE (preg->fastmap == NULL, 0))
+    return REG_ESPACE;
+
+  syntax |= (cflags & REG_ICASE) ? RE_ICASE : 0;
+
+  /* If REG_NEWLINE is set, newlines are treated differently.  */
+  if (cflags & REG_NEWLINE)
+    { /* REG_NEWLINE implies neither . nor [^...] match newline.  */
+      syntax &= ~RE_DOT_NEWLINE;
+      syntax |= RE_HAT_LISTS_NOT_NEWLINE;
+      /* It also changes the matching behavior.  */
+      preg->newline_anchor = 1;
+    }
+  else
+    preg->newline_anchor = 0;
+  preg->no_sub = !!(cflags & REG_NOSUB);
+  preg->translate = NULL;
+
+  ret = re_compile_internal (preg, pattern, strlen (pattern), syntax);
+
+  /* POSIX doesn't distinguish between an unmatched open-group and an
+     unmatched close-group: both are REG_EPAREN.  */
+  if (ret == REG_ERPAREN)
+    ret = REG_EPAREN;
+
+  /* We have already checked preg->fastmap != NULL.  */
+  if (BE (ret == REG_NOERROR, 1))
+    /* Compute the fastmap now, since regexec cannot modify the pattern
+       buffer.  This function never fails in this implementation.  */
+    (void) re_compile_fastmap (preg);
+  else
+    {
+      /* Some error occurred while compiling the expression.  */
+      re_free (preg->fastmap);
+      preg->fastmap = NULL;
+    }
+
+  return (int) ret;
+}
+#ifdef _LIBC
+weak_alias (__regcomp, regcomp)
+#endif
+
+/* Returns a message corresponding to an error code, ERRCODE, returned
+   from either regcomp or regexec.   We don't use PREG here.  */
+
+size_t
+regerror(int errcode, const regex_t *__restrict preg,
+	 char *__restrict errbuf, size_t errbuf_size)
+{
+  const char *msg;
+  size_t msg_size;
+
+  if (BE (errcode < 0
+	  || errcode >= (int) (sizeof (__re_error_msgid_idx)
+			       / sizeof (__re_error_msgid_idx[0])), 0))
+    /* Only error codes returned by the rest of the code should be passed
+       to this routine.  If we are given anything else, or if other regex
+       code generates an invalid error code, then the program has a bug.
+       Dump core so we can fix it.  */
+    abort ();
+
+  msg = gettext (__re_error_msgid + __re_error_msgid_idx[errcode]);
+
+  msg_size = strlen (msg) + 1; /* Includes the null.  */
+
+  if (BE (errbuf_size != 0, 1))
+    {
+      if (BE (msg_size > errbuf_size, 0))
+	{
+	  memcpy (errbuf, msg, errbuf_size - 1);
+	  errbuf[errbuf_size - 1] = 0;
+	}
+      else
+	memcpy (errbuf, msg, msg_size);
+    }
+
+  return msg_size;
+}
+#ifdef _LIBC
+weak_alias (__regerror, regerror)
+#endif
+
+
+#ifdef RE_ENABLE_I18N
+/* This static array is used for the map to single-byte characters when
+   UTF-8 is used.  Otherwise we would allocate memory just to initialize
+   it the same all the time.  UTF-8 is the preferred encoding so this is
+   a worthwhile optimization.  */
+#if __GNUC__ >= 3
+static const bitset_t utf8_sb_map = {
+  /* Set the first 128 bits.  */
+  [0 ... 0x80 / BITSET_WORD_BITS - 1] = BITSET_WORD_MAX
+};
+#else /* ! (__GNUC__ >= 3) */
+static bitset_t utf8_sb_map;
+#endif /* __GNUC__ >= 3 */
+#endif /* RE_ENABLE_I18N */
+
+
+static void
+free_dfa_content (re_dfa_t *dfa)
+{
+  int i, j;
+
+  if (dfa->nodes)
+    for (i = 0; i < dfa->nodes_len; ++i)
+      free_token (dfa->nodes + i);
+  re_free (dfa->nexts);
+  for (i = 0; i < dfa->nodes_len; ++i)
+    {
+      if (dfa->eclosures != NULL)
+	re_node_set_free (dfa->eclosures + i);
+      if (dfa->inveclosures != NULL)
+	re_node_set_free (dfa->inveclosures + i);
+      if (dfa->edests != NULL)
+	re_node_set_free (dfa->edests + i);
+    }
+  re_free (dfa->edests);
+  re_free (dfa->eclosures);
+  re_free (dfa->inveclosures);
+  re_free (dfa->nodes);
+
+  if (dfa->state_table)
+    for (i = 0; i <= dfa->state_hash_mask; ++i)
+      {
+	struct re_state_table_entry *entry = dfa->state_table + i;
+	for (j = 0; j < entry->num; ++j)
+	  {
+	    re_dfastate_t *state = entry->array[j];
+	    free_state (state);
+	  }
+	re_free (entry->array);
+      }
+  re_free (dfa->state_table);
+#ifdef RE_ENABLE_I18N
+  if (dfa->sb_char != utf8_sb_map)
+    re_free (dfa->sb_char);
+#endif
+  re_free (dfa->subexp_map);
+#ifdef DEBUG
+  re_free (dfa->re_str);
+#endif
+
+  re_free (dfa);
+}
+
+
+/* Free dynamically allocated space used by PREG.  */
+
+void
+regfree (regex_t *preg)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  if (BE (dfa != NULL, 1))
+    free_dfa_content (dfa);
+  preg->buffer = NULL;
+  preg->allocated = 0;
+
+  re_free (preg->fastmap);
+  preg->fastmap = NULL;
+
+  re_free (preg->translate);
+  preg->translate = NULL;
+}
+#ifdef _LIBC
+weak_alias (__regfree, regfree)
+#endif
+
+/* Entry points compatible with 4.2 BSD regex library.  We don't define
+   them unless specifically requested.  */
+
+#if defined _REGEX_RE_COMP || defined _LIBC
+
+/* BSD has one and only one pattern buffer.  */
+static struct re_pattern_buffer re_comp_buf;
+
+char *
+# ifdef _LIBC
+/* Make these definitions weak in libc, so POSIX programs can redefine
+   these names if they don't use our functions, and still use
+   regcomp/regexec above without link errors.  */
+weak_function
+# endif
+re_comp (s)
+     const char *s;
+{
+  reg_errcode_t ret;
+  char *fastmap;
+
+  if (!s)
+    {
+      if (!re_comp_buf.buffer)
+	return gettext ("No previous regular expression");
+      return 0;
+    }
+
+  if (re_comp_buf.buffer)
+    {
+      fastmap = re_comp_buf.fastmap;
+      re_comp_buf.fastmap = NULL;
+      __regfree (&re_comp_buf);
+      memset (&re_comp_buf, '\0', sizeof (re_comp_buf));
+      re_comp_buf.fastmap = fastmap;
+    }
+
+  if (re_comp_buf.fastmap == NULL)
+    {
+      re_comp_buf.fastmap = (char *) malloc (SBC_MAX);
+      if (re_comp_buf.fastmap == NULL)
+	return (char *) gettext (__re_error_msgid
+				 + __re_error_msgid_idx[(int) REG_ESPACE]);
+    }
+
+  /* Since `re_exec' always passes NULL for the `regs' argument, we
+     don't need to initialize the pattern buffer fields which affect it.  */
+
+  /* Match anchors at newlines.  */
+  re_comp_buf.newline_anchor = 1;
+
+  ret = re_compile_internal (&re_comp_buf, s, strlen (s), re_syntax_options);
+
+  if (!ret)
+    return NULL;
+
+  /* Yes, we're discarding `const' here if !HAVE_LIBINTL.  */
+  return (char *) gettext (__re_error_msgid + __re_error_msgid_idx[(int) ret]);
+}
+
+#ifdef _LIBC
+libc_freeres_fn (free_mem)
+{
+  __regfree (&re_comp_buf);
+}
+#endif
+
+#endif /* _REGEX_RE_COMP */
+
+/* Internal entry point.
+   Compile the regular expression PATTERN, whose length is LENGTH.
+   SYNTAX indicate regular expression's syntax.  */
+
+static reg_errcode_t
+re_compile_internal (regex_t *preg, const char * pattern, size_t length,
+		     reg_syntax_t syntax)
+{
+  reg_errcode_t err = REG_NOERROR;
+  re_dfa_t *dfa;
+  re_string_t regexp;
+
+  /* Initialize the pattern buffer.  */
+  preg->fastmap_accurate = 0;
+  preg->syntax = syntax;
+  preg->not_bol = preg->not_eol = 0;
+  preg->used = 0;
+  preg->re_nsub = 0;
+  preg->can_be_null = 0;
+  preg->regs_allocated = REGS_UNALLOCATED;
+
+  /* Initialize the dfa.  */
+  dfa = (re_dfa_t *) preg->buffer;
+  if (BE (preg->allocated < sizeof (re_dfa_t), 0))
+    {
+      /* If zero allocated, but buffer is non-null, try to realloc
+	 enough space.  This loses if buffer's address is bogus, but
+	 that is the user's responsibility.  If ->buffer is NULL this
+	 is a simple allocation.  */
+      dfa = re_realloc (preg->buffer, re_dfa_t, 1);
+      if (dfa == NULL)
+	return REG_ESPACE;
+      preg->allocated = sizeof (re_dfa_t);
+      preg->buffer = (unsigned char *) dfa;
+    }
+  preg->used = sizeof (re_dfa_t);
+
+  err = init_dfa (dfa, length);
+  if (BE (err != REG_NOERROR, 0))
+    {
+      free_dfa_content (dfa);
+      preg->buffer = NULL;
+      preg->allocated = 0;
+      return err;
+    }
+#ifdef DEBUG
+  /* Note: length+1 will not overflow since it is checked in init_dfa.  */
+  dfa->re_str = re_malloc (char, length + 1);
+  strncpy (dfa->re_str, pattern, length + 1);
+#endif
+
+  __libc_lock_init (dfa->lock);
+
+  err = re_string_construct (&regexp, pattern, length, preg->translate,
+			     syntax & RE_ICASE, dfa);
+  if (BE (err != REG_NOERROR, 0))
+    {
+    re_compile_internal_free_return:
+      free_workarea_compile (preg);
+      re_string_destruct (&regexp);
+      free_dfa_content (dfa);
+      preg->buffer = NULL;
+      preg->allocated = 0;
+      return err;
+    }
+
+  /* Parse the regular expression, and build a structure tree.  */
+  preg->re_nsub = 0;
+  dfa->str_tree = parse (&regexp, preg, syntax, &err);
+  if (BE (dfa->str_tree == NULL, 0))
+    goto re_compile_internal_free_return;
+
+  /* Analyze the tree and create the nfa.  */
+  err = analyze (preg);
+  if (BE (err != REG_NOERROR, 0))
+    goto re_compile_internal_free_return;
+
+#ifdef RE_ENABLE_I18N
+  /* If possible, do searching in single byte encoding to speed things up.  */
+  if (dfa->is_utf8 && !(syntax & RE_ICASE) && preg->translate == NULL)
+    optimize_utf8 (dfa);
+#endif
+
+  /* Then create the initial state of the dfa.  */
+  err = create_initial_state (dfa);
+
+  /* Release work areas.  */
+  free_workarea_compile (preg);
+  re_string_destruct (&regexp);
+
+  if (BE (err != REG_NOERROR, 0))
+    {
+      free_dfa_content (dfa);
+      preg->buffer = NULL;
+      preg->allocated = 0;
+    }
+
+  return err;
+}
+
+/* Initialize DFA.  We use the length of the regular expression PAT_LEN
+   as the initial length of some arrays.  */
+
+static reg_errcode_t
+init_dfa (re_dfa_t *dfa, size_t pat_len)
+{
+  unsigned int table_size;
+#ifndef _LIBC
+  char *codeset_name;
+#endif
+
+  memset (dfa, '\0', sizeof (re_dfa_t));
+
+  /* Force allocation of str_tree_storage the first time.  */
+  dfa->str_tree_storage_idx = BIN_TREE_STORAGE_SIZE;
+
+  /* Avoid overflows.  */
+  if (pat_len == SIZE_MAX)
+    return REG_ESPACE;
+
+  dfa->nodes_alloc = pat_len + 1;
+  dfa->nodes = re_malloc (re_token_t, dfa->nodes_alloc);
+
+  /*  table_size = 2 ^ ceil(log pat_len) */
+  for (table_size = 1; ; table_size <<= 1)
+    if (table_size > pat_len)
+      break;
+
+  dfa->state_table = calloc (sizeof (struct re_state_table_entry), table_size);
+  dfa->state_hash_mask = table_size - 1;
+
+  dfa->mb_cur_max = MB_CUR_MAX;
+#ifdef _LIBC
+  if (dfa->mb_cur_max == 6
+      && strcmp (_NL_CURRENT (LC_CTYPE, _NL_CTYPE_CODESET_NAME), "UTF-8") == 0)
+    dfa->is_utf8 = 1;
+  dfa->map_notascii = (_NL_CURRENT_WORD (LC_CTYPE, _NL_CTYPE_MAP_TO_NONASCII)
+		       != 0);
+#else
+# ifdef HAVE_LANGINFO_CODESET
+  codeset_name = nl_langinfo (CODESET);
+# else
+  codeset_name = getenv ("LC_ALL");
+  if (codeset_name == NULL || codeset_name[0] == '\0')
+    codeset_name = getenv ("LC_CTYPE");
+  if (codeset_name == NULL || codeset_name[0] == '\0')
+    codeset_name = getenv ("LANG");
+  if (codeset_name == NULL)
+    codeset_name = "";
+  else if (strchr (codeset_name, '.') !=  NULL)
+    codeset_name = strchr (codeset_name, '.') + 1;
+# endif
+
+  /* strcasecmp isn't a standard interface. brute force check */
+#if 0
+  if (strcasecmp (codeset_name, "UTF-8") == 0
+      || strcasecmp (codeset_name, "UTF8") == 0)
+    dfa->is_utf8 = 1;
+#else
+  if (   (codeset_name[0] == 'U' || codeset_name[0] == 'u')
+      && (codeset_name[1] == 'T' || codeset_name[1] == 't')
+      && (codeset_name[2] == 'F' || codeset_name[2] == 'f')
+      && (codeset_name[3] == '-'
+          ? codeset_name[4] == '8' && codeset_name[5] == '\0'
+          : codeset_name[3] == '8' && codeset_name[4] == '\0'))
+    dfa->is_utf8 = 1;
+#endif
+
+  /* We check exhaustively in the loop below if this charset is a
+     superset of ASCII.  */
+  dfa->map_notascii = 0;
+#endif
+
+#ifdef RE_ENABLE_I18N
+  if (dfa->mb_cur_max > 1)
+    {
+      if (dfa->is_utf8)
+        {
+#if !defined(__GNUC__) || __GNUC__ < 3
+	  static short utf8_sb_map_inited = 0;
+
+	  if (! utf8_sb_map_inited)
+	    {
+		int i;
+
+	  	utf8_sb_map_inited = 0;
+		for (i = 0; i <= 0x80 / BITSET_WORD_BITS - 1; i++)
+		  utf8_sb_map[i] = BITSET_WORD_MAX;
+	    }
+#endif
+	  dfa->sb_char = (re_bitset_ptr_t) utf8_sb_map;
+	}
+      else
+	{
+	  int i, j, ch;
+
+	  dfa->sb_char = (re_bitset_ptr_t) calloc (sizeof (bitset_t), 1);
+	  if (BE (dfa->sb_char == NULL, 0))
+	    return REG_ESPACE;
+
+	  /* Set the bits corresponding to single byte chars.  */
+	  for (i = 0, ch = 0; i < BITSET_WORDS; ++i)
+	    for (j = 0; j < BITSET_WORD_BITS; ++j, ++ch)
+	      {
+		wint_t wch = __btowc (ch);
+		if (wch != WEOF)
+		  dfa->sb_char[i] |= (bitset_word_t) 1 << j;
+# ifndef _LIBC
+		if (isascii (ch) && wch != ch)
+		  dfa->map_notascii = 1;
+# endif
+	      }
+	}
+    }
+#endif
+
+  if (BE (dfa->nodes == NULL || dfa->state_table == NULL, 0))
+    return REG_ESPACE;
+  return REG_NOERROR;
+}
+
+/* Initialize WORD_CHAR table, which indicate which character is
+   "word".  In this case "word" means that it is the word construction
+   character used by some operators like "\<", "\>", etc.  */
+
+static void
+internal_function
+init_word_char (re_dfa_t *dfa)
+{
+  int i, j, ch;
+  dfa->word_ops_used = 1;
+  for (i = 0, ch = 0; i < BITSET_WORDS; ++i)
+    for (j = 0; j < BITSET_WORD_BITS; ++j, ++ch)
+      if (isalnum (ch) || ch == '_')
+	dfa->word_char[i] |= (bitset_word_t) 1 << j;
+}
+
+/* Free the work area which are only used while compiling.  */
+
+static void
+free_workarea_compile (regex_t *preg)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  bin_tree_storage_t *storage, *next;
+  for (storage = dfa->str_tree_storage; storage; storage = next)
+    {
+      next = storage->next;
+      re_free (storage);
+    }
+  dfa->str_tree_storage = NULL;
+  dfa->str_tree_storage_idx = BIN_TREE_STORAGE_SIZE;
+  dfa->str_tree = NULL;
+  re_free (dfa->org_indices);
+  dfa->org_indices = NULL;
+}
+
+/* Create initial states for all contexts.  */
+
+static reg_errcode_t
+create_initial_state (re_dfa_t *dfa)
+{
+  int first, i;
+  reg_errcode_t err;
+  re_node_set init_nodes;
+
+  /* Initial states have the epsilon closure of the node which is
+     the first node of the regular expression.  */
+  first = dfa->str_tree->first->node_idx;
+  dfa->init_node = first;
+  err = re_node_set_init_copy (&init_nodes, dfa->eclosures + first);
+  if (BE (err != REG_NOERROR, 0))
+    return err;
+
+  /* The back-references which are in initial states can epsilon transit,
+     since in this case all of the subexpressions can be null.
+     Then we add epsilon closures of the nodes which are the next nodes of
+     the back-references.  */
+  if (dfa->nbackref > 0)
+    for (i = 0; i < init_nodes.nelem; ++i)
+      {
+	int node_idx = init_nodes.elems[i];
+	re_token_type_t type = dfa->nodes[node_idx].type;
+
+	int clexp_idx;
+	if (type != OP_BACK_REF)
+	  continue;
+	for (clexp_idx = 0; clexp_idx < init_nodes.nelem; ++clexp_idx)
+	  {
+	    re_token_t *clexp_node;
+	    clexp_node = dfa->nodes + init_nodes.elems[clexp_idx];
+	    if (clexp_node->type == OP_CLOSE_SUBEXP
+		&& clexp_node->opr.idx == dfa->nodes[node_idx].opr.idx)
+	      break;
+	  }
+	if (clexp_idx == init_nodes.nelem)
+	  continue;
+
+	if (type == OP_BACK_REF)
+	  {
+	    int dest_idx = dfa->edests[node_idx].elems[0];
+	    if (!re_node_set_contains (&init_nodes, dest_idx))
+	      {
+		reg_errcode_t err = re_node_set_merge (&init_nodes,
+						       dfa->eclosures
+						       + dest_idx);
+		if (err != REG_NOERROR)
+		  return err;
+		i = 0;
+	      }
+	  }
+      }
+
+  /* It must be the first time to invoke acquire_state.  */
+  dfa->init_state = re_acquire_state_context (&err, dfa, &init_nodes, 0);
+  /* We don't check ERR here, since the initial state must not be NULL.  */
+  if (BE (dfa->init_state == NULL, 0))
+    return err;
+  if (dfa->init_state->has_constraint)
+    {
+      dfa->init_state_word = re_acquire_state_context (&err, dfa, &init_nodes,
+						       CONTEXT_WORD);
+      dfa->init_state_nl = re_acquire_state_context (&err, dfa, &init_nodes,
+						     CONTEXT_NEWLINE);
+      dfa->init_state_begbuf = re_acquire_state_context (&err, dfa,
+							 &init_nodes,
+							 CONTEXT_NEWLINE
+							 | CONTEXT_BEGBUF);
+      if (BE (dfa->init_state_word == NULL || dfa->init_state_nl == NULL
+	      || dfa->init_state_begbuf == NULL, 0))
+	return err;
+    }
+  else
+    dfa->init_state_word = dfa->init_state_nl
+      = dfa->init_state_begbuf = dfa->init_state;
+
+  re_node_set_free (&init_nodes);
+  return REG_NOERROR;
+}
+
+#ifdef RE_ENABLE_I18N
+/* If it is possible to do searching in single byte encoding instead of UTF-8
+   to speed things up, set dfa->mb_cur_max to 1, clear is_utf8 and change
+   DFA nodes where needed.  */
+
+static void
+optimize_utf8 (re_dfa_t *dfa)
+{
+  int node, i, mb_chars = 0, has_period = 0;
+
+  for (node = 0; node < dfa->nodes_len; ++node)
+    switch (dfa->nodes[node].type)
+      {
+      case CHARACTER:
+	if (dfa->nodes[node].opr.c >= 0x80)
+	  mb_chars = 1;
+	break;
+      case ANCHOR:
+	switch (dfa->nodes[node].opr.ctx_type)
+	  {
+	  case LINE_FIRST:
+	  case LINE_LAST:
+	  case BUF_FIRST:
+	  case BUF_LAST:
+	    break;
+	  default:
+	    /* Word anchors etc. cannot be handled.  It's okay to test
+	       opr.ctx_type since constraints (for all DFA nodes) are
+	       created by ORing one or more opr.ctx_type values.  */
+	    return;
+	  }
+	break;
+      case OP_PERIOD:
+	has_period = 1;
+	break;
+      case OP_BACK_REF:
+      case OP_ALT:
+      case END_OF_RE:
+      case OP_DUP_ASTERISK:
+      case OP_OPEN_SUBEXP:
+      case OP_CLOSE_SUBEXP:
+	break;
+      case COMPLEX_BRACKET:
+	return;
+      case SIMPLE_BRACKET:
+	/* Just double check.  The non-ASCII range starts at 0x80.  */
+	assert (0x80 % BITSET_WORD_BITS == 0);
+	for (i = 0x80 / BITSET_WORD_BITS; i < BITSET_WORDS; ++i)
+	  if (dfa->nodes[node].opr.sbcset[i])
+	    return;
+	break;
+      default:
+	abort ();
+      }
+
+  if (mb_chars || has_period)
+    for (node = 0; node < dfa->nodes_len; ++node)
+      {
+	if (dfa->nodes[node].type == CHARACTER
+	    && dfa->nodes[node].opr.c >= 0x80)
+	  dfa->nodes[node].mb_partial = 0;
+	else if (dfa->nodes[node].type == OP_PERIOD)
+	  dfa->nodes[node].type = OP_UTF8_PERIOD;
+      }
+
+  /* The search can be in single byte locale.  */
+  dfa->mb_cur_max = 1;
+  dfa->is_utf8 = 0;
+  dfa->has_mb_node = dfa->nbackref > 0 || has_period;
+}
+#endif
+
+/* Analyze the structure tree, and calculate "first", "next", "edest",
+   "eclosure", and "inveclosure".  */
+
+static reg_errcode_t
+analyze (regex_t *preg)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  reg_errcode_t ret;
+
+  /* Allocate arrays.  */
+  dfa->nexts = re_malloc (int, dfa->nodes_alloc);
+  dfa->org_indices = re_malloc (int, dfa->nodes_alloc);
+  dfa->edests = re_malloc (re_node_set, dfa->nodes_alloc);
+  dfa->eclosures = re_malloc (re_node_set, dfa->nodes_alloc);
+  if (BE (dfa->nexts == NULL || dfa->org_indices == NULL || dfa->edests == NULL
+	  || dfa->eclosures == NULL, 0))
+    return REG_ESPACE;
+
+  dfa->subexp_map = re_malloc (int, preg->re_nsub);
+  if (dfa->subexp_map != NULL)
+    {
+      int i;
+      for (i = 0; i < preg->re_nsub; i++)
+	dfa->subexp_map[i] = i;
+      preorder (dfa->str_tree, optimize_subexps, dfa);
+      for (i = 0; i < preg->re_nsub; i++)
+	if (dfa->subexp_map[i] != i)
+	  break;
+      if (i == preg->re_nsub)
+	{
+	  free (dfa->subexp_map);
+	  dfa->subexp_map = NULL;
+	}
+    }
+
+  ret = postorder (dfa->str_tree, lower_subexps, preg);
+  if (BE (ret != REG_NOERROR, 0))
+    return ret;
+  ret = postorder (dfa->str_tree, calc_first, dfa);
+  if (BE (ret != REG_NOERROR, 0))
+    return ret;
+  preorder (dfa->str_tree, calc_next, dfa);
+  ret = preorder (dfa->str_tree, link_nfa_nodes, dfa);
+  if (BE (ret != REG_NOERROR, 0))
+    return ret;
+  ret = calc_eclosure (dfa);
+  if (BE (ret != REG_NOERROR, 0))
+    return ret;
+
+  /* We only need this during the prune_impossible_nodes pass in regexec.c;
+     skip it if p_i_n will not run, as calc_inveclosure can be quadratic.  */
+  if ((!preg->no_sub && preg->re_nsub > 0 && dfa->has_plural_match)
+      || dfa->nbackref)
+    {
+      dfa->inveclosures = re_malloc (re_node_set, dfa->nodes_len);
+      if (BE (dfa->inveclosures == NULL, 0))
+	return REG_ESPACE;
+      ret = calc_inveclosure (dfa);
+    }
+
+  return ret;
+}
+
+/* Our parse trees are very unbalanced, so we cannot use a stack to
+   implement parse tree visits.  Instead, we use parent pointers and
+   some hairy code in these two functions.  */
+static reg_errcode_t
+postorder (bin_tree_t *root, reg_errcode_t (fn (void *, bin_tree_t *)),
+	   void *extra)
+{
+  bin_tree_t *node, *prev;
+
+  for (node = root; ; )
+    {
+      /* Descend down the tree, preferably to the left (or to the right
+	 if that's the only child).  */
+      while (node->left || node->right)
+	if (node->left)
+	  node = node->left;
+	else
+	  node = node->right;
+
+      do
+	{
+	  reg_errcode_t err = fn (extra, node);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	  if (node->parent == NULL)
+	    return REG_NOERROR;
+	  prev = node;
+	  node = node->parent;
+	}
+      /* Go up while we have a node that is reached from the right.  */
+      while (node->right == prev || node->right == NULL);
+      node = node->right;
+    }
+}
+
+static reg_errcode_t
+preorder (bin_tree_t *root, reg_errcode_t (fn (void *, bin_tree_t *)),
+	  void *extra)
+{
+  bin_tree_t *node;
+
+  for (node = root; ; )
+    {
+      reg_errcode_t err = fn (extra, node);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+
+      /* Go to the left node, or up and to the right.  */
+      if (node->left)
+	node = node->left;
+      else
+	{
+	  bin_tree_t *prev = NULL;
+	  while (node->right == prev || node->right == NULL)
+	    {
+	      prev = node;
+	      node = node->parent;
+	      if (!node)
+		return REG_NOERROR;
+	    }
+	  node = node->right;
+	}
+    }
+}
+
+/* Optimization pass: if a SUBEXP is entirely contained, strip it and tell
+   re_search_internal to map the inner one's opr.idx to this one's.  Adjust
+   backreferences as well.  Requires a preorder visit.  */
+static reg_errcode_t
+optimize_subexps (void *extra, bin_tree_t *node)
+{
+  re_dfa_t *dfa = (re_dfa_t *) extra;
+
+  if (node->token.type == OP_BACK_REF && dfa->subexp_map)
+    {
+      int idx = node->token.opr.idx;
+      node->token.opr.idx = dfa->subexp_map[idx];
+      dfa->used_bkref_map |= 1 << node->token.opr.idx;
+    }
+
+  else if (node->token.type == SUBEXP
+	   && node->left && node->left->token.type == SUBEXP)
+    {
+      int other_idx = node->left->token.opr.idx;
+
+      node->left = node->left->left;
+      if (node->left)
+	node->left->parent = node;
+
+      dfa->subexp_map[other_idx] = dfa->subexp_map[node->token.opr.idx];
+      if (other_idx < BITSET_WORD_BITS)
+	  dfa->used_bkref_map &= ~((bitset_word_t) 1 << other_idx);
+    }
+
+  return REG_NOERROR;
+}
+
+/* Lowering pass: Turn each SUBEXP node into the appropriate concatenation
+   of OP_OPEN_SUBEXP, the body of the SUBEXP (if any) and OP_CLOSE_SUBEXP.  */
+static reg_errcode_t
+lower_subexps (void *extra, bin_tree_t *node)
+{
+  regex_t *preg = (regex_t *) extra;
+  reg_errcode_t err = REG_NOERROR;
+
+  if (node->left && node->left->token.type == SUBEXP)
+    {
+      node->left = lower_subexp (&err, preg, node->left);
+      if (node->left)
+	node->left->parent = node;
+    }
+  if (node->right && node->right->token.type == SUBEXP)
+    {
+      node->right = lower_subexp (&err, preg, node->right);
+      if (node->right)
+	node->right->parent = node;
+    }
+
+  return err;
+}
+
+static bin_tree_t *
+lower_subexp (reg_errcode_t *err, regex_t *preg, bin_tree_t *node)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  bin_tree_t *body = node->left;
+  bin_tree_t *op, *cls, *tree1, *tree;
+
+  if (preg->no_sub
+      /* We do not optimize empty subexpressions, because otherwise we may
+	 have bad CONCAT nodes with NULL children.  This is obviously not
+	 very common, so we do not lose much.  An example that triggers
+	 this case is the sed "script" /\(\)/x.  */
+      && node->left != NULL
+      && (node->token.opr.idx >= BITSET_WORD_BITS
+	  || !(dfa->used_bkref_map
+	       & ((bitset_word_t) 1 << node->token.opr.idx))))
+    return node->left;
+
+  /* Convert the SUBEXP node to the concatenation of an
+     OP_OPEN_SUBEXP, the contents, and an OP_CLOSE_SUBEXP.  */
+  op = create_tree (dfa, NULL, NULL, OP_OPEN_SUBEXP);
+  cls = create_tree (dfa, NULL, NULL, OP_CLOSE_SUBEXP);
+  tree1 = body ? create_tree (dfa, body, cls, CONCAT) : cls;
+  tree = create_tree (dfa, op, tree1, CONCAT);
+  if (BE (tree == NULL || tree1 == NULL || op == NULL || cls == NULL, 0))
+    {
+      *err = REG_ESPACE;
+      return NULL;
+    }
+
+  op->token.opr.idx = cls->token.opr.idx = node->token.opr.idx;
+  op->token.opt_subexp = cls->token.opt_subexp = node->token.opt_subexp;
+  return tree;
+}
+
+/* Pass 1 in building the NFA: compute FIRST and create unlinked automaton
+   nodes.  Requires a postorder visit.  */
+static reg_errcode_t
+calc_first (void *extra, bin_tree_t *node)
+{
+  re_dfa_t *dfa = (re_dfa_t *) extra;
+  if (node->token.type == CONCAT)
+    {
+      node->first = node->left->first;
+      node->node_idx = node->left->node_idx;
+    }
+  else
+    {
+      node->first = node;
+      node->node_idx = re_dfa_add_node (dfa, node->token);
+      if (BE (node->node_idx == -1, 0))
+	return REG_ESPACE;
+      if (node->token.type == ANCHOR)
+	dfa->nodes[node->node_idx].constraint = node->token.opr.ctx_type;
+    }
+  return REG_NOERROR;
+}
+
+/* Pass 2: compute NEXT on the tree.  Preorder visit.  */
+static reg_errcode_t
+calc_next (void *extra, bin_tree_t *node)
+{
+  switch (node->token.type)
+    {
+    case OP_DUP_ASTERISK:
+      node->left->next = node;
+      break;
+    case CONCAT:
+      node->left->next = node->right->first;
+      node->right->next = node->next;
+      break;
+    default:
+      if (node->left)
+	node->left->next = node->next;
+      if (node->right)
+	node->right->next = node->next;
+      break;
+    }
+  return REG_NOERROR;
+}
+
+/* Pass 3: link all DFA nodes to their NEXT node (any order will do).  */
+static reg_errcode_t
+link_nfa_nodes (void *extra, bin_tree_t *node)
+{
+  re_dfa_t *dfa = (re_dfa_t *) extra;
+  int idx = node->node_idx;
+  reg_errcode_t err = REG_NOERROR;
+
+  switch (node->token.type)
+    {
+    case CONCAT:
+      break;
+
+    case END_OF_RE:
+      assert (node->next == NULL);
+      break;
+
+    case OP_DUP_ASTERISK:
+    case OP_ALT:
+      {
+	int left, right;
+	dfa->has_plural_match = 1;
+	if (node->left != NULL)
+	  left = node->left->first->node_idx;
+	else
+	  left = node->next->node_idx;
+	if (node->right != NULL)
+	  right = node->right->first->node_idx;
+	else
+	  right = node->next->node_idx;
+	assert (left > -1);
+	assert (right > -1);
+	err = re_node_set_init_2 (dfa->edests + idx, left, right);
+      }
+      break;
+
+    case ANCHOR:
+    case OP_OPEN_SUBEXP:
+    case OP_CLOSE_SUBEXP:
+      err = re_node_set_init_1 (dfa->edests + idx, node->next->node_idx);
+      break;
+
+    case OP_BACK_REF:
+      dfa->nexts[idx] = node->next->node_idx;
+      if (node->token.type == OP_BACK_REF)
+	err = re_node_set_init_1 (dfa->edests + idx, dfa->nexts[idx]);
+      break;
+
+    default:
+      assert (!IS_EPSILON_NODE (node->token.type));
+      dfa->nexts[idx] = node->next->node_idx;
+      break;
+    }
+
+  return err;
+}
+
+/* Duplicate the epsilon closure of the node ROOT_NODE.
+   Note that duplicated nodes have constraint INIT_CONSTRAINT in addition
+   to their own constraint.  */
+
+static reg_errcode_t
+internal_function
+duplicate_node_closure (re_dfa_t *dfa, int top_org_node, int top_clone_node,
+			int root_node, unsigned int init_constraint)
+{
+  int org_node, clone_node, ret;
+  unsigned int constraint = init_constraint;
+  for (org_node = top_org_node, clone_node = top_clone_node;;)
+    {
+      int org_dest, clone_dest;
+      if (dfa->nodes[org_node].type == OP_BACK_REF)
+	{
+	  /* If the back reference epsilon-transit, its destination must
+	     also have the constraint.  Then duplicate the epsilon closure
+	     of the destination of the back reference, and store it in
+	     edests of the back reference.  */
+	  org_dest = dfa->nexts[org_node];
+	  re_node_set_empty (dfa->edests + clone_node);
+	  clone_dest = duplicate_node (dfa, org_dest, constraint);
+	  if (BE (clone_dest == -1, 0))
+	    return REG_ESPACE;
+	  dfa->nexts[clone_node] = dfa->nexts[org_node];
+	  ret = re_node_set_insert (dfa->edests + clone_node, clone_dest);
+	  if (BE (ret < 0, 0))
+	    return REG_ESPACE;
+	}
+      else if (dfa->edests[org_node].nelem == 0)
+	{
+	  /* In case of the node can't epsilon-transit, don't duplicate the
+	     destination and store the original destination as the
+	     destination of the node.  */
+	  dfa->nexts[clone_node] = dfa->nexts[org_node];
+	  break;
+	}
+      else if (dfa->edests[org_node].nelem == 1)
+	{
+	  /* In case of the node can epsilon-transit, and it has only one
+	     destination.  */
+	  org_dest = dfa->edests[org_node].elems[0];
+	  re_node_set_empty (dfa->edests + clone_node);
+	  /* If the node is root_node itself, it means the epsilon clsoure
+	     has a loop.   Then tie it to the destination of the root_node.  */
+	  if (org_node == root_node && clone_node != org_node)
+	    {
+	      ret = re_node_set_insert (dfa->edests + clone_node, org_dest);
+	      if (BE (ret < 0, 0))
+		return REG_ESPACE;
+	      break;
+	    }
+	  /* In case of the node has another constraint, add it.  */
+	  constraint |= dfa->nodes[org_node].constraint;
+	  clone_dest = duplicate_node (dfa, org_dest, constraint);
+	  if (BE (clone_dest == -1, 0))
+	    return REG_ESPACE;
+	  ret = re_node_set_insert (dfa->edests + clone_node, clone_dest);
+	  if (BE (ret < 0, 0))
+	    return REG_ESPACE;
+	}
+      else /* dfa->edests[org_node].nelem == 2 */
+	{
+	  /* In case of the node can epsilon-transit, and it has two
+	     destinations. In the bin_tree_t and DFA, that's '|' and '*'.   */
+	  org_dest = dfa->edests[org_node].elems[0];
+	  re_node_set_empty (dfa->edests + clone_node);
+	  /* Search for a duplicated node which satisfies the constraint.  */
+	  clone_dest = search_duplicated_node (dfa, org_dest, constraint);
+	  if (clone_dest == -1)
+	    {
+	      /* There is no such duplicated node, create a new one.  */
+	      reg_errcode_t err;
+	      clone_dest = duplicate_node (dfa, org_dest, constraint);
+	      if (BE (clone_dest == -1, 0))
+		return REG_ESPACE;
+	      ret = re_node_set_insert (dfa->edests + clone_node, clone_dest);
+	      if (BE (ret < 0, 0))
+		return REG_ESPACE;
+	      err = duplicate_node_closure (dfa, org_dest, clone_dest,
+					    root_node, constraint);
+	      if (BE (err != REG_NOERROR, 0))
+		return err;
+	    }
+	  else
+	    {
+	      /* There is a duplicated node which satisfies the constraint,
+		 use it to avoid infinite loop.  */
+	      ret = re_node_set_insert (dfa->edests + clone_node, clone_dest);
+	      if (BE (ret < 0, 0))
+		return REG_ESPACE;
+	    }
+
+	  org_dest = dfa->edests[org_node].elems[1];
+	  clone_dest = duplicate_node (dfa, org_dest, constraint);
+	  if (BE (clone_dest == -1, 0))
+	    return REG_ESPACE;
+	  ret = re_node_set_insert (dfa->edests + clone_node, clone_dest);
+	  if (BE (ret < 0, 0))
+	    return REG_ESPACE;
+	}
+      org_node = org_dest;
+      clone_node = clone_dest;
+    }
+  return REG_NOERROR;
+}
+
+/* Search for a node which is duplicated from the node ORG_NODE, and
+   satisfies the constraint CONSTRAINT.  */
+
+static int
+search_duplicated_node (const re_dfa_t *dfa, int org_node,
+			unsigned int constraint)
+{
+  int idx;
+  for (idx = dfa->nodes_len - 1; dfa->nodes[idx].duplicated && idx > 0; --idx)
+    {
+      if (org_node == dfa->org_indices[idx]
+	  && constraint == dfa->nodes[idx].constraint)
+	return idx; /* Found.  */
+    }
+  return -1; /* Not found.  */
+}
+
+/* Duplicate the node whose index is ORG_IDX and set the constraint CONSTRAINT.
+   Return the index of the new node, or -1 if insufficient storage is
+   available.  */
+
+static int
+duplicate_node (re_dfa_t *dfa, int org_idx, unsigned int constraint)
+{
+  int dup_idx = re_dfa_add_node (dfa, dfa->nodes[org_idx]);
+  if (BE (dup_idx != -1, 1))
+    {
+      dfa->nodes[dup_idx].constraint = constraint;
+      dfa->nodes[dup_idx].constraint |= dfa->nodes[org_idx].constraint;
+      dfa->nodes[dup_idx].duplicated = 1;
+
+      /* Store the index of the original node.  */
+      dfa->org_indices[dup_idx] = org_idx;
+    }
+  return dup_idx;
+}
+
+static reg_errcode_t
+calc_inveclosure (re_dfa_t *dfa)
+{
+  int src, idx, ret;
+  for (idx = 0; idx < dfa->nodes_len; ++idx)
+    re_node_set_init_empty (dfa->inveclosures + idx);
+
+  for (src = 0; src < dfa->nodes_len; ++src)
+    {
+      int *elems = dfa->eclosures[src].elems;
+      for (idx = 0; idx < dfa->eclosures[src].nelem; ++idx)
+	{
+	  ret = re_node_set_insert_last (dfa->inveclosures + elems[idx], src);
+	  if (BE (ret == -1, 0))
+	    return REG_ESPACE;
+	}
+    }
+
+  return REG_NOERROR;
+}
+
+/* Calculate "eclosure" for all the node in DFA.  */
+
+static reg_errcode_t
+calc_eclosure (re_dfa_t *dfa)
+{
+  int node_idx, incomplete;
+#ifdef DEBUG
+  assert (dfa->nodes_len > 0);
+#endif
+  incomplete = 0;
+  /* For each nodes, calculate epsilon closure.  */
+  for (node_idx = 0; ; ++node_idx)
+    {
+      reg_errcode_t err;
+      re_node_set eclosure_elem;
+      if (node_idx == dfa->nodes_len)
+	{
+	  if (!incomplete)
+	    break;
+	  incomplete = 0;
+	  node_idx = 0;
+	}
+
+#ifdef DEBUG
+      assert (dfa->eclosures[node_idx].nelem != -1);
+#endif
+
+      /* If we have already calculated, skip it.  */
+      if (dfa->eclosures[node_idx].nelem != 0)
+	continue;
+      /* Calculate epsilon closure of `node_idx'.  */
+      err = calc_eclosure_iter (&eclosure_elem, dfa, node_idx, 1);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+
+      if (dfa->eclosures[node_idx].nelem == 0)
+	{
+	  incomplete = 1;
+	  re_node_set_free (&eclosure_elem);
+	}
+    }
+  return REG_NOERROR;
+}
+
+/* Calculate epsilon closure of NODE.  */
+
+static reg_errcode_t
+calc_eclosure_iter (re_node_set *new_set, re_dfa_t *dfa, int node, int root)
+{
+  reg_errcode_t err;
+  int i;
+  re_node_set eclosure;
+  int ret;
+  int incomplete = 0;
+  err = re_node_set_alloc (&eclosure, dfa->edests[node].nelem + 1);
+  if (BE (err != REG_NOERROR, 0))
+    return err;
+
+  /* This indicates that we are calculating this node now.
+     We reference this value to avoid infinite loop.  */
+  dfa->eclosures[node].nelem = -1;
+
+  /* If the current node has constraints, duplicate all nodes
+     since they must inherit the constraints.  */
+  if (dfa->nodes[node].constraint
+      && dfa->edests[node].nelem
+      && !dfa->nodes[dfa->edests[node].elems[0]].duplicated)
+    {
+      err = duplicate_node_closure (dfa, node, node, node,
+				    dfa->nodes[node].constraint);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+    }
+
+  /* Expand each epsilon destination nodes.  */
+  if (IS_EPSILON_NODE(dfa->nodes[node].type))
+    for (i = 0; i < dfa->edests[node].nelem; ++i)
+      {
+	re_node_set eclosure_elem;
+	int edest = dfa->edests[node].elems[i];
+	/* If calculating the epsilon closure of `edest' is in progress,
+	   return intermediate result.  */
+	if (dfa->eclosures[edest].nelem == -1)
+	  {
+	    incomplete = 1;
+	    continue;
+	  }
+	/* If we haven't calculated the epsilon closure of `edest' yet,
+	   calculate now. Otherwise use calculated epsilon closure.  */
+	if (dfa->eclosures[edest].nelem == 0)
+	  {
+	    err = calc_eclosure_iter (&eclosure_elem, dfa, edest, 0);
+	    if (BE (err != REG_NOERROR, 0))
+	      return err;
+	  }
+	else
+	  eclosure_elem = dfa->eclosures[edest];
+	/* Merge the epsilon closure of `edest'.  */
+	err = re_node_set_merge (&eclosure, &eclosure_elem);
+	if (BE (err != REG_NOERROR, 0))
+	  return err;
+	/* If the epsilon closure of `edest' is incomplete,
+	   the epsilon closure of this node is also incomplete.  */
+	if (dfa->eclosures[edest].nelem == 0)
+	  {
+	    incomplete = 1;
+	    re_node_set_free (&eclosure_elem);
+	  }
+      }
+
+  /* An epsilon closure includes itself.  */
+  ret = re_node_set_insert (&eclosure, node);
+  if (BE (ret < 0, 0))
+    return REG_ESPACE;
+  if (incomplete && !root)
+    dfa->eclosures[node].nelem = 0;
+  else
+    dfa->eclosures[node] = eclosure;
+  *new_set = eclosure;
+  return REG_NOERROR;
+}
+
+/* Functions for token which are used in the parser.  */
+
+/* Fetch a token from INPUT.
+   We must not use this function inside bracket expressions.  */
+
+static void
+internal_function
+fetch_token (re_token_t *result, re_string_t *input, reg_syntax_t syntax)
+{
+  re_string_skip_bytes (input, peek_token (result, input, syntax));
+}
+
+/* Peek a token from INPUT, and return the length of the token.
+   We must not use this function inside bracket expressions.  */
+
+static int
+internal_function
+peek_token (re_token_t *token, re_string_t *input, reg_syntax_t syntax)
+{
+  unsigned char c;
+
+  if (re_string_eoi (input))
+    {
+      token->type = END_OF_RE;
+      return 0;
+    }
+
+  c = re_string_peek_byte (input, 0);
+  token->opr.c = c;
+
+  token->word_char = 0;
+#ifdef RE_ENABLE_I18N
+  token->mb_partial = 0;
+  if (input->mb_cur_max > 1 &&
+      !re_string_first_byte (input, re_string_cur_idx (input)))
+    {
+      token->type = CHARACTER;
+      token->mb_partial = 1;
+      return 1;
+    }
+#endif
+  if (c == '\\')
+    {
+      unsigned char c2;
+      if (re_string_cur_idx (input) + 1 >= re_string_length (input))
+	{
+	  token->type = BACK_SLASH;
+	  return 1;
+	}
+
+      c2 = re_string_peek_byte_case (input, 1);
+      token->opr.c = c2;
+      token->type = CHARACTER;
+#ifdef RE_ENABLE_I18N
+      if (input->mb_cur_max > 1)
+	{
+	  wint_t wc = re_string_wchar_at (input,
+					  re_string_cur_idx (input) + 1);
+	  token->word_char = IS_WIDE_WORD_CHAR (wc) != 0;
+	}
+      else
+#endif
+	token->word_char = IS_WORD_CHAR (c2) != 0;
+
+      switch (c2)
+	{
+	case '|':
+	  if (!(syntax & RE_LIMITED_OPS) && !(syntax & RE_NO_BK_VBAR))
+	    token->type = OP_ALT;
+	  break;
+	case '1': case '2': case '3': case '4': case '5':
+	case '6': case '7': case '8': case '9':
+	  if (!(syntax & RE_NO_BK_REFS))
+	    {
+	      token->type = OP_BACK_REF;
+	      token->opr.idx = c2 - '1';
+	    }
+	  break;
+	case '<':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    {
+	      token->type = ANCHOR;
+	      token->opr.ctx_type = WORD_FIRST;
+	    }
+	  break;
+	case '>':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    {
+	      token->type = ANCHOR;
+	      token->opr.ctx_type = WORD_LAST;
+	    }
+	  break;
+	case 'b':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    {
+	      token->type = ANCHOR;
+	      token->opr.ctx_type = WORD_DELIM;
+	    }
+	  break;
+	case 'B':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    {
+	      token->type = ANCHOR;
+	      token->opr.ctx_type = NOT_WORD_DELIM;
+	    }
+	  break;
+	case 'w':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    token->type = OP_WORD;
+	  break;
+	case 'W':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    token->type = OP_NOTWORD;
+	  break;
+	case 's':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    token->type = OP_SPACE;
+	  break;
+	case 'S':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    token->type = OP_NOTSPACE;
+	  break;
+	case '`':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    {
+	      token->type = ANCHOR;
+	      token->opr.ctx_type = BUF_FIRST;
+	    }
+	  break;
+	case '\'':
+	  if (!(syntax & RE_NO_GNU_OPS))
+	    {
+	      token->type = ANCHOR;
+	      token->opr.ctx_type = BUF_LAST;
+	    }
+	  break;
+	case '(':
+	  if (!(syntax & RE_NO_BK_PARENS))
+	    token->type = OP_OPEN_SUBEXP;
+	  break;
+	case ')':
+	  if (!(syntax & RE_NO_BK_PARENS))
+	    token->type = OP_CLOSE_SUBEXP;
+	  break;
+	case '+':
+	  if (!(syntax & RE_LIMITED_OPS) && (syntax & RE_BK_PLUS_QM))
+	    token->type = OP_DUP_PLUS;
+	  break;
+	case '?':
+	  if (!(syntax & RE_LIMITED_OPS) && (syntax & RE_BK_PLUS_QM))
+	    token->type = OP_DUP_QUESTION;
+	  break;
+	case '{':
+	  if ((syntax & RE_INTERVALS) && (!(syntax & RE_NO_BK_BRACES)))
+	    token->type = OP_OPEN_DUP_NUM;
+	  break;
+	case '}':
+	  if ((syntax & RE_INTERVALS) && (!(syntax & RE_NO_BK_BRACES)))
+	    token->type = OP_CLOSE_DUP_NUM;
+	  break;
+	default:
+	  break;
+	}
+      return 2;
+    }
+
+  token->type = CHARACTER;
+#ifdef RE_ENABLE_I18N
+  if (input->mb_cur_max > 1)
+    {
+      wint_t wc = re_string_wchar_at (input, re_string_cur_idx (input));
+      token->word_char = IS_WIDE_WORD_CHAR (wc) != 0;
+    }
+  else
+#endif
+    token->word_char = IS_WORD_CHAR (token->opr.c);
+
+  switch (c)
+    {
+    case '\n':
+      if (syntax & RE_NEWLINE_ALT)
+	token->type = OP_ALT;
+      break;
+    case '|':
+      if (!(syntax & RE_LIMITED_OPS) && (syntax & RE_NO_BK_VBAR))
+	token->type = OP_ALT;
+      break;
+    case '*':
+      token->type = OP_DUP_ASTERISK;
+      break;
+    case '+':
+      if (!(syntax & RE_LIMITED_OPS) && !(syntax & RE_BK_PLUS_QM))
+	token->type = OP_DUP_PLUS;
+      break;
+    case '?':
+      if (!(syntax & RE_LIMITED_OPS) && !(syntax & RE_BK_PLUS_QM))
+	token->type = OP_DUP_QUESTION;
+      break;
+    case '{':
+      if ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
+	token->type = OP_OPEN_DUP_NUM;
+      break;
+    case '}':
+      if ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
+	token->type = OP_CLOSE_DUP_NUM;
+      break;
+    case '(':
+      if (syntax & RE_NO_BK_PARENS)
+	token->type = OP_OPEN_SUBEXP;
+      break;
+    case ')':
+      if (syntax & RE_NO_BK_PARENS)
+	token->type = OP_CLOSE_SUBEXP;
+      break;
+    case '[':
+      token->type = OP_OPEN_BRACKET;
+      break;
+    case '.':
+      token->type = OP_PERIOD;
+      break;
+    case '^':
+      if (!(syntax & (RE_CONTEXT_INDEP_ANCHORS | RE_CARET_ANCHORS_HERE)) &&
+	  re_string_cur_idx (input) != 0)
+	{
+	  char prev = re_string_peek_byte (input, -1);
+	  if (!(syntax & RE_NEWLINE_ALT) || prev != '\n')
+	    break;
+	}
+      token->type = ANCHOR;
+      token->opr.ctx_type = LINE_FIRST;
+      break;
+    case '$':
+      if (!(syntax & RE_CONTEXT_INDEP_ANCHORS) &&
+	  re_string_cur_idx (input) + 1 != re_string_length (input))
+	{
+	  re_token_t next;
+	  re_string_skip_bytes (input, 1);
+	  peek_token (&next, input, syntax);
+	  re_string_skip_bytes (input, -1);
+	  if (next.type != OP_ALT && next.type != OP_CLOSE_SUBEXP)
+	    break;
+	}
+      token->type = ANCHOR;
+      token->opr.ctx_type = LINE_LAST;
+      break;
+    default:
+      break;
+    }
+  return 1;
+}
+
+/* Peek a token from INPUT, and return the length of the token.
+   We must not use this function out of bracket expressions.  */
+
+static int
+internal_function
+peek_token_bracket (re_token_t *token, re_string_t *input, reg_syntax_t syntax)
+{
+  unsigned char c;
+  if (re_string_eoi (input))
+    {
+      token->type = END_OF_RE;
+      return 0;
+    }
+  c = re_string_peek_byte (input, 0);
+  token->opr.c = c;
+
+#ifdef RE_ENABLE_I18N
+  if (input->mb_cur_max > 1 &&
+      !re_string_first_byte (input, re_string_cur_idx (input)))
+    {
+      token->type = CHARACTER;
+      return 1;
+    }
+#endif /* RE_ENABLE_I18N */
+
+  if (c == '\\' && (syntax & RE_BACKSLASH_ESCAPE_IN_LISTS)
+      && re_string_cur_idx (input) + 1 < re_string_length (input))
+    {
+      /* In this case, '\' escape a character.  */
+      unsigned char c2;
+      re_string_skip_bytes (input, 1);
+      c2 = re_string_peek_byte (input, 0);
+      token->opr.c = c2;
+      token->type = CHARACTER;
+      return 1;
+    }
+  if (c == '[') /* '[' is a special char in a bracket exps.  */
+    {
+      unsigned char c2;
+      int token_len;
+      if (re_string_cur_idx (input) + 1 < re_string_length (input))
+	c2 = re_string_peek_byte (input, 1);
+      else
+	c2 = 0;
+      token->opr.c = c2;
+      token_len = 2;
+      switch (c2)
+	{
+	case '.':
+	  token->type = OP_OPEN_COLL_ELEM;
+	  break;
+	case '=':
+	  token->type = OP_OPEN_EQUIV_CLASS;
+	  break;
+	case ':':
+	  if (syntax & RE_CHAR_CLASSES)
+	    {
+	      token->type = OP_OPEN_CHAR_CLASS;
+	      break;
+	    }
+	  /* else fall through.  */
+	default:
+	  token->type = CHARACTER;
+	  token->opr.c = c;
+	  token_len = 1;
+	  break;
+	}
+      return token_len;
+    }
+  switch (c)
+    {
+    case '-':
+      token->type = OP_CHARSET_RANGE;
+      break;
+    case ']':
+      token->type = OP_CLOSE_BRACKET;
+      break;
+    case '^':
+      token->type = OP_NON_MATCH_LIST;
+      break;
+    default:
+      token->type = CHARACTER;
+    }
+  return 1;
+}
+
+/* Functions for parser.  */
+
+/* Entry point of the parser.
+   Parse the regular expression REGEXP and return the structure tree.
+   If an error has occurred, ERR is set by error code, and return NULL.
+   This function build the following tree, from regular expression <reg_exp>:
+	   CAT
+	   / \
+	  /   \
+   <reg_exp>  EOR
+
+   CAT means concatenation.
+   EOR means end of regular expression.  */
+
+static bin_tree_t *
+parse (re_string_t *regexp, regex_t *preg, reg_syntax_t syntax,
+       reg_errcode_t *err)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  bin_tree_t *tree, *eor, *root;
+  re_token_t current_token;
+  dfa->syntax = syntax;
+  fetch_token (&current_token, regexp, syntax | RE_CARET_ANCHORS_HERE);
+  tree = parse_reg_exp (regexp, preg, &current_token, syntax, 0, err);
+  if (BE (*err != REG_NOERROR && tree == NULL, 0))
+    return NULL;
+  eor = create_tree (dfa, NULL, NULL, END_OF_RE);
+  if (tree != NULL)
+    root = create_tree (dfa, tree, eor, CONCAT);
+  else
+    root = eor;
+  if (BE (eor == NULL || root == NULL, 0))
+    {
+      *err = REG_ESPACE;
+      return NULL;
+    }
+  return root;
+}
+
+/* This function build the following tree, from regular expression
+   <branch1>|<branch2>:
+	   ALT
+	   / \
+	  /   \
+   <branch1> <branch2>
+
+   ALT means alternative, which represents the operator `|'.  */
+
+static bin_tree_t *
+parse_reg_exp (re_string_t *regexp, regex_t *preg, re_token_t *token,
+	       reg_syntax_t syntax, int nest, reg_errcode_t *err)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  bin_tree_t *tree, *branch = NULL;
+  tree = parse_branch (regexp, preg, token, syntax, nest, err);
+  if (BE (*err != REG_NOERROR && tree == NULL, 0))
+    return NULL;
+
+  while (token->type == OP_ALT)
+    {
+      fetch_token (token, regexp, syntax | RE_CARET_ANCHORS_HERE);
+      if (token->type != OP_ALT && token->type != END_OF_RE
+	  && (nest == 0 || token->type != OP_CLOSE_SUBEXP))
+	{
+	  branch = parse_branch (regexp, preg, token, syntax, nest, err);
+	  if (BE (*err != REG_NOERROR && branch == NULL, 0))
+	    return NULL;
+	}
+      else
+	branch = NULL;
+      tree = create_tree (dfa, tree, branch, OP_ALT);
+      if (BE (tree == NULL, 0))
+	{
+	  *err = REG_ESPACE;
+	  return NULL;
+	}
+    }
+  return tree;
+}
+
+/* This function build the following tree, from regular expression
+   <exp1><exp2>:
+	CAT
+	/ \
+       /   \
+   <exp1> <exp2>
+
+   CAT means concatenation.  */
+
+static bin_tree_t *
+parse_branch (re_string_t *regexp, regex_t *preg, re_token_t *token,
+	      reg_syntax_t syntax, int nest, reg_errcode_t *err)
+{
+  bin_tree_t *tree, *exp;
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  tree = parse_expression (regexp, preg, token, syntax, nest, err);
+  if (BE (*err != REG_NOERROR && tree == NULL, 0))
+    return NULL;
+
+  while (token->type != OP_ALT && token->type != END_OF_RE
+	 && (nest == 0 || token->type != OP_CLOSE_SUBEXP))
+    {
+      exp = parse_expression (regexp, preg, token, syntax, nest, err);
+      if (BE (*err != REG_NOERROR && exp == NULL, 0))
+	{
+	  return NULL;
+	}
+      if (tree != NULL && exp != NULL)
+	{
+	  tree = create_tree (dfa, tree, exp, CONCAT);
+	  if (tree == NULL)
+	    {
+	      *err = REG_ESPACE;
+	      return NULL;
+	    }
+	}
+      else if (tree == NULL)
+	tree = exp;
+      /* Otherwise exp == NULL, we don't need to create new tree.  */
+    }
+  return tree;
+}
+
+/* This function build the following tree, from regular expression a*:
+	 *
+	 |
+	 a
+*/
+
+static bin_tree_t *
+parse_expression (re_string_t *regexp, regex_t *preg, re_token_t *token,
+		  reg_syntax_t syntax, int nest, reg_errcode_t *err)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  bin_tree_t *tree;
+  switch (token->type)
+    {
+    case CHARACTER:
+      tree = create_token_tree (dfa, NULL, NULL, token);
+      if (BE (tree == NULL, 0))
+	{
+	  *err = REG_ESPACE;
+	  return NULL;
+	}
+#ifdef RE_ENABLE_I18N
+      if (dfa->mb_cur_max > 1)
+	{
+	  while (!re_string_eoi (regexp)
+		 && !re_string_first_byte (regexp, re_string_cur_idx (regexp)))
+	    {
+	      bin_tree_t *mbc_remain;
+	      fetch_token (token, regexp, syntax);
+	      mbc_remain = create_token_tree (dfa, NULL, NULL, token);
+	      tree = create_tree (dfa, tree, mbc_remain, CONCAT);
+	      if (BE (mbc_remain == NULL || tree == NULL, 0))
+		{
+		  *err = REG_ESPACE;
+		  return NULL;
+		}
+	    }
+	}
+#endif
+      break;
+    case OP_OPEN_SUBEXP:
+      tree = parse_sub_exp (regexp, preg, token, syntax, nest + 1, err);
+      if (BE (*err != REG_NOERROR && tree == NULL, 0))
+	return NULL;
+      break;
+    case OP_OPEN_BRACKET:
+      tree = parse_bracket_exp (regexp, dfa, token, syntax, err);
+      if (BE (*err != REG_NOERROR && tree == NULL, 0))
+	return NULL;
+      break;
+    case OP_BACK_REF:
+      if (!BE (dfa->completed_bkref_map & (1 << token->opr.idx), 1))
+	{
+	  *err = REG_ESUBREG;
+	  return NULL;
+	}
+      dfa->used_bkref_map |= 1 << token->opr.idx;
+      tree = create_token_tree (dfa, NULL, NULL, token);
+      if (BE (tree == NULL, 0))
+	{
+	  *err = REG_ESPACE;
+	  return NULL;
+	}
+      ++dfa->nbackref;
+      dfa->has_mb_node = 1;
+      break;
+    case OP_OPEN_DUP_NUM:
+      if (syntax & RE_CONTEXT_INVALID_DUP)
+	{
+	  *err = REG_BADRPT;
+	  return NULL;
+	}
+      /* FALLTHROUGH */
+    case OP_DUP_ASTERISK:
+    case OP_DUP_PLUS:
+    case OP_DUP_QUESTION:
+      if (syntax & RE_CONTEXT_INVALID_OPS)
+	{
+	  *err = REG_BADRPT;
+	  return NULL;
+	}
+      else if (syntax & RE_CONTEXT_INDEP_OPS)
+	{
+	  fetch_token (token, regexp, syntax);
+	  return parse_expression (regexp, preg, token, syntax, nest, err);
+	}
+      /* else fall through  */
+    case OP_CLOSE_SUBEXP:
+      if ((token->type == OP_CLOSE_SUBEXP) &&
+	  !(syntax & RE_UNMATCHED_RIGHT_PAREN_ORD))
+	{
+	  *err = REG_ERPAREN;
+	  return NULL;
+	}
+      /* else fall through  */
+    case OP_CLOSE_DUP_NUM:
+      /* We treat it as a normal character.  */
+
+      /* Then we can these characters as normal characters.  */
+      token->type = CHARACTER;
+      /* mb_partial and word_char bits should be initialized already
+	 by peek_token.  */
+      tree = create_token_tree (dfa, NULL, NULL, token);
+      if (BE (tree == NULL, 0))
+	{
+	  *err = REG_ESPACE;
+	  return NULL;
+	}
+      break;
+    case ANCHOR:
+      if ((token->opr.ctx_type
+	   & (WORD_DELIM | NOT_WORD_DELIM | WORD_FIRST | WORD_LAST))
+	  && dfa->word_ops_used == 0)
+	init_word_char (dfa);
+      if (token->opr.ctx_type == WORD_DELIM
+	  || token->opr.ctx_type == NOT_WORD_DELIM)
+	{
+	  bin_tree_t *tree_first, *tree_last;
+	  if (token->opr.ctx_type == WORD_DELIM)
+	    {
+	      token->opr.ctx_type = WORD_FIRST;
+	      tree_first = create_token_tree (dfa, NULL, NULL, token);
+	      token->opr.ctx_type = WORD_LAST;
+	    }
+	  else
+	    {
+	      token->opr.ctx_type = INSIDE_WORD;
+	      tree_first = create_token_tree (dfa, NULL, NULL, token);
+	      token->opr.ctx_type = INSIDE_NOTWORD;
+	    }
+	  tree_last = create_token_tree (dfa, NULL, NULL, token);
+	  tree = create_tree (dfa, tree_first, tree_last, OP_ALT);
+	  if (BE (tree_first == NULL || tree_last == NULL || tree == NULL, 0))
+	    {
+	      *err = REG_ESPACE;
+	      return NULL;
+	    }
+	}
+      else
+	{
+	  tree = create_token_tree (dfa, NULL, NULL, token);
+	  if (BE (tree == NULL, 0))
+	    {
+	      *err = REG_ESPACE;
+	      return NULL;
+	    }
+	}
+      /* We must return here, since ANCHORs can't be followed
+	 by repetition operators.
+	 eg. RE"^*" is invalid or "<ANCHOR(^)><CHAR(*)>",
+	     it must not be "<ANCHOR(^)><REPEAT(*)>".  */
+      fetch_token (token, regexp, syntax);
+      return tree;
+    case OP_PERIOD:
+      tree = create_token_tree (dfa, NULL, NULL, token);
+      if (BE (tree == NULL, 0))
+	{
+	  *err = REG_ESPACE;
+	  return NULL;
+	}
+      if (dfa->mb_cur_max > 1)
+	dfa->has_mb_node = 1;
+      break;
+    case OP_WORD:
+    case OP_NOTWORD:
+      tree = build_charclass_op (dfa, regexp->trans,
+				 "alnum",
+				 "_",
+				 token->type == OP_NOTWORD, err);
+      if (BE (*err != REG_NOERROR && tree == NULL, 0))
+	return NULL;
+      break;
+    case OP_SPACE:
+    case OP_NOTSPACE:
+      tree = build_charclass_op (dfa, regexp->trans,
+				 "space",
+				 "",
+				 token->type == OP_NOTSPACE, err);
+      if (BE (*err != REG_NOERROR && tree == NULL, 0))
+	return NULL;
+      break;
+    case OP_ALT:
+    case END_OF_RE:
+      return NULL;
+    case BACK_SLASH:
+      *err = REG_EESCAPE;
+      return NULL;
+    default:
+      /* Must not happen?  */
+#ifdef DEBUG
+      assert (0);
+#endif
+      return NULL;
+    }
+  fetch_token (token, regexp, syntax);
+
+  while (token->type == OP_DUP_ASTERISK || token->type == OP_DUP_PLUS
+	 || token->type == OP_DUP_QUESTION || token->type == OP_OPEN_DUP_NUM)
+    {
+      tree = parse_dup_op (tree, regexp, dfa, token, syntax, err);
+      if (BE (*err != REG_NOERROR && tree == NULL, 0))
+	return NULL;
+      /* In BRE consecutive duplications are not allowed.  */
+      if ((syntax & RE_CONTEXT_INVALID_DUP)
+	  && (token->type == OP_DUP_ASTERISK
+	      || token->type == OP_OPEN_DUP_NUM))
+	{
+	  *err = REG_BADRPT;
+	  return NULL;
+	}
+    }
+
+  return tree;
+}
+
+/* This function build the following tree, from regular expression
+   (<reg_exp>):
+	 SUBEXP
+	    |
+	<reg_exp>
+*/
+
+static bin_tree_t *
+parse_sub_exp (re_string_t *regexp, regex_t *preg, re_token_t *token,
+	       reg_syntax_t syntax, int nest, reg_errcode_t *err)
+{
+  re_dfa_t *dfa = (re_dfa_t *) preg->buffer;
+  bin_tree_t *tree;
+  size_t cur_nsub;
+  cur_nsub = preg->re_nsub++;
+
+  fetch_token (token, regexp, syntax | RE_CARET_ANCHORS_HERE);
+
+  /* The subexpression may be a null string.  */
+  if (token->type == OP_CLOSE_SUBEXP)
+    tree = NULL;
+  else
+    {
+      tree = parse_reg_exp (regexp, preg, token, syntax, nest, err);
+      if (BE (*err == REG_NOERROR && token->type != OP_CLOSE_SUBEXP, 0))
+	*err = REG_EPAREN;
+      if (BE (*err != REG_NOERROR, 0))
+	return NULL;
+    }
+
+  if (cur_nsub <= '9' - '1')
+    dfa->completed_bkref_map |= 1 << cur_nsub;
+
+  tree = create_tree (dfa, tree, NULL, SUBEXP);
+  if (BE (tree == NULL, 0))
+    {
+      *err = REG_ESPACE;
+      return NULL;
+    }
+  tree->token.opr.idx = cur_nsub;
+  return tree;
+}
+
+/* This function parse repetition operators like "*", "+", "{1,3}" etc.  */
+
+static bin_tree_t *
+parse_dup_op (bin_tree_t *elem, re_string_t *regexp, re_dfa_t *dfa,
+	      re_token_t *token, reg_syntax_t syntax, reg_errcode_t *err)
+{
+  bin_tree_t *tree = NULL, *old_tree = NULL;
+  int i, start, end, start_idx = re_string_cur_idx (regexp);
+#ifndef RE_TOKEN_INIT_BUG
+  re_token_t start_token = *token;
+#else
+  re_token_t start_token;
+
+  memcpy ((void *) &start_token, (void *) token, sizeof start_token);
+#endif
+
+  if (token->type == OP_OPEN_DUP_NUM)
+    {
+      end = 0;
+      start = fetch_number (regexp, token, syntax);
+      if (start == -1)
+	{
+	  if (token->type == CHARACTER && token->opr.c == ',')
+	    start = 0; /* We treat "{,m}" as "{0,m}".  */
+	  else
+	    {
+	      *err = REG_BADBR; /* <re>{} is invalid.  */
+	      return NULL;
+	    }
+	}
+      if (BE (start != -2, 1))
+	{
+	  /* We treat "{n}" as "{n,n}".  */
+	  end = ((token->type == OP_CLOSE_DUP_NUM) ? start
+		 : ((token->type == CHARACTER && token->opr.c == ',')
+		    ? fetch_number (regexp, token, syntax) : -2));
+	}
+      if (BE (start == -2 || end == -2, 0))
+	{
+	  /* Invalid sequence.  */
+	  if (BE (!(syntax & RE_INVALID_INTERVAL_ORD), 0))
+	    {
+	      if (token->type == END_OF_RE)
+		*err = REG_EBRACE;
+	      else
+		*err = REG_BADBR;
+
+	      return NULL;
+	    }
+
+	  /* If the syntax bit is set, rollback.  */
+	  re_string_set_index (regexp, start_idx);
+	  *token = start_token;
+	  token->type = CHARACTER;
+	  /* mb_partial and word_char bits should be already initialized by
+	     peek_token.  */
+	  return elem;
+	}
+
+      if (BE ((end != -1 && start > end) || token->type != OP_CLOSE_DUP_NUM, 0))
+	{
+	  /* First number greater than second.  */
+	  *err = REG_BADBR;
+	  return NULL;
+	}
+    }
+  else
+    {
+      start = (token->type == OP_DUP_PLUS) ? 1 : 0;
+      end = (token->type == OP_DUP_QUESTION) ? 1 : -1;
+    }
+
+  fetch_token (token, regexp, syntax);
+
+  if (BE (elem == NULL, 0))
+    return NULL;
+  if (BE (start == 0 && end == 0, 0))
+    {
+      postorder (elem, free_tree, NULL);
+      return NULL;
+    }
+
+  /* Extract "<re>{n,m}" to "<re><re>...<re><re>{0,<m-n>}".  */
+  if (BE (start > 0, 0))
+    {
+      tree = elem;
+      for (i = 2; i <= start; ++i)
+	{
+	  elem = duplicate_tree (elem, dfa);
+	  tree = create_tree (dfa, tree, elem, CONCAT);
+	  if (BE (elem == NULL || tree == NULL, 0))
+	    goto parse_dup_op_espace;
+	}
+
+      if (start == end)
+	return tree;
+
+      /* Duplicate ELEM before it is marked optional.  */
+      elem = duplicate_tree (elem, dfa);
+      old_tree = tree;
+    }
+  else
+    old_tree = NULL;
+
+  if (elem->token.type == SUBEXP)
+    postorder (elem, mark_opt_subexp, (void *) (intptr_t) elem->token.opr.idx);
+
+  tree = create_tree (dfa, elem, NULL, (end == -1 ? OP_DUP_ASTERISK : OP_ALT));
+  if (BE (tree == NULL, 0))
+    goto parse_dup_op_espace;
+
+  /* This loop is actually executed only when end != -1,
+     to rewrite <re>{0,n} as (<re>(<re>...<re>?)?)?...  We have
+     already created the start+1-th copy.  */
+  for (i = start + 2; i <= end; ++i)
+    {
+      elem = duplicate_tree (elem, dfa);
+      tree = create_tree (dfa, tree, elem, CONCAT);
+      if (BE (elem == NULL || tree == NULL, 0))
+	goto parse_dup_op_espace;
+
+      tree = create_tree (dfa, tree, NULL, OP_ALT);
+      if (BE (tree == NULL, 0))
+	goto parse_dup_op_espace;
+    }
+
+  if (old_tree)
+    tree = create_tree (dfa, old_tree, tree, CONCAT);
+
+  return tree;
+
+ parse_dup_op_espace:
+  *err = REG_ESPACE;
+  return NULL;
+}
+
+/* Size of the names for collating symbol/equivalence_class/character_class.
+   I'm not sure, but maybe enough.  */
+#define BRACKET_NAME_BUF_SIZE 32
+
+#ifndef _LIBC
+  /* Local function for parse_bracket_exp only used in case of NOT _LIBC.
+     Build the range expression which starts from START_ELEM, and ends
+     at END_ELEM.  The result are written to MBCSET and SBCSET.
+     RANGE_ALLOC is the allocated size of mbcset->range_starts, and
+     mbcset->range_ends, is a pointer argument since we may
+     update it.  */
+
+static reg_errcode_t
+internal_function
+# ifdef RE_ENABLE_I18N
+build_range_exp (bitset_t sbcset, re_charset_t *mbcset, int *range_alloc,
+		 bracket_elem_t *start_elem, bracket_elem_t *end_elem)
+# else /* not RE_ENABLE_I18N */
+build_range_exp (bitset_t sbcset, bracket_elem_t *start_elem,
+		 bracket_elem_t *end_elem)
+# endif /* not RE_ENABLE_I18N */
+{
+  unsigned int start_ch, end_ch;
+  /* Equivalence Classes and Character Classes can't be a range start/end.  */
+  if (BE (start_elem->type == EQUIV_CLASS || start_elem->type == CHAR_CLASS
+	  || end_elem->type == EQUIV_CLASS || end_elem->type == CHAR_CLASS,
+	  0))
+    return REG_ERANGE;
+
+  /* We can handle no multi character collating elements without libc
+     support.  */
+  if (BE ((start_elem->type == COLL_SYM
+	   && strlen ((char *) start_elem->opr.name) > 1)
+	  || (end_elem->type == COLL_SYM
+	      && strlen ((char *) end_elem->opr.name) > 1), 0))
+    return REG_ECOLLATE;
+
+# ifdef RE_ENABLE_I18N
+  {
+    wchar_t wc;
+    wint_t start_wc;
+    wint_t end_wc;
+    wchar_t cmp_buf[6] = {L'\0', L'\0', L'\0', L'\0', L'\0', L'\0'};
+
+    start_ch = ((start_elem->type == SB_CHAR) ? start_elem->opr.ch
+		: ((start_elem->type == COLL_SYM) ? start_elem->opr.name[0]
+		   : 0));
+    end_ch = ((end_elem->type == SB_CHAR) ? end_elem->opr.ch
+	      : ((end_elem->type == COLL_SYM) ? end_elem->opr.name[0]
+		 : 0));
+#ifdef GAWK
+    /*
+     * Fedora Core 2, maybe others, have broken `btowc' that returns -1
+     * for any value > 127. Sigh. Note that `start_ch' and `end_ch' are
+     * unsigned, so we don't have sign extension problems.
+     */
+    start_wc = ((start_elem->type == SB_CHAR || start_elem->type == COLL_SYM)
+		? start_ch : start_elem->opr.wch);
+    end_wc = ((end_elem->type == SB_CHAR || end_elem->type == COLL_SYM)
+	      ? end_ch : end_elem->opr.wch);
+#else
+    start_wc = ((start_elem->type == SB_CHAR || start_elem->type == COLL_SYM)
+		? __btowc (start_ch) : start_elem->opr.wch);
+    end_wc = ((end_elem->type == SB_CHAR || end_elem->type == COLL_SYM)
+	      ? __btowc (end_ch) : end_elem->opr.wch);
+#endif
+    if (start_wc == WEOF || end_wc == WEOF)
+      return REG_ECOLLATE;
+    cmp_buf[0] = start_wc;
+    cmp_buf[4] = end_wc;
+    if (wcscoll (cmp_buf, cmp_buf + 4) > 0)
+      return REG_ERANGE;
+
+    /* Got valid collation sequence values, add them as a new entry.
+       However, for !_LIBC we have no collation elements: if the
+       character set is single byte, the single byte character set
+       that we build below suffices.  parse_bracket_exp passes
+       no MBCSET if dfa->mb_cur_max == 1.  */
+    if (mbcset)
+      {
+	/* Check the space of the arrays.  */
+	if (BE (*range_alloc == mbcset->nranges, 0))
+	  {
+	    /* There is not enough space, need realloc.  */
+	    wchar_t *new_array_start, *new_array_end;
+	    int new_nranges;
+
+	    /* +1 in case of mbcset->nranges is 0.  */
+	    new_nranges = 2 * mbcset->nranges + 1;
+	    /* Use realloc since mbcset->range_starts and mbcset->range_ends
+	       are NULL if *range_alloc == 0.  */
+	    new_array_start = re_realloc (mbcset->range_starts, wchar_t,
+					  new_nranges);
+	    new_array_end = re_realloc (mbcset->range_ends, wchar_t,
+					new_nranges);
+
+	    if (BE (new_array_start == NULL || new_array_end == NULL, 0))
+	      return REG_ESPACE;
+
+	    mbcset->range_starts = new_array_start;
+	    mbcset->range_ends = new_array_end;
+	    *range_alloc = new_nranges;
+	  }
+
+	mbcset->range_starts[mbcset->nranges] = start_wc;
+	mbcset->range_ends[mbcset->nranges++] = end_wc;
+      }
+
+    /* Build the table for single byte characters.  */
+    for (wc = 0; wc < SBC_MAX; ++wc)
+      {
+	cmp_buf[2] = wc;
+	if (wcscoll (cmp_buf, cmp_buf + 2) <= 0
+	    && wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0)
+	  bitset_set (sbcset, wc);
+      }
+  }
+# else /* not RE_ENABLE_I18N */
+  {
+    unsigned int ch;
+    start_ch = ((start_elem->type == SB_CHAR ) ? start_elem->opr.ch
+		: ((start_elem->type == COLL_SYM) ? start_elem->opr.name[0]
+		   : 0));
+    end_ch = ((end_elem->type == SB_CHAR ) ? end_elem->opr.ch
+	      : ((end_elem->type == COLL_SYM) ? end_elem->opr.name[0]
+		 : 0));
+    if (start_ch > end_ch)
+      return REG_ERANGE;
+    /* Build the table for single byte characters.  */
+    for (ch = 0; ch < SBC_MAX; ++ch)
+      if (start_ch <= ch  && ch <= end_ch)
+	bitset_set (sbcset, ch);
+  }
+# endif /* not RE_ENABLE_I18N */
+  return REG_NOERROR;
+}
+#endif /* not _LIBC */
+
+#ifndef _LIBC
+/* Helper function for parse_bracket_exp only used in case of NOT _LIBC..
+   Build the collating element which is represented by NAME.
+   The result are written to MBCSET and SBCSET.
+   COLL_SYM_ALLOC is the allocated size of mbcset->coll_sym, is a
+   pointer argument since we may update it.  */
+
+static reg_errcode_t
+internal_function
+# ifdef RE_ENABLE_I18N
+build_collating_symbol (bitset_t sbcset, re_charset_t *mbcset,
+			int *coll_sym_alloc, const unsigned char *name)
+# else /* not RE_ENABLE_I18N */
+build_collating_symbol (bitset_t sbcset, const unsigned char *name)
+# endif /* not RE_ENABLE_I18N */
+{
+  size_t name_len = strlen ((const char *) name);
+  if (BE (name_len != 1, 0))
+    return REG_ECOLLATE;
+  else
+    {
+      bitset_set (sbcset, name[0]);
+      return REG_NOERROR;
+    }
+}
+#endif /* not _LIBC */
+
+/* This function parse bracket expression like "[abc]", "[a-c]",
+   "[[.a-a.]]" etc.  */
+
+static bin_tree_t *
+parse_bracket_exp (re_string_t *regexp, re_dfa_t *dfa, re_token_t *token,
+		   reg_syntax_t syntax, reg_errcode_t *err)
+{
+#ifdef _LIBC
+  const unsigned char *collseqmb;
+  const char *collseqwc;
+  uint32_t nrules;
+  int32_t table_size;
+  const int32_t *symb_table;
+  const unsigned char *extra;
+
+  /* Local function for parse_bracket_exp used in _LIBC environment.
+     Seek the collating symbol entry correspondings to NAME.
+     Return the index of the symbol in the SYMB_TABLE.  */
+
+  auto inline int32_t
+  __attribute ((always_inline))
+  seek_collating_symbol_entry (name, name_len)
+	 const unsigned char *name;
+	 size_t name_len;
+    {
+      int32_t hash = elem_hash ((const char *) name, name_len);
+      int32_t elem = hash % table_size;
+      if (symb_table[2 * elem] != 0)
+	{
+	  int32_t second = hash % (table_size - 2) + 1;
+
+	  do
+	    {
+	      /* First compare the hashing value.  */
+	      if (symb_table[2 * elem] == hash
+		  /* Compare the length of the name.  */
+		  && name_len == extra[symb_table[2 * elem + 1]]
+		  /* Compare the name.  */
+		  && memcmp (name, &extra[symb_table[2 * elem + 1] + 1],
+			     name_len) == 0)
+		{
+		  /* Yep, this is the entry.  */
+		  break;
+		}
+
+	      /* Next entry.  */
+	      elem += second;
+	    }
+	  while (symb_table[2 * elem] != 0);
+	}
+      return elem;
+    }
+
+  /* Local function for parse_bracket_exp used in _LIBC environment.
+     Look up the collation sequence value of BR_ELEM.
+     Return the value if succeeded, UINT_MAX otherwise.  */
+
+  auto inline unsigned int
+  __attribute ((always_inline))
+  lookup_collation_sequence_value (br_elem)
+	 bracket_elem_t *br_elem;
+    {
+      if (br_elem->type == SB_CHAR)
+	{
+	  /*
+	  if (MB_CUR_MAX == 1)
+	  */
+	  if (nrules == 0)
+	    return collseqmb[br_elem->opr.ch];
+	  else
+	    {
+	      wint_t wc = __btowc (br_elem->opr.ch);
+	      return __collseq_table_lookup (collseqwc, wc);
+	    }
+	}
+      else if (br_elem->type == MB_CHAR)
+	{
+	  if (nrules != 0)
+	    return __collseq_table_lookup (collseqwc, br_elem->opr.wch);
+	}
+      else if (br_elem->type == COLL_SYM)
+	{
+	  size_t sym_name_len = strlen ((char *) br_elem->opr.name);
+	  if (nrules != 0)
+	    {
+	      int32_t elem, idx;
+	      elem = seek_collating_symbol_entry (br_elem->opr.name,
+						  sym_name_len);
+	      if (symb_table[2 * elem] != 0)
+		{
+		  /* We found the entry.  */
+		  idx = symb_table[2 * elem + 1];
+		  /* Skip the name of collating element name.  */
+		  idx += 1 + extra[idx];
+		  /* Skip the byte sequence of the collating element.  */
+		  idx += 1 + extra[idx];
+		  /* Adjust for the alignment.  */
+		  idx = (idx + 3) & ~3;
+		  /* Skip the multibyte collation sequence value.  */
+		  idx += sizeof (unsigned int);
+		  /* Skip the wide char sequence of the collating element.  */
+		  idx += sizeof (unsigned int) *
+		    (1 + *(unsigned int *) (extra + idx));
+		  /* Return the collation sequence value.  */
+		  return *(unsigned int *) (extra + idx);
+		}
+	      else if (symb_table[2 * elem] == 0 && sym_name_len == 1)
+		{
+		  /* No valid character.  Match it as a single byte
+		     character.  */
+		  return collseqmb[br_elem->opr.name[0]];
+		}
+	    }
+	  else if (sym_name_len == 1)
+	    return collseqmb[br_elem->opr.name[0]];
+	}
+      return UINT_MAX;
+    }
+
+  /* Local function for parse_bracket_exp used in _LIBC environment.
+     Build the range expression which starts from START_ELEM, and ends
+     at END_ELEM.  The result are written to MBCSET and SBCSET.
+     RANGE_ALLOC is the allocated size of mbcset->range_starts, and
+     mbcset->range_ends, is a pointer argument since we may
+     update it.  */
+
+  auto inline reg_errcode_t
+  __attribute ((always_inline))
+  build_range_exp (sbcset, mbcset, range_alloc, start_elem, end_elem)
+	 re_charset_t *mbcset;
+	 int *range_alloc;
+	 bitset_t sbcset;
+	 bracket_elem_t *start_elem, *end_elem;
+    {
+      unsigned int ch;
+      uint32_t start_collseq;
+      uint32_t end_collseq;
+
+      /* Equivalence Classes and Character Classes can't be a range
+	 start/end.  */
+      if (BE (start_elem->type == EQUIV_CLASS || start_elem->type == CHAR_CLASS
+	      || end_elem->type == EQUIV_CLASS || end_elem->type == CHAR_CLASS,
+	      0))
+	return REG_ERANGE;
+
+      start_collseq = lookup_collation_sequence_value (start_elem);
+      end_collseq = lookup_collation_sequence_value (end_elem);
+      /* Check start/end collation sequence values.  */
+      if (BE (start_collseq == UINT_MAX || end_collseq == UINT_MAX, 0))
+	return REG_ECOLLATE;
+      if (BE ((syntax & RE_NO_EMPTY_RANGES) && start_collseq > end_collseq, 0))
+	return REG_ERANGE;
+
+      /* Got valid collation sequence values, add them as a new entry.
+	 However, if we have no collation elements, and the character set
+	 is single byte, the single byte character set that we
+	 build below suffices. */
+      if (nrules > 0 || dfa->mb_cur_max > 1)
+	{
+	  /* Check the space of the arrays.  */
+	  if (BE (*range_alloc == mbcset->nranges, 0))
+	    {
+	      /* There is not enough space, need realloc.  */
+	      uint32_t *new_array_start;
+	      uint32_t *new_array_end;
+	      int new_nranges;
+
+	      /* +1 in case of mbcset->nranges is 0.  */
+	      new_nranges = 2 * mbcset->nranges + 1;
+	      new_array_start = re_realloc (mbcset->range_starts, uint32_t,
+					    new_nranges);
+	      new_array_end = re_realloc (mbcset->range_ends, uint32_t,
+					  new_nranges);
+
+	      if (BE (new_array_start == NULL || new_array_end == NULL, 0))
+		return REG_ESPACE;
+
+	      mbcset->range_starts = new_array_start;
+	      mbcset->range_ends = new_array_end;
+	      *range_alloc = new_nranges;
+	    }
+
+	  mbcset->range_starts[mbcset->nranges] = start_collseq;
+	  mbcset->range_ends[mbcset->nranges++] = end_collseq;
+	}
+
+      /* Build the table for single byte characters.  */
+      for (ch = 0; ch < SBC_MAX; ch++)
+	{
+	  uint32_t ch_collseq;
+	  /*
+	  if (MB_CUR_MAX == 1)
+	  */
+	  if (nrules == 0)
+	    ch_collseq = collseqmb[ch];
+	  else
+	    ch_collseq = __collseq_table_lookup (collseqwc, __btowc (ch));
+	  if (start_collseq <= ch_collseq && ch_collseq <= end_collseq)
+	    bitset_set (sbcset, ch);
+	}
+      return REG_NOERROR;
+    }
+
+  /* Local function for parse_bracket_exp used in _LIBC environment.
+     Build the collating element which is represented by NAME.
+     The result are written to MBCSET and SBCSET.
+     COLL_SYM_ALLOC is the allocated size of mbcset->coll_sym, is a
+     pointer argument since we may update it.  */
+
+  auto inline reg_errcode_t
+  __attribute ((always_inline))
+  build_collating_symbol (sbcset, mbcset, coll_sym_alloc, name)
+	 re_charset_t *mbcset;
+	 int *coll_sym_alloc;
+	 bitset_t sbcset;
+	 const unsigned char *name;
+    {
+      int32_t elem, idx;
+      size_t name_len = strlen ((const char *) name);
+      if (nrules != 0)
+	{
+	  elem = seek_collating_symbol_entry (name, name_len);
+	  if (symb_table[2 * elem] != 0)
+	    {
+	      /* We found the entry.  */
+	      idx = symb_table[2 * elem + 1];
+	      /* Skip the name of collating element name.  */
+	      idx += 1 + extra[idx];
+	    }
+	  else if (symb_table[2 * elem] == 0 && name_len == 1)
+	    {
+	      /* No valid character, treat it as a normal
+		 character.  */
+	      bitset_set (sbcset, name[0]);
+	      return REG_NOERROR;
+	    }
+	  else
+	    return REG_ECOLLATE;
+
+	  /* Got valid collation sequence, add it as a new entry.  */
+	  /* Check the space of the arrays.  */
+	  if (BE (*coll_sym_alloc == mbcset->ncoll_syms, 0))
+	    {
+	      /* Not enough, realloc it.  */
+	      /* +1 in case of mbcset->ncoll_syms is 0.  */
+	      int new_coll_sym_alloc = 2 * mbcset->ncoll_syms + 1;
+	      /* Use realloc since mbcset->coll_syms is NULL
+		 if *alloc == 0.  */
+	      int32_t *new_coll_syms = re_realloc (mbcset->coll_syms, int32_t,
+						   new_coll_sym_alloc);
+	      if (BE (new_coll_syms == NULL, 0))
+		return REG_ESPACE;
+	      mbcset->coll_syms = new_coll_syms;
+	      *coll_sym_alloc = new_coll_sym_alloc;
+	    }
+	  mbcset->coll_syms[mbcset->ncoll_syms++] = idx;
+	  return REG_NOERROR;
+	}
+      else
+	{
+	  if (BE (name_len != 1, 0))
+	    return REG_ECOLLATE;
+	  else
+	    {
+	      bitset_set (sbcset, name[0]);
+	      return REG_NOERROR;
+	    }
+	}
+    }
+#endif
+
+  re_token_t br_token;
+  re_bitset_ptr_t sbcset;
+#ifdef RE_ENABLE_I18N
+  re_charset_t *mbcset;
+  int coll_sym_alloc = 0, range_alloc = 0, mbchar_alloc = 0;
+  int equiv_class_alloc = 0, char_class_alloc = 0;
+#endif /* not RE_ENABLE_I18N */
+  int non_match = 0;
+  bin_tree_t *work_tree;
+  int token_len;
+  int first_round = 1;
+#ifdef _LIBC
+  collseqmb = (const unsigned char *)
+    _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQMB);
+  nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
+  if (nrules)
+    {
+      /*
+      if (MB_CUR_MAX > 1)
+      */
+      collseqwc = _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQWC);
+      table_size = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_SYMB_HASH_SIZEMB);
+      symb_table = (const int32_t *) _NL_CURRENT (LC_COLLATE,
+						  _NL_COLLATE_SYMB_TABLEMB);
+      extra = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
+						   _NL_COLLATE_SYMB_EXTRAMB);
+    }
+#endif
+  sbcset = (re_bitset_ptr_t) calloc (sizeof (bitset_t), 1);
+#ifdef RE_ENABLE_I18N
+  mbcset = (re_charset_t *) calloc (sizeof (re_charset_t), 1);
+#endif /* RE_ENABLE_I18N */
+#ifdef RE_ENABLE_I18N
+  if (BE (sbcset == NULL || mbcset == NULL, 0))
+#else
+  if (BE (sbcset == NULL, 0))
+#endif /* RE_ENABLE_I18N */
+    {
+      *err = REG_ESPACE;
+      return NULL;
+    }
+
+  token_len = peek_token_bracket (token, regexp, syntax);
+  if (BE (token->type == END_OF_RE, 0))
+    {
+      *err = REG_BADPAT;
+      goto parse_bracket_exp_free_return;
+    }
+  if (token->type == OP_NON_MATCH_LIST)
+    {
+#ifdef RE_ENABLE_I18N
+      mbcset->non_match = 1;
+#endif /* not RE_ENABLE_I18N */
+      non_match = 1;
+      if (syntax & RE_HAT_LISTS_NOT_NEWLINE)
+	bitset_set (sbcset, '\n');
+      re_string_skip_bytes (regexp, token_len); /* Skip a token.  */
+      token_len = peek_token_bracket (token, regexp, syntax);
+      if (BE (token->type == END_OF_RE, 0))
+	{
+	  *err = REG_BADPAT;
+	  goto parse_bracket_exp_free_return;
+	}
+    }
+
+  /* We treat the first ']' as a normal character.  */
+  if (token->type == OP_CLOSE_BRACKET)
+    token->type = CHARACTER;
+
+  while (1)
+    {
+      bracket_elem_t start_elem, end_elem;
+      unsigned char start_name_buf[BRACKET_NAME_BUF_SIZE];
+      unsigned char end_name_buf[BRACKET_NAME_BUF_SIZE];
+      reg_errcode_t ret;
+      int token_len2 = 0, is_range_exp = 0;
+      re_token_t token2;
+
+      start_elem.opr.name = start_name_buf;
+      ret = parse_bracket_element (&start_elem, regexp, token, token_len, dfa,
+				   syntax, first_round);
+      if (BE (ret != REG_NOERROR, 0))
+	{
+	  *err = ret;
+	  goto parse_bracket_exp_free_return;
+	}
+      first_round = 0;
+
+      /* Get information about the next token.  We need it in any case.  */
+      token_len = peek_token_bracket (token, regexp, syntax);
+
+      /* Do not check for ranges if we know they are not allowed.  */
+      if (start_elem.type != CHAR_CLASS && start_elem.type != EQUIV_CLASS)
+	{
+	  if (BE (token->type == END_OF_RE, 0))
+	    {
+	      *err = REG_EBRACK;
+	      goto parse_bracket_exp_free_return;
+	    }
+	  if (token->type == OP_CHARSET_RANGE)
+	    {
+	      re_string_skip_bytes (regexp, token_len); /* Skip '-'.  */
+	      token_len2 = peek_token_bracket (&token2, regexp, syntax);
+	      if (BE (token2.type == END_OF_RE, 0))
+		{
+		  *err = REG_EBRACK;
+		  goto parse_bracket_exp_free_return;
+		}
+	      if (token2.type == OP_CLOSE_BRACKET)
+		{
+		  /* We treat the last '-' as a normal character.  */
+		  re_string_skip_bytes (regexp, -token_len);
+		  token->type = CHARACTER;
+		}
+	      else
+		is_range_exp = 1;
+	    }
+	}
+
+      if (is_range_exp == 1)
+	{
+	  end_elem.opr.name = end_name_buf;
+	  ret = parse_bracket_element (&end_elem, regexp, &token2, token_len2,
+				       dfa, syntax, 1);
+	  if (BE (ret != REG_NOERROR, 0))
+	    {
+	      *err = ret;
+	      goto parse_bracket_exp_free_return;
+	    }
+
+	  token_len = peek_token_bracket (token, regexp, syntax);
+
+#ifdef _LIBC
+	  *err = build_range_exp (sbcset, mbcset, &range_alloc,
+				  &start_elem, &end_elem);
+#else
+# ifdef RE_ENABLE_I18N
+	  *err = build_range_exp (sbcset,
+				  dfa->mb_cur_max > 1 ? mbcset : NULL,
+				  &range_alloc, &start_elem, &end_elem);
+# else
+	  *err = build_range_exp (sbcset, &start_elem, &end_elem);
+# endif
+#endif /* RE_ENABLE_I18N */
+	  if (BE (*err != REG_NOERROR, 0))
+	    goto parse_bracket_exp_free_return;
+	}
+      else
+	{
+	  switch (start_elem.type)
+	    {
+	    case SB_CHAR:
+	      bitset_set (sbcset, start_elem.opr.ch);
+	      break;
+#ifdef RE_ENABLE_I18N
+	    case MB_CHAR:
+	      /* Check whether the array has enough space.  */
+	      if (BE (mbchar_alloc == mbcset->nmbchars, 0))
+		{
+		  wchar_t *new_mbchars;
+		  /* Not enough, realloc it.  */
+		  /* +1 in case of mbcset->nmbchars is 0.  */
+		  mbchar_alloc = 2 * mbcset->nmbchars + 1;
+		  /* Use realloc since array is NULL if *alloc == 0.  */
+		  new_mbchars = re_realloc (mbcset->mbchars, wchar_t,
+					    mbchar_alloc);
+		  if (BE (new_mbchars == NULL, 0))
+		    goto parse_bracket_exp_espace;
+		  mbcset->mbchars = new_mbchars;
+		}
+	      mbcset->mbchars[mbcset->nmbchars++] = start_elem.opr.wch;
+	      break;
+#endif /* RE_ENABLE_I18N */
+	    case EQUIV_CLASS:
+	      *err = build_equiv_class (sbcset,
+#ifdef RE_ENABLE_I18N
+					mbcset, &equiv_class_alloc,
+#endif /* RE_ENABLE_I18N */
+					start_elem.opr.name);
+	      if (BE (*err != REG_NOERROR, 0))
+		goto parse_bracket_exp_free_return;
+	      break;
+	    case COLL_SYM:
+	      *err = build_collating_symbol (sbcset,
+#ifdef RE_ENABLE_I18N
+					     mbcset, &coll_sym_alloc,
+#endif /* RE_ENABLE_I18N */
+					     start_elem.opr.name);
+	      if (BE (*err != REG_NOERROR, 0))
+		goto parse_bracket_exp_free_return;
+	      break;
+	    case CHAR_CLASS:
+	      *err = build_charclass (regexp->trans, sbcset,
+#ifdef RE_ENABLE_I18N
+				      mbcset, &char_class_alloc,
+#endif /* RE_ENABLE_I18N */
+				      (const char *) start_elem.opr.name, syntax);
+	      if (BE (*err != REG_NOERROR, 0))
+	       goto parse_bracket_exp_free_return;
+	      break;
+	    default:
+	      assert (0);
+	      break;
+	    }
+	}
+      if (BE (token->type == END_OF_RE, 0))
+	{
+	  *err = REG_EBRACK;
+	  goto parse_bracket_exp_free_return;
+	}
+      if (token->type == OP_CLOSE_BRACKET)
+	break;
+    }
+
+  re_string_skip_bytes (regexp, token_len); /* Skip a token.  */
+
+  /* If it is non-matching list.  */
+  if (non_match)
+    bitset_not (sbcset);
+
+#ifdef RE_ENABLE_I18N
+  /* Ensure only single byte characters are set.  */
+  if (dfa->mb_cur_max > 1)
+    bitset_mask (sbcset, dfa->sb_char);
+
+  if (mbcset->nmbchars || mbcset->ncoll_syms || mbcset->nequiv_classes
+      || mbcset->nranges || (dfa->mb_cur_max > 1 && (mbcset->nchar_classes
+						     || mbcset->non_match)))
+    {
+      bin_tree_t *mbc_tree;
+      int sbc_idx;
+      /* Build a tree for complex bracket.  */
+      dfa->has_mb_node = 1;
+      br_token.type = COMPLEX_BRACKET;
+      br_token.opr.mbcset = mbcset;
+      mbc_tree = create_token_tree (dfa, NULL, NULL, &br_token);
+      if (BE (mbc_tree == NULL, 0))
+	goto parse_bracket_exp_espace;
+      for (sbc_idx = 0; sbc_idx < BITSET_WORDS; ++sbc_idx)
+	if (sbcset[sbc_idx])
+	  break;
+      /* If there are no bits set in sbcset, there is no point
+	 of having both SIMPLE_BRACKET and COMPLEX_BRACKET.  */
+      if (sbc_idx < BITSET_WORDS)
+	{
+	  /* Build a tree for simple bracket.  */
+	  br_token.type = SIMPLE_BRACKET;
+	  br_token.opr.sbcset = sbcset;
+	  work_tree = create_token_tree (dfa, NULL, NULL, &br_token);
+	  if (BE (work_tree == NULL, 0))
+	    goto parse_bracket_exp_espace;
+
+	  /* Then join them by ALT node.  */
+	  work_tree = create_tree (dfa, work_tree, mbc_tree, OP_ALT);
+	  if (BE (work_tree == NULL, 0))
+	    goto parse_bracket_exp_espace;
+	}
+      else
+	{
+	  re_free (sbcset);
+	  work_tree = mbc_tree;
+	}
+    }
+  else
+#endif /* not RE_ENABLE_I18N */
+    {
+#ifdef RE_ENABLE_I18N
+      free_charset (mbcset);
+#endif
+      /* Build a tree for simple bracket.  */
+      br_token.type = SIMPLE_BRACKET;
+      br_token.opr.sbcset = sbcset;
+      work_tree = create_token_tree (dfa, NULL, NULL, &br_token);
+      if (BE (work_tree == NULL, 0))
+	goto parse_bracket_exp_espace;
+    }
+  return work_tree;
+
+ parse_bracket_exp_espace:
+  *err = REG_ESPACE;
+ parse_bracket_exp_free_return:
+  re_free (sbcset);
+#ifdef RE_ENABLE_I18N
+  free_charset (mbcset);
+#endif /* RE_ENABLE_I18N */
+  return NULL;
+}
+
+/* Parse an element in the bracket expression.  */
+
+static reg_errcode_t
+parse_bracket_element (bracket_elem_t *elem, re_string_t *regexp,
+		       re_token_t *token, int token_len, re_dfa_t *dfa,
+		       reg_syntax_t syntax, int accept_hyphen)
+{
+#ifdef RE_ENABLE_I18N
+  int cur_char_size;
+  cur_char_size = re_string_char_size_at (regexp, re_string_cur_idx (regexp));
+  if (cur_char_size > 1)
+    {
+      elem->type = MB_CHAR;
+      elem->opr.wch = re_string_wchar_at (regexp, re_string_cur_idx (regexp));
+      re_string_skip_bytes (regexp, cur_char_size);
+      return REG_NOERROR;
+    }
+#endif /* RE_ENABLE_I18N */
+  re_string_skip_bytes (regexp, token_len); /* Skip a token.  */
+  if (token->type == OP_OPEN_COLL_ELEM || token->type == OP_OPEN_CHAR_CLASS
+      || token->type == OP_OPEN_EQUIV_CLASS)
+    return parse_bracket_symbol (elem, regexp, token);
+  if (BE (token->type == OP_CHARSET_RANGE, 0) && !accept_hyphen)
+    {
+      /* A '-' must only appear as anything but a range indicator before
+	 the closing bracket.  Everything else is an error.  */
+      re_token_t token2;
+      (void) peek_token_bracket (&token2, regexp, syntax);
+      if (token2.type != OP_CLOSE_BRACKET)
+	/* The actual error value is not standardized since this whole
+	   case is undefined.  But ERANGE makes good sense.  */
+	return REG_ERANGE;
+    }
+  elem->type = SB_CHAR;
+  elem->opr.ch = token->opr.c;
+  return REG_NOERROR;
+}
+
+/* Parse a bracket symbol in the bracket expression.  Bracket symbols are
+   such as [:<character_class>:], [.<collating_element>.], and
+   [=<equivalent_class>=].  */
+
+static reg_errcode_t
+parse_bracket_symbol (bracket_elem_t *elem, re_string_t *regexp,
+		      re_token_t *token)
+{
+  unsigned char ch, delim = token->opr.c;
+  int i = 0;
+  if (re_string_eoi(regexp))
+    return REG_EBRACK;
+  for (;; ++i)
+    {
+      if (i >= BRACKET_NAME_BUF_SIZE)
+	return REG_EBRACK;
+      if (token->type == OP_OPEN_CHAR_CLASS)
+	ch = re_string_fetch_byte_case (regexp);
+      else
+	ch = re_string_fetch_byte (regexp);
+      if (re_string_eoi(regexp))
+	return REG_EBRACK;
+      if (ch == delim && re_string_peek_byte (regexp, 0) == ']')
+	break;
+      elem->opr.name[i] = ch;
+    }
+  re_string_skip_bytes (regexp, 1);
+  elem->opr.name[i] = '\0';
+  switch (token->type)
+    {
+    case OP_OPEN_COLL_ELEM:
+      elem->type = COLL_SYM;
+      break;
+    case OP_OPEN_EQUIV_CLASS:
+      elem->type = EQUIV_CLASS;
+      break;
+    case OP_OPEN_CHAR_CLASS:
+      elem->type = CHAR_CLASS;
+      break;
+    default:
+      break;
+    }
+  return REG_NOERROR;
+}
+
+  /* Helper function for parse_bracket_exp.
+     Build the equivalence class which is represented by NAME.
+     The result are written to MBCSET and SBCSET.
+     EQUIV_CLASS_ALLOC is the allocated size of mbcset->equiv_classes,
+     is a pointer argument since we may update it.  */
+
+static reg_errcode_t
+#ifdef RE_ENABLE_I18N
+build_equiv_class (bitset_t sbcset, re_charset_t *mbcset,
+		   int *equiv_class_alloc, const unsigned char *name)
+#else /* not RE_ENABLE_I18N */
+build_equiv_class (bitset_t sbcset, const unsigned char *name)
+#endif /* not RE_ENABLE_I18N */
+{
+#ifdef _LIBC
+  uint32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
+  if (nrules != 0)
+    {
+      const int32_t *table, *indirect;
+      const unsigned char *weights, *extra, *cp;
+      unsigned char char_buf[2];
+      int32_t idx1, idx2;
+      unsigned int ch;
+      size_t len;
+      /* This #include defines a local function!  */
+# include <locale/weight.h>
+      /* Calculate the index for equivalence class.  */
+      cp = name;
+      table = (const int32_t *) _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
+      weights = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
+					       _NL_COLLATE_WEIGHTMB);
+      extra = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
+						   _NL_COLLATE_EXTRAMB);
+      indirect = (const int32_t *) _NL_CURRENT (LC_COLLATE,
+						_NL_COLLATE_INDIRECTMB);
+      idx1 = findidx (&cp);
+      if (BE (idx1 == 0 || cp < name + strlen ((const char *) name), 0))
+	/* This isn't a valid character.  */
+	return REG_ECOLLATE;
+
+      /* Build single byte matcing table for this equivalence class.  */
+      char_buf[1] = (unsigned char) '\0';
+      len = weights[idx1 & 0xffffff];
+      for (ch = 0; ch < SBC_MAX; ++ch)
+	{
+	  char_buf[0] = ch;
+	  cp = char_buf;
+	  idx2 = findidx (&cp);
+/*
+	  idx2 = table[ch];
+*/
+	  if (idx2 == 0)
+	    /* This isn't a valid character.  */
+	    continue;
+	  /* Compare only if the length matches and the collation rule
+	     index is the same.  */
+	  if (len == weights[idx2 & 0xffffff] && (idx1 >> 24) == (idx2 >> 24))
+	    {
+	      int cnt = 0;
+
+	      while (cnt <= len &&
+		     weights[(idx1 & 0xffffff) + 1 + cnt]
+		     == weights[(idx2 & 0xffffff) + 1 + cnt])
+		++cnt;
+
+	      if (cnt > len)
+		bitset_set (sbcset, ch);
+	    }
+	}
+      /* Check whether the array has enough space.  */
+      if (BE (*equiv_class_alloc == mbcset->nequiv_classes, 0))
+	{
+	  /* Not enough, realloc it.  */
+	  /* +1 in case of mbcset->nequiv_classes is 0.  */
+	  int new_equiv_class_alloc = 2 * mbcset->nequiv_classes + 1;
+	  /* Use realloc since the array is NULL if *alloc == 0.  */
+	  int32_t *new_equiv_classes = re_realloc (mbcset->equiv_classes,
+						   int32_t,
+						   new_equiv_class_alloc);
+	  if (BE (new_equiv_classes == NULL, 0))
+	    return REG_ESPACE;
+	  mbcset->equiv_classes = new_equiv_classes;
+	  *equiv_class_alloc = new_equiv_class_alloc;
+	}
+      mbcset->equiv_classes[mbcset->nequiv_classes++] = idx1;
+    }
+  else
+#endif /* _LIBC */
+    {
+      if (BE (strlen ((const char *) name) != 1, 0))
+	return REG_ECOLLATE;
+      bitset_set (sbcset, *name);
+    }
+  return REG_NOERROR;
+}
+
+  /* Helper function for parse_bracket_exp.
+     Build the character class which is represented by NAME.
+     The result are written to MBCSET and SBCSET.
+     CHAR_CLASS_ALLOC is the allocated size of mbcset->char_classes,
+     is a pointer argument since we may update it.  */
+
+static reg_errcode_t
+#ifdef RE_ENABLE_I18N
+build_charclass (RE_TRANSLATE_TYPE trans, bitset_t sbcset,
+		 re_charset_t *mbcset, int *char_class_alloc,
+		 const char *class_name, reg_syntax_t syntax)
+#else /* not RE_ENABLE_I18N */
+build_charclass (RE_TRANSLATE_TYPE trans, bitset_t sbcset,
+		 const char *class_name, reg_syntax_t syntax)
+#endif /* not RE_ENABLE_I18N */
+{
+  int i;
+
+  /* In case of REG_ICASE "upper" and "lower" match the both of
+     upper and lower cases.  */
+  if ((syntax & RE_ICASE)
+      && (strcmp (class_name, "upper") == 0 || strcmp (class_name, "lower") == 0))
+    class_name = "alpha";
+
+#ifdef RE_ENABLE_I18N
+  /* Check the space of the arrays.  */
+  if (BE (*char_class_alloc == mbcset->nchar_classes, 0))
+    {
+      /* Not enough, realloc it.  */
+      /* +1 in case of mbcset->nchar_classes is 0.  */
+      int new_char_class_alloc = 2 * mbcset->nchar_classes + 1;
+      /* Use realloc since array is NULL if *alloc == 0.  */
+      wctype_t *new_char_classes = re_realloc (mbcset->char_classes, wctype_t,
+					       new_char_class_alloc);
+      if (BE (new_char_classes == NULL, 0))
+	return REG_ESPACE;
+      mbcset->char_classes = new_char_classes;
+      *char_class_alloc = new_char_class_alloc;
+    }
+  mbcset->char_classes[mbcset->nchar_classes++] = __wctype (class_name);
+#endif /* RE_ENABLE_I18N */
+
+#define BUILD_CHARCLASS_LOOP(ctype_func)	\
+  do {						\
+    if (BE (trans != NULL, 0))			\
+      {						\
+	for (i = 0; i < SBC_MAX; ++i)		\
+  	  if (ctype_func (i))			\
+	    bitset_set (sbcset, trans[i]);	\
+      }						\
+    else					\
+      {						\
+	for (i = 0; i < SBC_MAX; ++i)		\
+  	  if (ctype_func (i))			\
+	    bitset_set (sbcset, i);		\
+      }						\
+  } while (0)
+
+  if (strcmp (class_name, "alnum") == 0)
+    BUILD_CHARCLASS_LOOP (isalnum);
+  else if (strcmp (class_name, "cntrl") == 0)
+    BUILD_CHARCLASS_LOOP (iscntrl);
+  else if (strcmp (class_name, "lower") == 0)
+    BUILD_CHARCLASS_LOOP (islower);
+  else if (strcmp (class_name, "space") == 0)
+    BUILD_CHARCLASS_LOOP (isspace);
+  else if (strcmp (class_name, "alpha") == 0)
+    BUILD_CHARCLASS_LOOP (isalpha);
+  else if (strcmp (class_name, "digit") == 0)
+    BUILD_CHARCLASS_LOOP (isdigit);
+  else if (strcmp (class_name, "print") == 0)
+    BUILD_CHARCLASS_LOOP (isprint);
+  else if (strcmp (class_name, "upper") == 0)
+    BUILD_CHARCLASS_LOOP (isupper);
+  else if (strcmp (class_name, "blank") == 0)
+#ifndef GAWK
+    BUILD_CHARCLASS_LOOP (isblank);
+#else
+    /* see comments above */
+    BUILD_CHARCLASS_LOOP (is_blank);
+#endif
+  else if (strcmp (class_name, "graph") == 0)
+    BUILD_CHARCLASS_LOOP (isgraph);
+  else if (strcmp (class_name, "punct") == 0)
+    BUILD_CHARCLASS_LOOP (ispunct);
+  else if (strcmp (class_name, "xdigit") == 0)
+    BUILD_CHARCLASS_LOOP (isxdigit);
+  else
+    return REG_ECTYPE;
+
+  return REG_NOERROR;
+}
+
+static bin_tree_t *
+build_charclass_op (re_dfa_t *dfa, RE_TRANSLATE_TYPE trans,
+		    const char *class_name,
+		    const char *extra, int non_match,
+		    reg_errcode_t *err)
+{
+  re_bitset_ptr_t sbcset;
+#ifdef RE_ENABLE_I18N
+  re_charset_t *mbcset;
+  int alloc = 0;
+#endif /* not RE_ENABLE_I18N */
+  reg_errcode_t ret;
+  re_token_t br_token;
+  bin_tree_t *tree;
+
+  sbcset = (re_bitset_ptr_t) calloc (sizeof (bitset_t), 1);
+#ifdef RE_ENABLE_I18N
+  mbcset = (re_charset_t *) calloc (sizeof (re_charset_t), 1);
+#endif /* RE_ENABLE_I18N */
+
+#ifdef RE_ENABLE_I18N
+  if (BE (sbcset == NULL || mbcset == NULL, 0))
+#else /* not RE_ENABLE_I18N */
+  if (BE (sbcset == NULL, 0))
+#endif /* not RE_ENABLE_I18N */
+    {
+      *err = REG_ESPACE;
+      return NULL;
+    }
+
+  if (non_match)
+    {
+#ifdef RE_ENABLE_I18N
+      mbcset->non_match = 1;
+#endif /* not RE_ENABLE_I18N */
+    }
+
+  /* We don't care the syntax in this case.  */
+  ret = build_charclass (trans, sbcset,
+#ifdef RE_ENABLE_I18N
+			 mbcset, &alloc,
+#endif /* RE_ENABLE_I18N */
+			 class_name, 0);
+
+  if (BE (ret != REG_NOERROR, 0))
+    {
+      re_free (sbcset);
+#ifdef RE_ENABLE_I18N
+      free_charset (mbcset);
+#endif /* RE_ENABLE_I18N */
+      *err = ret;
+      return NULL;
+    }
+  /* \w match '_' also.  */
+  for (; *extra; extra++)
+    bitset_set (sbcset, *extra);
+
+  /* If it is non-matching list.  */
+  if (non_match)
+    bitset_not (sbcset);
+
+#ifdef RE_ENABLE_I18N
+  /* Ensure only single byte characters are set.  */
+  if (dfa->mb_cur_max > 1)
+    bitset_mask (sbcset, dfa->sb_char);
+#endif
+
+  /* Build a tree for simple bracket.  */
+  br_token.type = SIMPLE_BRACKET;
+  br_token.opr.sbcset = sbcset;
+  tree = create_token_tree (dfa, NULL, NULL, &br_token);
+  if (BE (tree == NULL, 0))
+    goto build_word_op_espace;
+
+#ifdef RE_ENABLE_I18N
+  if (dfa->mb_cur_max > 1)
+    {
+      bin_tree_t *mbc_tree;
+      /* Build a tree for complex bracket.  */
+      br_token.type = COMPLEX_BRACKET;
+      br_token.opr.mbcset = mbcset;
+      dfa->has_mb_node = 1;
+      mbc_tree = create_token_tree (dfa, NULL, NULL, &br_token);
+      if (BE (mbc_tree == NULL, 0))
+	goto build_word_op_espace;
+      /* Then join them by ALT node.  */
+      tree = create_tree (dfa, tree, mbc_tree, OP_ALT);
+      if (BE (mbc_tree != NULL, 1))
+	return tree;
+    }
+  else
+    {
+      free_charset (mbcset);
+      return tree;
+    }
+#else /* not RE_ENABLE_I18N */
+  return tree;
+#endif /* not RE_ENABLE_I18N */
+
+ build_word_op_espace:
+  re_free (sbcset);
+#ifdef RE_ENABLE_I18N
+  free_charset (mbcset);
+#endif /* RE_ENABLE_I18N */
+  *err = REG_ESPACE;
+  return NULL;
+}
+
+/* This is intended for the expressions like "a{1,3}".
+   Fetch a number from `input', and return the number.
+   Return -1, if the number field is empty like "{,1}".
+   Return -2, if an error has occurred.  */
+
+static int
+fetch_number (re_string_t *input, re_token_t *token, reg_syntax_t syntax)
+{
+  int num = -1;
+  unsigned char c;
+  while (1)
+    {
+      fetch_token (token, input, syntax);
+      c = token->opr.c;
+      if (BE (token->type == END_OF_RE, 0))
+	return -2;
+      if (token->type == OP_CLOSE_DUP_NUM || c == ',')
+	break;
+      num = ((token->type != CHARACTER || c < '0' || '9' < c || num == -2)
+	     ? -2 : ((num == -1) ? c - '0' : num * 10 + c - '0'));
+      num = (num > RE_DUP_MAX) ? -2 : num;
+    }
+  return num;
+}
+
+#ifdef RE_ENABLE_I18N
+static void
+free_charset (re_charset_t *cset)
+{
+  re_free (cset->mbchars);
+# ifdef _LIBC
+  re_free (cset->coll_syms);
+  re_free (cset->equiv_classes);
+  re_free (cset->range_starts);
+  re_free (cset->range_ends);
+# endif
+  re_free (cset->char_classes);
+  re_free (cset);
+}
+#endif /* RE_ENABLE_I18N */
+
+/* Functions for binary tree operation.  */
+
+/* Create a tree node.  */
+
+static bin_tree_t *
+create_tree (re_dfa_t *dfa, bin_tree_t *left, bin_tree_t *right,
+	     re_token_type_t type)
+{
+  re_token_t t;
+  t.type = type;
+  return create_token_tree (dfa, left, right, &t);
+}
+
+static bin_tree_t *
+create_token_tree (re_dfa_t *dfa, bin_tree_t *left, bin_tree_t *right,
+		   const re_token_t *token)
+{
+  bin_tree_t *tree;
+  if (BE (dfa->str_tree_storage_idx == BIN_TREE_STORAGE_SIZE, 0))
+    {
+      bin_tree_storage_t *storage = re_malloc (bin_tree_storage_t, 1);
+
+      if (storage == NULL)
+	return NULL;
+      storage->next = dfa->str_tree_storage;
+      dfa->str_tree_storage = storage;
+      dfa->str_tree_storage_idx = 0;
+    }
+  tree = &dfa->str_tree_storage->data[dfa->str_tree_storage_idx++];
+
+  tree->parent = NULL;
+  tree->left = left;
+  tree->right = right;
+  tree->token = *token;
+  tree->token.duplicated = 0;
+  tree->token.opt_subexp = 0;
+  tree->first = NULL;
+  tree->next = NULL;
+  tree->node_idx = -1;
+
+  if (left != NULL)
+    left->parent = tree;
+  if (right != NULL)
+    right->parent = tree;
+  return tree;
+}
+
+/* Mark the tree SRC as an optional subexpression.
+   To be called from preorder or postorder.  */
+
+static reg_errcode_t
+mark_opt_subexp (void *extra, bin_tree_t *node)
+{
+  int idx = (int) (intptr_t) extra;
+  if (node->token.type == SUBEXP && node->token.opr.idx == idx)
+    node->token.opt_subexp = 1;
+
+  return REG_NOERROR;
+}
+
+/* Free the allocated memory inside NODE. */
+
+static void
+free_token (re_token_t *node)
+{
+#ifdef RE_ENABLE_I18N
+  if (node->type == COMPLEX_BRACKET && node->duplicated == 0)
+    free_charset (node->opr.mbcset);
+  else
+#endif /* RE_ENABLE_I18N */
+    if (node->type == SIMPLE_BRACKET && node->duplicated == 0)
+      re_free (node->opr.sbcset);
+}
+
+/* Worker function for tree walking.  Free the allocated memory inside NODE
+   and its children. */
+
+static reg_errcode_t
+free_tree (void *extra, bin_tree_t *node)
+{
+  free_token (&node->token);
+  return REG_NOERROR;
+}
+
+
+/* Duplicate the node SRC, and return new node.  This is a preorder
+   visit similar to the one implemented by the generic visitor, but
+   we need more infrastructure to maintain two parallel trees --- so,
+   it's easier to duplicate.  */
+
+static bin_tree_t *
+duplicate_tree (const bin_tree_t *root, re_dfa_t *dfa)
+{
+  const bin_tree_t *node;
+  bin_tree_t *dup_root;
+  bin_tree_t **p_new = &dup_root, *dup_node = root->parent;
+
+  for (node = root; ; )
+    {
+      /* Create a new tree and link it back to the current parent.  */
+      *p_new = create_token_tree (dfa, NULL, NULL, &node->token);
+      if (*p_new == NULL)
+	return NULL;
+      (*p_new)->parent = dup_node;
+      (*p_new)->token.duplicated = 1;
+      dup_node = *p_new;
+
+      /* Go to the left node, or up and to the right.  */
+      if (node->left)
+	{
+	  node = node->left;
+	  p_new = &dup_node->left;
+	}
+      else
+	{
+	  const bin_tree_t *prev = NULL;
+	  while (node->right == prev || node->right == NULL)
+	    {
+	      prev = node;
+	      node = node->parent;
+	      dup_node = dup_node->parent;
+	      if (!node)
+		return dup_root;
+	    }
+	  node = node->right;
+	  p_new = &dup_node->right;
+	}
+    }
+}
diff --git a/compat/regex/regex.c b/compat/regex/regex.c
new file mode 100644
index 000000000000..f3e03a9eabee
--- /dev/null
+++ b/compat/regex/regex.c
@@ -0,0 +1,87 @@
+/* Extended regular expression matching and search library.
+   Copyright (C) 2002, 2003, 2005 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+/* Make sure no one compiles this code with a C++ compiler.  */
+#ifdef __cplusplus
+# error "This is C code, use a C compiler"
+#endif
+
+#ifdef _LIBC
+/* We have to keep the namespace clean.  */
+# define regfree(preg) __regfree (preg)
+# define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
+# define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
+# define regerror(errcode, preg, errbuf, errbuf_size) \
+	__regerror(errcode, preg, errbuf, errbuf_size)
+# define re_set_registers(bu, re, nu, st, en) \
+	__re_set_registers (bu, re, nu, st, en)
+# define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
+	__re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
+# define re_match(bufp, string, size, pos, regs) \
+	__re_match (bufp, string, size, pos, regs)
+# define re_search(bufp, string, size, startpos, range, regs) \
+	__re_search (bufp, string, size, startpos, range, regs)
+# define re_compile_pattern(pattern, length, bufp) \
+	__re_compile_pattern (pattern, length, bufp)
+# define re_set_syntax(syntax) __re_set_syntax (syntax)
+# define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
+	__re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
+# define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
+
+# include "../locale/localeinfo.h"
+#endif
+
+#if defined (_MSC_VER)
+#include <stdio.h> /* for size_t */
+#endif
+
+/* On some systems, limits.h sets RE_DUP_MAX to a lower value than
+   GNU regex allows.  Include it before <regex.h>, which correctly
+   #undefs RE_DUP_MAX and sets it to the right value.  */
+#include <limits.h>
+#include <stdint.h>
+
+#ifdef GAWK
+#undef alloca
+#define alloca alloca_is_bad_you_should_never_use_it
+#endif
+#include <regex.h>
+#include "regex_internal.h"
+
+#include "regex_internal.c"
+#ifdef GAWK
+#define bool int
+#define true (1)
+#define false (0)
+#endif
+#include "regcomp.c"
+#include "regexec.c"
+
+/* Binary backward compatibility.  */
+#if _LIBC
+# include <shlib-compat.h>
+# if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3)
+link_warning (re_max_failures, "the 're_max_failures' variable is obsolete and will go away.")
+int re_max_failures = 2000;
+# endif
+#endif
diff --git a/compat/regex/regex.h b/compat/regex/regex.h
new file mode 100644
index 000000000000..4d81358a83d0
--- /dev/null
+++ b/compat/regex/regex.h
@@ -0,0 +1,581 @@
+#include <stdio.h>
+#include <stddef.h>
+
+/* Definitions for data structures and routines for the regular
+   expression library.
+   Copyright (C) 1985,1989-93,1995-98,2000,2001,2002,2003,2005,2006,2008
+   Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _REGEX_H
+#define _REGEX_H 1
+
+#ifdef HAVE_STDDEF_H
+#include <stddef.h>
+#endif
+
+#ifdef HAVE_SYS_TYPES_H
+#include <sys/types.h>
+#endif
+
+#ifndef _LIBC
+#define __USE_GNU	1
+#endif
+
+/* Allow the use in C++ code.  */
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* The following two types have to be signed and unsigned integer type
+   wide enough to hold a value of a pointer.  For most ANSI compilers
+   ptrdiff_t and size_t should be likely OK.  Still size of these two
+   types is 2 for Microsoft C.  Ugh... */
+typedef long int s_reg_t;
+typedef unsigned long int active_reg_t;
+
+/* The following bits are used to determine the regexp syntax we
+   recognize.  The set/not-set meanings are chosen so that Emacs syntax
+   remains the value 0.  The bits are given in alphabetical order, and
+   the definitions shifted by one from the previous bit; thus, when we
+   add or remove a bit, only one other definition need change.  */
+typedef unsigned long int reg_syntax_t;
+
+#ifdef __USE_GNU
+/* If this bit is not set, then \ inside a bracket expression is literal.
+   If set, then such a \ quotes the following character.  */
+# define RE_BACKSLASH_ESCAPE_IN_LISTS ((unsigned long int) 1)
+
+/* If this bit is not set, then + and ? are operators, and \+ and \? are
+     literals.
+   If set, then \+ and \? are operators and + and ? are literals.  */
+# define RE_BK_PLUS_QM (RE_BACKSLASH_ESCAPE_IN_LISTS << 1)
+
+/* If this bit is set, then character classes are supported.  They are:
+     [:alpha:], [:upper:], [:lower:],  [:digit:], [:alnum:], [:xdigit:],
+     [:space:], [:print:], [:punct:], [:graph:], and [:cntrl:].
+   If not set, then character classes are not supported.  */
+# define RE_CHAR_CLASSES (RE_BK_PLUS_QM << 1)
+
+/* If this bit is set, then ^ and $ are always anchors (outside bracket
+     expressions, of course).
+   If this bit is not set, then it depends:
+        ^  is an anchor if it is at the beginning of a regular
+           expression or after an open-group or an alternation operator;
+        $  is an anchor if it is at the end of a regular expression, or
+           before a close-group or an alternation operator.
+
+   This bit could be (re)combined with RE_CONTEXT_INDEP_OPS, because
+   POSIX draft 11.2 says that * etc. in leading positions is undefined.
+   We already implemented a previous draft which made those constructs
+   invalid, though, so we haven't changed the code back.  */
+# define RE_CONTEXT_INDEP_ANCHORS (RE_CHAR_CLASSES << 1)
+
+/* If this bit is set, then special characters are always special
+     regardless of where they are in the pattern.
+   If this bit is not set, then special characters are special only in
+     some contexts; otherwise they are ordinary.  Specifically,
+     * + ? and intervals are only special when not after the beginning,
+     open-group, or alternation operator.  */
+# define RE_CONTEXT_INDEP_OPS (RE_CONTEXT_INDEP_ANCHORS << 1)
+
+/* If this bit is set, then *, +, ?, and { cannot be first in an re or
+     immediately after an alternation or begin-group operator.  */
+# define RE_CONTEXT_INVALID_OPS (RE_CONTEXT_INDEP_OPS << 1)
+
+/* If this bit is set, then . matches newline.
+   If not set, then it doesn't.  */
+# define RE_DOT_NEWLINE (RE_CONTEXT_INVALID_OPS << 1)
+
+/* If this bit is set, then . doesn't match NUL.
+   If not set, then it does.  */
+# define RE_DOT_NOT_NULL (RE_DOT_NEWLINE << 1)
+
+/* If this bit is set, nonmatching lists [^...] do not match newline.
+   If not set, they do.  */
+# define RE_HAT_LISTS_NOT_NEWLINE (RE_DOT_NOT_NULL << 1)
+
+/* If this bit is set, either \{...\} or {...} defines an
+     interval, depending on RE_NO_BK_BRACES.
+   If not set, \{, \}, {, and } are literals.  */
+# define RE_INTERVALS (RE_HAT_LISTS_NOT_NEWLINE << 1)
+
+/* If this bit is set, +, ? and | aren't recognized as operators.
+   If not set, they are.  */
+# define RE_LIMITED_OPS (RE_INTERVALS << 1)
+
+/* If this bit is set, newline is an alternation operator.
+   If not set, newline is literal.  */
+# define RE_NEWLINE_ALT (RE_LIMITED_OPS << 1)
+
+/* If this bit is set, then `{...}' defines an interval, and \{ and \}
+     are literals.
+  If not set, then `\{...\}' defines an interval.  */
+# define RE_NO_BK_BRACES (RE_NEWLINE_ALT << 1)
+
+/* If this bit is set, (...) defines a group, and \( and \) are literals.
+   If not set, \(...\) defines a group, and ( and ) are literals.  */
+# define RE_NO_BK_PARENS (RE_NO_BK_BRACES << 1)
+
+/* If this bit is set, then \<digit> matches <digit>.
+   If not set, then \<digit> is a back-reference.  */
+# define RE_NO_BK_REFS (RE_NO_BK_PARENS << 1)
+
+/* If this bit is set, then | is an alternation operator, and \| is literal.
+   If not set, then \| is an alternation operator, and | is literal.  */
+# define RE_NO_BK_VBAR (RE_NO_BK_REFS << 1)
+
+/* If this bit is set, then an ending range point collating higher
+     than the starting range point, as in [z-a], is invalid.
+   If not set, then when ending range point collates higher than the
+     starting range point, the range is ignored.  */
+# define RE_NO_EMPTY_RANGES (RE_NO_BK_VBAR << 1)
+
+/* If this bit is set, then an unmatched ) is ordinary.
+   If not set, then an unmatched ) is invalid.  */
+# define RE_UNMATCHED_RIGHT_PAREN_ORD (RE_NO_EMPTY_RANGES << 1)
+
+/* If this bit is set, succeed as soon as we match the whole pattern,
+   without further backtracking.  */
+# define RE_NO_POSIX_BACKTRACKING (RE_UNMATCHED_RIGHT_PAREN_ORD << 1)
+
+/* If this bit is set, do not process the GNU regex operators.
+   If not set, then the GNU regex operators are recognized. */
+# define RE_NO_GNU_OPS (RE_NO_POSIX_BACKTRACKING << 1)
+
+/* If this bit is set, a syntactically invalid interval is treated as
+   a string of ordinary characters.  For example, the ERE 'a{1' is
+   treated as 'a\{1'.  */
+# define RE_INVALID_INTERVAL_ORD (RE_NO_GNU_OPS << 1)
+
+/* If this bit is set, then ignore case when matching.
+   If not set, then case is significant.  */
+# define RE_ICASE (RE_INVALID_INTERVAL_ORD << 1)
+
+/* This bit is used internally like RE_CONTEXT_INDEP_ANCHORS but only
+   for ^, because it is difficult to scan the regex backwards to find
+   whether ^ should be special.  */
+# define RE_CARET_ANCHORS_HERE (RE_ICASE << 1)
+
+/* If this bit is set, then \{ cannot be first in an bre or
+   immediately after an alternation or begin-group operator.  */
+# define RE_CONTEXT_INVALID_DUP (RE_CARET_ANCHORS_HERE << 1)
+
+/* If this bit is set, then no_sub will be set to 1 during
+   re_compile_pattern.  */
+#define RE_NO_SUB (RE_CONTEXT_INVALID_DUP << 1)
+#endif
+
+/* This global variable defines the particular regexp syntax to use (for
+   some interfaces).  When a regexp is compiled, the syntax used is
+   stored in the pattern buffer, so changing this does not affect
+   already-compiled regexps.  */
+extern reg_syntax_t re_syntax_options;
+
+#ifdef __USE_GNU
+/* Define combinations of the above bits for the standard possibilities.
+   (The [[[ comments delimit what gets put into the Texinfo file, so
+   don't delete them!)  */
+/* [[[begin syntaxes]]] */
+#define RE_SYNTAX_EMACS 0
+
+#define RE_SYNTAX_AWK							\
+  (RE_BACKSLASH_ESCAPE_IN_LISTS   | RE_DOT_NOT_NULL			\
+   | RE_NO_BK_PARENS              | RE_NO_BK_REFS			\
+   | RE_NO_BK_VBAR                | RE_NO_EMPTY_RANGES			\
+   | RE_DOT_NEWLINE		  | RE_CONTEXT_INDEP_ANCHORS		\
+   | RE_UNMATCHED_RIGHT_PAREN_ORD | RE_NO_GNU_OPS)
+
+#define RE_SYNTAX_GNU_AWK						\
+  ((RE_SYNTAX_POSIX_EXTENDED | RE_BACKSLASH_ESCAPE_IN_LISTS		\
+   | RE_INVALID_INTERVAL_ORD)						\
+   & ~(RE_DOT_NOT_NULL | RE_CONTEXT_INDEP_OPS				\
+       | RE_CONTEXT_INVALID_OPS ))
+
+#define RE_SYNTAX_POSIX_AWK						\
+  (RE_SYNTAX_POSIX_EXTENDED | RE_BACKSLASH_ESCAPE_IN_LISTS		\
+   | RE_INTERVALS	    | RE_NO_GNU_OPS				\
+   | RE_INVALID_INTERVAL_ORD)
+
+#define RE_SYNTAX_GREP							\
+  (RE_BK_PLUS_QM              | RE_CHAR_CLASSES				\
+   | RE_HAT_LISTS_NOT_NEWLINE | RE_INTERVALS				\
+   | RE_NEWLINE_ALT)
+
+#define RE_SYNTAX_EGREP							\
+  (RE_CHAR_CLASSES        | RE_CONTEXT_INDEP_ANCHORS			\
+   | RE_CONTEXT_INDEP_OPS | RE_HAT_LISTS_NOT_NEWLINE			\
+   | RE_NEWLINE_ALT       | RE_NO_BK_PARENS				\
+   | RE_NO_BK_VBAR)
+
+#define RE_SYNTAX_POSIX_EGREP						\
+  (RE_SYNTAX_EGREP | RE_INTERVALS | RE_NO_BK_BRACES			\
+   | RE_INVALID_INTERVAL_ORD)
+
+/* P1003.2/D11.2, section 4.20.7.1, lines 5078ff.  */
+#define RE_SYNTAX_ED RE_SYNTAX_POSIX_BASIC
+
+#define RE_SYNTAX_SED RE_SYNTAX_POSIX_BASIC
+
+/* Syntax bits common to both basic and extended POSIX regex syntax.  */
+#define _RE_SYNTAX_POSIX_COMMON						\
+  (RE_CHAR_CLASSES | RE_DOT_NEWLINE      | RE_DOT_NOT_NULL		\
+   | RE_INTERVALS  | RE_NO_EMPTY_RANGES)
+
+#define RE_SYNTAX_POSIX_BASIC						\
+  (_RE_SYNTAX_POSIX_COMMON | RE_BK_PLUS_QM | RE_CONTEXT_INVALID_DUP)
+
+/* Differs from ..._POSIX_BASIC only in that RE_BK_PLUS_QM becomes
+   RE_LIMITED_OPS, i.e., \? \+ \| are not recognized.  Actually, this
+   isn't minimal, since other operators, such as \`, aren't disabled.  */
+#define RE_SYNTAX_POSIX_MINIMAL_BASIC					\
+  (_RE_SYNTAX_POSIX_COMMON | RE_LIMITED_OPS)
+
+#define RE_SYNTAX_POSIX_EXTENDED					\
+  (_RE_SYNTAX_POSIX_COMMON  | RE_CONTEXT_INDEP_ANCHORS			\
+   | RE_CONTEXT_INDEP_OPS   | RE_NO_BK_BRACES				\
+   | RE_NO_BK_PARENS        | RE_NO_BK_VBAR				\
+   | RE_CONTEXT_INVALID_OPS | RE_UNMATCHED_RIGHT_PAREN_ORD)
+
+/* Differs from ..._POSIX_EXTENDED in that RE_CONTEXT_INDEP_OPS is
+   removed and RE_NO_BK_REFS is added.  */
+#define RE_SYNTAX_POSIX_MINIMAL_EXTENDED				\
+  (_RE_SYNTAX_POSIX_COMMON  | RE_CONTEXT_INDEP_ANCHORS			\
+   | RE_CONTEXT_INVALID_OPS | RE_NO_BK_BRACES				\
+   | RE_NO_BK_PARENS        | RE_NO_BK_REFS				\
+   | RE_NO_BK_VBAR	    | RE_UNMATCHED_RIGHT_PAREN_ORD)
+/* [[[end syntaxes]]] */
+
+/* Maximum number of duplicates an interval can allow.  Some systems
+   (erroneously) define this in other header files, but we want our
+   value, so remove any previous define.  */
+# ifdef RE_DUP_MAX
+#  undef RE_DUP_MAX
+# endif
+/* If sizeof(int) == 2, then ((1 << 15) - 1) overflows.  */
+# define RE_DUP_MAX (0x7fff)
+#endif
+
+
+/* POSIX `cflags' bits (i.e., information for `regcomp').  */
+
+/* If this bit is set, then use extended regular expression syntax.
+   If not set, then use basic regular expression syntax.  */
+#define REG_EXTENDED 1
+
+/* If this bit is set, then ignore case when matching.
+   If not set, then case is significant.  */
+#define REG_ICASE (REG_EXTENDED << 1)
+
+/* If this bit is set, then anchors do not match at newline
+     characters in the string.
+   If not set, then anchors do match at newlines.  */
+#define REG_NEWLINE (REG_ICASE << 1)
+
+/* If this bit is set, then report only success or fail in regexec.
+   If not set, then returns differ between not matching and errors.  */
+#define REG_NOSUB (REG_NEWLINE << 1)
+
+
+/* POSIX `eflags' bits (i.e., information for regexec).  */
+
+/* If this bit is set, then the beginning-of-line operator doesn't match
+     the beginning of the string (presumably because it's not the
+     beginning of a line).
+   If not set, then the beginning-of-line operator does match the
+     beginning of the string.  */
+#define REG_NOTBOL 1
+
+/* Like REG_NOTBOL, except for the end-of-line.  */
+#define REG_NOTEOL (1 << 1)
+
+/* Use PMATCH[0] to delimit the start and end of the search in the
+   buffer.  */
+#define REG_STARTEND (1 << 2)
+
+
+/* If any error codes are removed, changed, or added, update the
+   `re_error_msg' table in regex.c.  */
+typedef enum
+{
+#if defined _XOPEN_SOURCE || defined __USE_XOPEN2K
+  REG_ENOSYS = -1,	/* This will never happen for this implementation.  */
+#endif
+
+  REG_NOERROR = 0,	/* Success.  */
+  REG_NOMATCH,		/* Didn't find a match (for regexec).  */
+
+  /* POSIX regcomp return error codes.  (In the order listed in the
+     standard.)  */
+  REG_BADPAT,		/* Invalid pattern.  */
+  REG_ECOLLATE,		/* Inalid collating element.  */
+  REG_ECTYPE,		/* Invalid character class name.  */
+  REG_EESCAPE,		/* Trailing backslash.  */
+  REG_ESUBREG,		/* Invalid back reference.  */
+  REG_EBRACK,		/* Unmatched left bracket.  */
+  REG_EPAREN,		/* Parenthesis imbalance.  */
+  REG_EBRACE,		/* Unmatched \{.  */
+  REG_BADBR,		/* Invalid contents of \{\}.  */
+  REG_ERANGE,		/* Invalid range end.  */
+  REG_ESPACE,		/* Ran out of memory.  */
+  REG_BADRPT,		/* No preceding re for repetition op.  */
+
+  /* Error codes we've added.  */
+  REG_EEND,		/* Premature end.  */
+  REG_ESIZE,		/* Compiled pattern bigger than 2^16 bytes.  */
+  REG_ERPAREN		/* Unmatched ) or \); not returned from regcomp.  */
+} reg_errcode_t;
+
+/* This data structure represents a compiled pattern.  Before calling
+   the pattern compiler, the fields `buffer', `allocated', `fastmap',
+   `translate', and `no_sub' can be set.  After the pattern has been
+   compiled, the `re_nsub' field is available.  All other fields are
+   private to the regex routines.  */
+
+#ifndef RE_TRANSLATE_TYPE
+# define __RE_TRANSLATE_TYPE unsigned char *
+# ifdef __USE_GNU
+#  define RE_TRANSLATE_TYPE __RE_TRANSLATE_TYPE
+# endif
+#endif
+
+#ifdef __USE_GNU
+# define __REPB_PREFIX(name) name
+#else
+# define __REPB_PREFIX(name) __##name
+#endif
+
+struct re_pattern_buffer
+{
+  /* Space that holds the compiled pattern.  It is declared as
+     `unsigned char *' because its elements are sometimes used as
+     array indexes.  */
+  unsigned char *__REPB_PREFIX(buffer);
+
+  /* Number of bytes to which `buffer' points.  */
+  unsigned long int __REPB_PREFIX(allocated);
+
+  /* Number of bytes actually used in `buffer'.  */
+  unsigned long int __REPB_PREFIX(used);
+
+  /* Syntax setting with which the pattern was compiled.  */
+  reg_syntax_t __REPB_PREFIX(syntax);
+
+  /* Pointer to a fastmap, if any, otherwise zero.  re_search uses the
+     fastmap, if there is one, to skip over impossible starting points
+     for matches.  */
+  char *__REPB_PREFIX(fastmap);
+
+  /* Either a translate table to apply to all characters before
+     comparing them, or zero for no translation.  The translation is
+     applied to a pattern when it is compiled and to a string when it
+     is matched.  */
+  __RE_TRANSLATE_TYPE __REPB_PREFIX(translate);
+
+  /* Number of subexpressions found by the compiler.  */
+  size_t re_nsub;
+
+  /* Zero if this pattern cannot match the empty string, one else.
+     Well, in truth it's used only in `re_search_2', to see whether or
+     not we should use the fastmap, so we don't set this absolutely
+     perfectly; see `re_compile_fastmap' (the `duplicate' case).  */
+  unsigned __REPB_PREFIX(can_be_null) : 1;
+
+  /* If REGS_UNALLOCATED, allocate space in the `regs' structure
+     for `max (RE_NREGS, re_nsub + 1)' groups.
+     If REGS_REALLOCATE, reallocate space if necessary.
+     If REGS_FIXED, use what's there.  */
+#ifdef __USE_GNU
+# define REGS_UNALLOCATED 0
+# define REGS_REALLOCATE 1
+# define REGS_FIXED 2
+#endif
+  unsigned __REPB_PREFIX(regs_allocated) : 2;
+
+  /* Set to zero when `regex_compile' compiles a pattern; set to one
+     by `re_compile_fastmap' if it updates the fastmap.  */
+  unsigned __REPB_PREFIX(fastmap_accurate) : 1;
+
+  /* If set, `re_match_2' does not return information about
+     subexpressions.  */
+  unsigned __REPB_PREFIX(no_sub) : 1;
+
+  /* If set, a beginning-of-line anchor doesn't match at the beginning
+     of the string.  */
+  unsigned __REPB_PREFIX(not_bol) : 1;
+
+  /* Similarly for an end-of-line anchor.  */
+  unsigned __REPB_PREFIX(not_eol) : 1;
+
+  /* If true, an anchor at a newline matches.  */
+  unsigned __REPB_PREFIX(newline_anchor) : 1;
+};
+
+typedef struct re_pattern_buffer regex_t;
+
+/* Type for byte offsets within the string.  POSIX mandates this.  */
+typedef int regoff_t;
+
+
+#ifdef __USE_GNU
+/* This is the structure we store register match data in.  See
+   regex.texinfo for a full description of what registers match.  */
+struct re_registers
+{
+  unsigned num_regs;
+  regoff_t *start;
+  regoff_t *end;
+};
+
+
+/* If `regs_allocated' is REGS_UNALLOCATED in the pattern buffer,
+   `re_match_2' returns information about at least this many registers
+   the first time a `regs' structure is passed.  */
+# ifndef RE_NREGS
+#  define RE_NREGS 30
+# endif
+#endif
+
+
+/* POSIX specification for registers.  Aside from the different names than
+   `re_registers', POSIX uses an array of structures, instead of a
+   structure of arrays.  */
+typedef struct
+{
+  regoff_t rm_so;  /* Byte offset from string's start to substring's start.  */
+  regoff_t rm_eo;  /* Byte offset from string's start to substring's end.  */
+} regmatch_t;
+
+/* Declarations for routines.  */
+
+#ifdef __USE_GNU
+/* Sets the current default syntax to SYNTAX, and return the old syntax.
+   You can also simply assign to the `re_syntax_options' variable.  */
+extern reg_syntax_t re_set_syntax (reg_syntax_t __syntax);
+
+/* Compile the regular expression PATTERN, with length LENGTH
+   and syntax given by the global `re_syntax_options', into the buffer
+   BUFFER.  Return NULL if successful, and an error string if not.  */
+extern const char *re_compile_pattern (const char *__pattern, size_t __length,
+				       struct re_pattern_buffer *__buffer);
+
+
+/* Compile a fastmap for the compiled pattern in BUFFER; used to
+   accelerate searches.  Return 0 if successful and -2 if was an
+   internal error.  */
+extern int re_compile_fastmap (struct re_pattern_buffer *__buffer);
+
+
+/* Search in the string STRING (with length LENGTH) for the pattern
+   compiled into BUFFER.  Start searching at position START, for RANGE
+   characters.  Return the starting position of the match, -1 for no
+   match, or -2 for an internal error.  Also return register
+   information in REGS (if REGS and BUFFER->no_sub are nonzero).  */
+extern int re_search (struct re_pattern_buffer *__buffer, const char *__cstring,
+		      int __length, int __start, int __range,
+		      struct re_registers *__regs);
+
+
+/* Like `re_search', but search in the concatenation of STRING1 and
+   STRING2.  Also, stop searching at index START + STOP.  */
+extern int re_search_2 (struct re_pattern_buffer *__buffer,
+			const char *__string1, int __length1,
+			const char *__string2, int __length2, int __start,
+			int __range, struct re_registers *__regs, int __stop);
+
+
+/* Like `re_search', but return how many characters in STRING the regexp
+   in BUFFER matched, starting at position START.  */
+extern int re_match (struct re_pattern_buffer *__buffer, const char *__cstring,
+		     int __length, int __start, struct re_registers *__regs);
+
+
+/* Relates to `re_match' as `re_search_2' relates to `re_search'.  */
+extern int re_match_2 (struct re_pattern_buffer *__buffer,
+		       const char *__string1, int __length1,
+		       const char *__string2, int __length2, int __start,
+		       struct re_registers *__regs, int __stop);
+
+
+/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
+   ENDS.  Subsequent matches using BUFFER and REGS will use this memory
+   for recording register information.  STARTS and ENDS must be
+   allocated with malloc, and must each be at least `NUM_REGS * sizeof
+   (regoff_t)' bytes long.
+
+   If NUM_REGS == 0, then subsequent matches should allocate their own
+   register data.
+
+   Unless this function is called, the first search or match using
+   PATTERN_BUFFER will allocate its own register data, without
+   freeing the old data.  */
+extern void re_set_registers (struct re_pattern_buffer *__buffer,
+			      struct re_registers *__regs,
+			      unsigned int __num_regs,
+			      regoff_t *__starts, regoff_t *__ends);
+#endif	/* Use GNU */
+
+#if defined _REGEX_RE_COMP || (defined _LIBC && defined __USE_BSD)
+# ifndef _CRAY
+/* 4.2 bsd compatibility.  */
+extern char *re_comp (const char *);
+extern int re_exec (const char *);
+# endif
+#endif
+
+/* GCC 2.95 and later have "__restrict"; C99 compilers have
+   "restrict", and "configure" may have defined "restrict".  */
+#ifndef __restrict
+# if ! (2 < __GNUC__ || (2 == __GNUC__ && 95 <= __GNUC_MINOR__))
+#  if defined restrict || 199901L <= __STDC_VERSION__
+#   define __restrict restrict
+#  else
+#   define __restrict
+#  endif
+# endif
+#endif
+/* gcc 3.1 and up support the [restrict] syntax.  */
+#ifndef __restrict_arr
+# if (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1)) \
+     && !defined __GNUG__
+#  define __restrict_arr __restrict
+# else
+#  define __restrict_arr
+# endif
+#endif
+
+/* POSIX compatibility.  */
+extern int regcomp (regex_t *__restrict __preg,
+		    const char *__restrict __pattern,
+		    int __cflags);
+
+extern int regexec (const regex_t *__restrict __preg,
+		    const char *__restrict __cstring, size_t __nmatch,
+		    regmatch_t __pmatch[__restrict_arr],
+		    int __eflags);
+
+extern size_t regerror (int __errcode, const regex_t *__restrict __preg,
+			char *__restrict __errbuf, size_t __errbuf_size);
+
+extern void regfree (regex_t *__preg);
+
+
+#ifdef __cplusplus
+}
+#endif	/* C++ */
+
+#endif /* regex.h */
diff --git a/compat/regex/regex_internal.c b/compat/regex/regex_internal.c
new file mode 100644
index 000000000000..59bf151336c2
--- /dev/null
+++ b/compat/regex/regex_internal.c
@@ -0,0 +1,1743 @@
+/* Extended regular expression matching and search library.
+   Copyright (C) 2002-2006, 2010 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+static void re_string_construct_common (const char *str, int len,
+					re_string_t *pstr,
+					RE_TRANSLATE_TYPE trans, int icase,
+					const re_dfa_t *dfa) internal_function;
+static re_dfastate_t *create_ci_newstate (const re_dfa_t *dfa,
+					  const re_node_set *nodes,
+					  unsigned int hash) internal_function;
+static re_dfastate_t *create_cd_newstate (const re_dfa_t *dfa,
+					  const re_node_set *nodes,
+					  unsigned int context,
+					  unsigned int hash) internal_function;
+
+#ifdef GAWK
+#undef MAX	/* safety */
+static int
+MAX(size_t a, size_t b)
+{
+	return (a > b ? a : b);
+}
+#endif
+
+/* Functions for string operation.  */
+
+/* This function allocate the buffers.  It is necessary to call
+   re_string_reconstruct before using the object.  */
+
+static reg_errcode_t
+internal_function
+re_string_allocate (re_string_t *pstr, const char *str, int len, int init_len,
+		    RE_TRANSLATE_TYPE trans, int icase, const re_dfa_t *dfa)
+{
+  reg_errcode_t ret;
+  int init_buf_len;
+
+  /* Ensure at least one character fits into the buffers.  */
+  if (init_len < dfa->mb_cur_max)
+    init_len = dfa->mb_cur_max;
+  init_buf_len = (len + 1 < init_len) ? len + 1: init_len;
+  re_string_construct_common (str, len, pstr, trans, icase, dfa);
+
+  ret = re_string_realloc_buffers (pstr, init_buf_len);
+  if (BE (ret != REG_NOERROR, 0))
+    return ret;
+
+  pstr->word_char = dfa->word_char;
+  pstr->word_ops_used = dfa->word_ops_used;
+  pstr->mbs = pstr->mbs_allocated ? pstr->mbs : (unsigned char *) str;
+  pstr->valid_len = (pstr->mbs_allocated || dfa->mb_cur_max > 1) ? 0 : len;
+  pstr->valid_raw_len = pstr->valid_len;
+  return REG_NOERROR;
+}
+
+/* This function allocate the buffers, and initialize them.  */
+
+static reg_errcode_t
+internal_function
+re_string_construct (re_string_t *pstr, const char *str, int len,
+		     RE_TRANSLATE_TYPE trans, int icase, const re_dfa_t *dfa)
+{
+  reg_errcode_t ret;
+  memset (pstr, '\0', sizeof (re_string_t));
+  re_string_construct_common (str, len, pstr, trans, icase, dfa);
+
+  if (len > 0)
+    {
+      ret = re_string_realloc_buffers (pstr, len + 1);
+      if (BE (ret != REG_NOERROR, 0))
+	return ret;
+    }
+  pstr->mbs = pstr->mbs_allocated ? pstr->mbs : (unsigned char *) str;
+
+  if (icase)
+    {
+#ifdef RE_ENABLE_I18N
+      if (dfa->mb_cur_max > 1)
+	{
+	  while (1)
+	    {
+	      ret = build_wcs_upper_buffer (pstr);
+	      if (BE (ret != REG_NOERROR, 0))
+		return ret;
+	      if (pstr->valid_raw_len >= len)
+		break;
+	      if (pstr->bufs_len > pstr->valid_len + dfa->mb_cur_max)
+		break;
+	      ret = re_string_realloc_buffers (pstr, pstr->bufs_len * 2);
+	      if (BE (ret != REG_NOERROR, 0))
+		return ret;
+	    }
+	}
+      else
+#endif /* RE_ENABLE_I18N  */
+	build_upper_buffer (pstr);
+    }
+  else
+    {
+#ifdef RE_ENABLE_I18N
+      if (dfa->mb_cur_max > 1)
+	build_wcs_buffer (pstr);
+      else
+#endif /* RE_ENABLE_I18N  */
+	{
+	  if (trans != NULL)
+	    re_string_translate_buffer (pstr);
+	  else
+	    {
+	      pstr->valid_len = pstr->bufs_len;
+	      pstr->valid_raw_len = pstr->bufs_len;
+	    }
+	}
+    }
+
+  return REG_NOERROR;
+}
+
+/* Helper functions for re_string_allocate, and re_string_construct.  */
+
+static reg_errcode_t
+internal_function
+re_string_realloc_buffers (re_string_t *pstr, int new_buf_len)
+{
+#ifdef RE_ENABLE_I18N
+  if (pstr->mb_cur_max > 1)
+    {
+      wint_t *new_wcs;
+
+      /* Avoid overflow in realloc.  */
+      const size_t max_object_size = MAX (sizeof (wint_t), sizeof (int));
+      if (BE (SIZE_MAX / max_object_size < new_buf_len, 0))
+	return REG_ESPACE;
+
+      new_wcs = re_realloc (pstr->wcs, wint_t, new_buf_len);
+      if (BE (new_wcs == NULL, 0))
+	return REG_ESPACE;
+      pstr->wcs = new_wcs;
+      if (pstr->offsets != NULL)
+	{
+	  int *new_offsets = re_realloc (pstr->offsets, int, new_buf_len);
+	  if (BE (new_offsets == NULL, 0))
+	    return REG_ESPACE;
+	  pstr->offsets = new_offsets;
+	}
+    }
+#endif /* RE_ENABLE_I18N  */
+  if (pstr->mbs_allocated)
+    {
+      unsigned char *new_mbs = re_realloc (pstr->mbs, unsigned char,
+					   new_buf_len);
+      if (BE (new_mbs == NULL, 0))
+	return REG_ESPACE;
+      pstr->mbs = new_mbs;
+    }
+  pstr->bufs_len = new_buf_len;
+  return REG_NOERROR;
+}
+
+
+static void
+internal_function
+re_string_construct_common (const char *str, int len, re_string_t *pstr,
+			    RE_TRANSLATE_TYPE trans, int icase,
+			    const re_dfa_t *dfa)
+{
+  pstr->raw_mbs = (const unsigned char *) str;
+  pstr->len = len;
+  pstr->raw_len = len;
+  pstr->trans = trans;
+  pstr->icase = icase ? 1 : 0;
+  pstr->mbs_allocated = (trans != NULL || icase);
+  pstr->mb_cur_max = dfa->mb_cur_max;
+  pstr->is_utf8 = dfa->is_utf8;
+  pstr->map_notascii = dfa->map_notascii;
+  pstr->stop = pstr->len;
+  pstr->raw_stop = pstr->stop;
+}
+
+#ifdef RE_ENABLE_I18N
+
+/* Build wide character buffer PSTR->WCS.
+   If the byte sequence of the string are:
+     <mb1>(0), <mb1>(1), <mb2>(0), <mb2>(1), <sb3>
+   Then wide character buffer will be:
+     <wc1>   , WEOF    , <wc2>   , WEOF    , <wc3>
+   We use WEOF for padding, they indicate that the position isn't
+   a first byte of a multibyte character.
+
+   Note that this function assumes PSTR->VALID_LEN elements are already
+   built and starts from PSTR->VALID_LEN.  */
+
+static void
+internal_function
+build_wcs_buffer (re_string_t *pstr)
+{
+#ifdef _LIBC
+  unsigned char buf[MB_LEN_MAX];
+  assert (MB_LEN_MAX >= pstr->mb_cur_max);
+#else
+  unsigned char buf[64];
+#endif
+  mbstate_t prev_st;
+  int byte_idx, end_idx, remain_len;
+  size_t mbclen;
+
+  /* Build the buffers from pstr->valid_len to either pstr->len or
+     pstr->bufs_len.  */
+  end_idx = (pstr->bufs_len > pstr->len) ? pstr->len : pstr->bufs_len;
+  for (byte_idx = pstr->valid_len; byte_idx < end_idx;)
+    {
+      wchar_t wc;
+      const char *p;
+
+      remain_len = end_idx - byte_idx;
+      prev_st = pstr->cur_state;
+      /* Apply the translation if we need.  */
+      if (BE (pstr->trans != NULL, 0))
+	{
+	  int i, ch;
+
+	  for (i = 0; i < pstr->mb_cur_max && i < remain_len; ++i)
+	    {
+	      ch = pstr->raw_mbs [pstr->raw_mbs_idx + byte_idx + i];
+	      buf[i] = pstr->mbs[byte_idx + i] = pstr->trans[ch];
+	    }
+	  p = (const char *) buf;
+	}
+      else
+	p = (const char *) pstr->raw_mbs + pstr->raw_mbs_idx + byte_idx;
+      mbclen = __mbrtowc (&wc, p, remain_len, &pstr->cur_state);
+      if (BE (mbclen == (size_t) -2, 0))
+	{
+	  /* The buffer doesn't have enough space, finish to build.  */
+	  pstr->cur_state = prev_st;
+	  break;
+	}
+      else if (BE (mbclen == (size_t) -1 || mbclen == 0, 0))
+	{
+	  /* We treat these cases as a singlebyte character.  */
+	  mbclen = 1;
+	  wc = (wchar_t) pstr->raw_mbs[pstr->raw_mbs_idx + byte_idx];
+	  if (BE (pstr->trans != NULL, 0))
+	    wc = pstr->trans[wc];
+	  pstr->cur_state = prev_st;
+	}
+
+      /* Write wide character and padding.  */
+      pstr->wcs[byte_idx++] = wc;
+      /* Write paddings.  */
+      for (remain_len = byte_idx + mbclen - 1; byte_idx < remain_len ;)
+	pstr->wcs[byte_idx++] = WEOF;
+    }
+  pstr->valid_len = byte_idx;
+  pstr->valid_raw_len = byte_idx;
+}
+
+/* Build wide character buffer PSTR->WCS like build_wcs_buffer,
+   but for REG_ICASE.  */
+
+static reg_errcode_t
+internal_function
+build_wcs_upper_buffer (re_string_t *pstr)
+{
+  mbstate_t prev_st;
+  int src_idx, byte_idx, end_idx, remain_len;
+  size_t mbclen;
+#ifdef _LIBC
+  char buf[MB_LEN_MAX];
+  assert (MB_LEN_MAX >= pstr->mb_cur_max);
+#else
+  char buf[64];
+#endif
+
+  byte_idx = pstr->valid_len;
+  end_idx = (pstr->bufs_len > pstr->len) ? pstr->len : pstr->bufs_len;
+
+  /* The following optimization assumes that ASCII characters can be
+     mapped to wide characters with a simple cast.  */
+  if (! pstr->map_notascii && pstr->trans == NULL && !pstr->offsets_needed)
+    {
+      while (byte_idx < end_idx)
+	{
+	  wchar_t wc;
+
+	  if (isascii (pstr->raw_mbs[pstr->raw_mbs_idx + byte_idx])
+	      && mbsinit (&pstr->cur_state))
+	    {
+	      /* In case of a singlebyte character.  */
+	      pstr->mbs[byte_idx]
+		= toupper (pstr->raw_mbs[pstr->raw_mbs_idx + byte_idx]);
+	      /* The next step uses the assumption that wchar_t is encoded
+		 ASCII-safe: all ASCII values can be converted like this.  */
+	      pstr->wcs[byte_idx] = (wchar_t) pstr->mbs[byte_idx];
+	      ++byte_idx;
+	      continue;
+	    }
+
+	  remain_len = end_idx - byte_idx;
+	  prev_st = pstr->cur_state;
+	  mbclen = __mbrtowc (&wc,
+			      ((const char *) pstr->raw_mbs + pstr->raw_mbs_idx
+			       + byte_idx), remain_len, &pstr->cur_state);
+	  if (BE (mbclen + 2 > 2, 1))
+	    {
+	      wchar_t wcu = wc;
+	      if (iswlower (wc))
+		{
+		  size_t mbcdlen;
+
+		  wcu = towupper (wc);
+		  mbcdlen = wcrtomb (buf, wcu, &prev_st);
+		  if (BE (mbclen == mbcdlen, 1))
+		    memcpy (pstr->mbs + byte_idx, buf, mbclen);
+		  else
+		    {
+		      src_idx = byte_idx;
+		      goto offsets_needed;
+		    }
+		}
+	      else
+		memcpy (pstr->mbs + byte_idx,
+			pstr->raw_mbs + pstr->raw_mbs_idx + byte_idx, mbclen);
+	      pstr->wcs[byte_idx++] = wcu;
+	      /* Write paddings.  */
+	      for (remain_len = byte_idx + mbclen - 1; byte_idx < remain_len ;)
+		pstr->wcs[byte_idx++] = WEOF;
+	    }
+	  else if (mbclen == (size_t) -1 || mbclen == 0)
+	    {
+	      /* It is an invalid character or '\0'.  Just use the byte.  */
+	      int ch = pstr->raw_mbs[pstr->raw_mbs_idx + byte_idx];
+	      pstr->mbs[byte_idx] = ch;
+	      /* And also cast it to wide char.  */
+	      pstr->wcs[byte_idx++] = (wchar_t) ch;
+	      if (BE (mbclen == (size_t) -1, 0))
+		pstr->cur_state = prev_st;
+	    }
+	  else
+	    {
+	      /* The buffer doesn't have enough space, finish to build.  */
+	      pstr->cur_state = prev_st;
+	      break;
+	    }
+	}
+      pstr->valid_len = byte_idx;
+      pstr->valid_raw_len = byte_idx;
+      return REG_NOERROR;
+    }
+  else
+    for (src_idx = pstr->valid_raw_len; byte_idx < end_idx;)
+      {
+	wchar_t wc;
+	const char *p;
+      offsets_needed:
+	remain_len = end_idx - byte_idx;
+	prev_st = pstr->cur_state;
+	if (BE (pstr->trans != NULL, 0))
+	  {
+	    int i, ch;
+
+	    for (i = 0; i < pstr->mb_cur_max && i < remain_len; ++i)
+	      {
+		ch = pstr->raw_mbs [pstr->raw_mbs_idx + src_idx + i];
+		buf[i] = pstr->trans[ch];
+	      }
+	    p = (const char *) buf;
+	  }
+	else
+	  p = (const char *) pstr->raw_mbs + pstr->raw_mbs_idx + src_idx;
+	mbclen = __mbrtowc (&wc, p, remain_len, &pstr->cur_state);
+	if (BE (mbclen + 2 > 2, 1))
+	  {
+	    wchar_t wcu = wc;
+	    if (iswlower (wc))
+	      {
+		size_t mbcdlen;
+
+		wcu = towupper (wc);
+		mbcdlen = wcrtomb ((char *) buf, wcu, &prev_st);
+		if (BE (mbclen == mbcdlen, 1))
+		  memcpy (pstr->mbs + byte_idx, buf, mbclen);
+		else if (mbcdlen != (size_t) -1)
+		  {
+		    size_t i;
+
+		    if (byte_idx + mbcdlen > pstr->bufs_len)
+		      {
+			pstr->cur_state = prev_st;
+			break;
+		      }
+
+		    if (pstr->offsets == NULL)
+		      {
+			pstr->offsets = re_malloc (int, pstr->bufs_len);
+
+			if (pstr->offsets == NULL)
+			  return REG_ESPACE;
+		      }
+		    if (!pstr->offsets_needed)
+		      {
+			for (i = 0; i < (size_t) byte_idx; ++i)
+			  pstr->offsets[i] = i;
+			pstr->offsets_needed = 1;
+		      }
+
+		    memcpy (pstr->mbs + byte_idx, buf, mbcdlen);
+		    pstr->wcs[byte_idx] = wcu;
+		    pstr->offsets[byte_idx] = src_idx;
+		    for (i = 1; i < mbcdlen; ++i)
+		      {
+			pstr->offsets[byte_idx + i]
+			  = src_idx + (i < mbclen ? i : mbclen - 1);
+			pstr->wcs[byte_idx + i] = WEOF;
+		      }
+		    pstr->len += mbcdlen - mbclen;
+		    if (pstr->raw_stop > src_idx)
+		      pstr->stop += mbcdlen - mbclen;
+		    end_idx = (pstr->bufs_len > pstr->len)
+			      ? pstr->len : pstr->bufs_len;
+		    byte_idx += mbcdlen;
+		    src_idx += mbclen;
+		    continue;
+		  }
+		else
+		  memcpy (pstr->mbs + byte_idx, p, mbclen);
+	      }
+	    else
+	      memcpy (pstr->mbs + byte_idx, p, mbclen);
+
+	    if (BE (pstr->offsets_needed != 0, 0))
+	      {
+		size_t i;
+		for (i = 0; i < mbclen; ++i)
+		  pstr->offsets[byte_idx + i] = src_idx + i;
+	      }
+	    src_idx += mbclen;
+
+	    pstr->wcs[byte_idx++] = wcu;
+	    /* Write paddings.  */
+	    for (remain_len = byte_idx + mbclen - 1; byte_idx < remain_len ;)
+	      pstr->wcs[byte_idx++] = WEOF;
+	  }
+	else if (mbclen == (size_t) -1 || mbclen == 0)
+	  {
+	    /* It is an invalid character or '\0'.  Just use the byte.  */
+	    int ch = pstr->raw_mbs[pstr->raw_mbs_idx + src_idx];
+
+	    if (BE (pstr->trans != NULL, 0))
+	      ch = pstr->trans [ch];
+	    pstr->mbs[byte_idx] = ch;
+
+	    if (BE (pstr->offsets_needed != 0, 0))
+	      pstr->offsets[byte_idx] = src_idx;
+	    ++src_idx;
+
+	    /* And also cast it to wide char.  */
+	    pstr->wcs[byte_idx++] = (wchar_t) ch;
+	    if (BE (mbclen == (size_t) -1, 0))
+	      pstr->cur_state = prev_st;
+	  }
+	else
+	  {
+	    /* The buffer doesn't have enough space, finish to build.  */
+	    pstr->cur_state = prev_st;
+	    break;
+	  }
+      }
+  pstr->valid_len = byte_idx;
+  pstr->valid_raw_len = src_idx;
+  return REG_NOERROR;
+}
+
+/* Skip characters until the index becomes greater than NEW_RAW_IDX.
+   Return the index.  */
+
+static int
+internal_function
+re_string_skip_chars (re_string_t *pstr, int new_raw_idx, wint_t *last_wc)
+{
+  mbstate_t prev_st;
+  int rawbuf_idx;
+  size_t mbclen;
+  wint_t wc = WEOF;
+
+  /* Skip the characters which are not necessary to check.  */
+  for (rawbuf_idx = pstr->raw_mbs_idx + pstr->valid_raw_len;
+       rawbuf_idx < new_raw_idx;)
+    {
+      wchar_t wc2;
+      int remain_len = pstr->len - rawbuf_idx;
+      prev_st = pstr->cur_state;
+      mbclen = __mbrtowc (&wc2, (const char *) pstr->raw_mbs + rawbuf_idx,
+			  remain_len, &pstr->cur_state);
+      if (BE (mbclen == (size_t) -2 || mbclen == (size_t) -1 || mbclen == 0, 0))
+	{
+	  /* We treat these cases as a single byte character.  */
+	  if (mbclen == 0 || remain_len == 0)
+	    wc = L'\0';
+	  else
+	    wc = *(unsigned char *) (pstr->raw_mbs + rawbuf_idx);
+	  mbclen = 1;
+	  pstr->cur_state = prev_st;
+	}
+      else
+	wc = (wint_t) wc2;
+      /* Then proceed the next character.  */
+      rawbuf_idx += mbclen;
+    }
+  *last_wc = (wint_t) wc;
+  return rawbuf_idx;
+}
+#endif /* RE_ENABLE_I18N  */
+
+/* Build the buffer PSTR->MBS, and apply the translation if we need.
+   This function is used in case of REG_ICASE.  */
+
+static void
+internal_function
+build_upper_buffer (re_string_t *pstr)
+{
+  int char_idx, end_idx;
+  end_idx = (pstr->bufs_len > pstr->len) ? pstr->len : pstr->bufs_len;
+
+  for (char_idx = pstr->valid_len; char_idx < end_idx; ++char_idx)
+    {
+      int ch = pstr->raw_mbs[pstr->raw_mbs_idx + char_idx];
+      if (BE (pstr->trans != NULL, 0))
+	ch = pstr->trans[ch];
+      if (islower (ch))
+	pstr->mbs[char_idx] = toupper (ch);
+      else
+	pstr->mbs[char_idx] = ch;
+    }
+  pstr->valid_len = char_idx;
+  pstr->valid_raw_len = char_idx;
+}
+
+/* Apply TRANS to the buffer in PSTR.  */
+
+static void
+internal_function
+re_string_translate_buffer (re_string_t *pstr)
+{
+  int buf_idx, end_idx;
+  end_idx = (pstr->bufs_len > pstr->len) ? pstr->len : pstr->bufs_len;
+
+  for (buf_idx = pstr->valid_len; buf_idx < end_idx; ++buf_idx)
+    {
+      int ch = pstr->raw_mbs[pstr->raw_mbs_idx + buf_idx];
+      pstr->mbs[buf_idx] = pstr->trans[ch];
+    }
+
+  pstr->valid_len = buf_idx;
+  pstr->valid_raw_len = buf_idx;
+}
+
+/* This function re-construct the buffers.
+   Concretely, convert to wide character in case of pstr->mb_cur_max > 1,
+   convert to upper case in case of REG_ICASE, apply translation.  */
+
+static reg_errcode_t
+internal_function
+re_string_reconstruct (re_string_t *pstr, int idx, int eflags)
+{
+  int offset = idx - pstr->raw_mbs_idx;
+  if (BE (offset < 0, 0))
+    {
+      /* Reset buffer.  */
+#ifdef RE_ENABLE_I18N
+      if (pstr->mb_cur_max > 1)
+	memset (&pstr->cur_state, '\0', sizeof (mbstate_t));
+#endif /* RE_ENABLE_I18N */
+      pstr->len = pstr->raw_len;
+      pstr->stop = pstr->raw_stop;
+      pstr->valid_len = 0;
+      pstr->raw_mbs_idx = 0;
+      pstr->valid_raw_len = 0;
+      pstr->offsets_needed = 0;
+      pstr->tip_context = ((eflags & REG_NOTBOL) ? CONTEXT_BEGBUF
+			   : CONTEXT_NEWLINE | CONTEXT_BEGBUF);
+      if (!pstr->mbs_allocated)
+	pstr->mbs = (unsigned char *) pstr->raw_mbs;
+      offset = idx;
+    }
+
+  if (BE (offset != 0, 1))
+    {
+      /* Should the already checked characters be kept?  */
+      if (BE (offset < pstr->valid_raw_len, 1))
+	{
+	  /* Yes, move them to the front of the buffer.  */
+#ifdef RE_ENABLE_I18N
+	  if (BE (pstr->offsets_needed, 0))
+	    {
+	      int low = 0, high = pstr->valid_len, mid;
+	      do
+		{
+		  mid = low + (high - low) / 2;
+		  if (pstr->offsets[mid] > offset)
+		    high = mid;
+		  else if (pstr->offsets[mid] < offset)
+		    low = mid + 1;
+		  else
+		    break;
+		}
+	      while (low < high);
+	      if (pstr->offsets[mid] < offset)
+		++mid;
+	      pstr->tip_context = re_string_context_at (pstr, mid - 1,
+							eflags);
+	      /* This can be quite complicated, so handle specially
+		 only the common and easy case where the character with
+		 different length representation of lower and upper
+		 case is present at or after offset.  */
+	      if (pstr->valid_len > offset
+		  && mid == offset && pstr->offsets[mid] == offset)
+		{
+		  memmove (pstr->wcs, pstr->wcs + offset,
+			   (pstr->valid_len - offset) * sizeof (wint_t));
+		  memmove (pstr->mbs, pstr->mbs + offset, pstr->valid_len - offset);
+		  pstr->valid_len -= offset;
+		  pstr->valid_raw_len -= offset;
+		  for (low = 0; low < pstr->valid_len; low++)
+		    pstr->offsets[low] = pstr->offsets[low + offset] - offset;
+		}
+	      else
+		{
+		  /* Otherwise, just find out how long the partial multibyte
+		     character at offset is and fill it with WEOF/255.  */
+		  pstr->len = pstr->raw_len - idx + offset;
+		  pstr->stop = pstr->raw_stop - idx + offset;
+		  pstr->offsets_needed = 0;
+		  while (mid > 0 && pstr->offsets[mid - 1] == offset)
+		    --mid;
+		  while (mid < pstr->valid_len)
+		    if (pstr->wcs[mid] != WEOF)
+		      break;
+		    else
+		      ++mid;
+		  if (mid == pstr->valid_len)
+		    pstr->valid_len = 0;
+		  else
+		    {
+		      pstr->valid_len = pstr->offsets[mid] - offset;
+		      if (pstr->valid_len)
+			{
+			  for (low = 0; low < pstr->valid_len; ++low)
+			    pstr->wcs[low] = WEOF;
+			  memset (pstr->mbs, 255, pstr->valid_len);
+			}
+		    }
+		  pstr->valid_raw_len = pstr->valid_len;
+		}
+	    }
+	  else
+#endif
+	    {
+	      pstr->tip_context = re_string_context_at (pstr, offset - 1,
+							eflags);
+#ifdef RE_ENABLE_I18N
+	      if (pstr->mb_cur_max > 1)
+		memmove (pstr->wcs, pstr->wcs + offset,
+			 (pstr->valid_len - offset) * sizeof (wint_t));
+#endif /* RE_ENABLE_I18N */
+	      if (BE (pstr->mbs_allocated, 0))
+		memmove (pstr->mbs, pstr->mbs + offset,
+			 pstr->valid_len - offset);
+	      pstr->valid_len -= offset;
+	      pstr->valid_raw_len -= offset;
+#if DEBUG
+	      assert (pstr->valid_len > 0);
+#endif
+	    }
+	}
+      else
+	{
+#ifdef RE_ENABLE_I18N
+	  /* No, skip all characters until IDX.  */
+	  int prev_valid_len = pstr->valid_len;
+
+	  if (BE (pstr->offsets_needed, 0))
+	    {
+	      pstr->len = pstr->raw_len - idx + offset;
+	      pstr->stop = pstr->raw_stop - idx + offset;
+	      pstr->offsets_needed = 0;
+	    }
+#endif
+	  pstr->valid_len = 0;
+#ifdef RE_ENABLE_I18N
+	  if (pstr->mb_cur_max > 1)
+	    {
+	      int wcs_idx;
+	      wint_t wc = WEOF;
+
+	      if (pstr->is_utf8)
+		{
+		  const unsigned char *raw, *p, *end;
+
+		  /* Special case UTF-8.  Multi-byte chars start with any
+		     byte other than 0x80 - 0xbf.  */
+		  raw = pstr->raw_mbs + pstr->raw_mbs_idx;
+		  end = raw + (offset - pstr->mb_cur_max);
+		  if (end < pstr->raw_mbs)
+		    end = pstr->raw_mbs;
+		  p = raw + offset - 1;
+#ifdef _LIBC
+		  /* We know the wchar_t encoding is UCS4, so for the simple
+		     case, ASCII characters, skip the conversion step.  */
+		  if (isascii (*p) && BE (pstr->trans == NULL, 1))
+		    {
+		      memset (&pstr->cur_state, '\0', sizeof (mbstate_t));
+		      /* pstr->valid_len = 0; */
+		      wc = (wchar_t) *p;
+		    }
+		  else
+#endif
+		    for (; p >= end; --p)
+		      if ((*p & 0xc0) != 0x80)
+			{
+			  mbstate_t cur_state;
+			  wchar_t wc2;
+			  int mlen = raw + pstr->len - p;
+			  unsigned char buf[6];
+			  size_t mbclen;
+
+			  if (BE (pstr->trans != NULL, 0))
+			    {
+			      int i = mlen < 6 ? mlen : 6;
+			      while (--i >= 0)
+				buf[i] = pstr->trans[p[i]];
+			    }
+			  /* XXX Don't use mbrtowc, we know which conversion
+			     to use (UTF-8 -> UCS4).  */
+			  memset (&cur_state, 0, sizeof (cur_state));
+			  mbclen = __mbrtowc (&wc2, (const char *) p, mlen,
+					      &cur_state);
+			  if (raw + offset - p <= mbclen
+			      && mbclen < (size_t) -2)
+			    {
+			      memset (&pstr->cur_state, '\0',
+				      sizeof (mbstate_t));
+			      pstr->valid_len = mbclen - (raw + offset - p);
+			      wc = wc2;
+			    }
+			  break;
+			}
+		}
+
+	      if (wc == WEOF)
+		pstr->valid_len = re_string_skip_chars (pstr, idx, &wc) - idx;
+	      if (wc == WEOF)
+		pstr->tip_context
+		  = re_string_context_at (pstr, prev_valid_len - 1, eflags);
+	      else
+		pstr->tip_context = ((BE (pstr->word_ops_used != 0, 0)
+				      && IS_WIDE_WORD_CHAR (wc))
+				     ? CONTEXT_WORD
+				     : ((IS_WIDE_NEWLINE (wc)
+					 && pstr->newline_anchor)
+					? CONTEXT_NEWLINE : 0));
+	      if (BE (pstr->valid_len, 0))
+		{
+		  for (wcs_idx = 0; wcs_idx < pstr->valid_len; ++wcs_idx)
+		    pstr->wcs[wcs_idx] = WEOF;
+		  if (pstr->mbs_allocated)
+		    memset (pstr->mbs, 255, pstr->valid_len);
+		}
+	      pstr->valid_raw_len = pstr->valid_len;
+	    }
+	  else
+#endif /* RE_ENABLE_I18N */
+	    {
+	      int c = pstr->raw_mbs[pstr->raw_mbs_idx + offset - 1];
+	      pstr->valid_raw_len = 0;
+	      if (pstr->trans)
+		c = pstr->trans[c];
+	      pstr->tip_context = (bitset_contain (pstr->word_char, c)
+				   ? CONTEXT_WORD
+				   : ((IS_NEWLINE (c) && pstr->newline_anchor)
+				      ? CONTEXT_NEWLINE : 0));
+	    }
+	}
+      if (!BE (pstr->mbs_allocated, 0))
+	pstr->mbs += offset;
+    }
+  pstr->raw_mbs_idx = idx;
+  pstr->len -= offset;
+  pstr->stop -= offset;
+
+  /* Then build the buffers.  */
+#ifdef RE_ENABLE_I18N
+  if (pstr->mb_cur_max > 1)
+    {
+      if (pstr->icase)
+	{
+	  reg_errcode_t ret = build_wcs_upper_buffer (pstr);
+	  if (BE (ret != REG_NOERROR, 0))
+	    return ret;
+	}
+      else
+	build_wcs_buffer (pstr);
+    }
+  else
+#endif /* RE_ENABLE_I18N */
+    if (BE (pstr->mbs_allocated, 0))
+      {
+	if (pstr->icase)
+	  build_upper_buffer (pstr);
+	else if (pstr->trans != NULL)
+	  re_string_translate_buffer (pstr);
+      }
+    else
+      pstr->valid_len = pstr->len;
+
+  pstr->cur_idx = 0;
+  return REG_NOERROR;
+}
+
+static unsigned char
+internal_function __attribute ((pure))
+re_string_peek_byte_case (const re_string_t *pstr, int idx)
+{
+  int ch, off;
+
+  /* Handle the common (easiest) cases first.  */
+  if (BE (!pstr->mbs_allocated, 1))
+    return re_string_peek_byte (pstr, idx);
+
+#ifdef RE_ENABLE_I18N
+  if (pstr->mb_cur_max > 1
+      && ! re_string_is_single_byte_char (pstr, pstr->cur_idx + idx))
+    return re_string_peek_byte (pstr, idx);
+#endif
+
+  off = pstr->cur_idx + idx;
+#ifdef RE_ENABLE_I18N
+  if (pstr->offsets_needed)
+    off = pstr->offsets[off];
+#endif
+
+  ch = pstr->raw_mbs[pstr->raw_mbs_idx + off];
+
+#ifdef RE_ENABLE_I18N
+  /* Ensure that e.g. for tr_TR.UTF-8 BACKSLASH DOTLESS SMALL LETTER I
+     this function returns CAPITAL LETTER I instead of first byte of
+     DOTLESS SMALL LETTER I.  The latter would confuse the parser,
+     since peek_byte_case doesn't advance cur_idx in any way.  */
+  if (pstr->offsets_needed && !isascii (ch))
+    return re_string_peek_byte (pstr, idx);
+#endif
+
+  return ch;
+}
+
+static unsigned char
+internal_function __attribute ((pure))
+re_string_fetch_byte_case (re_string_t *pstr)
+{
+  if (BE (!pstr->mbs_allocated, 1))
+    return re_string_fetch_byte (pstr);
+
+#ifdef RE_ENABLE_I18N
+  if (pstr->offsets_needed)
+    {
+      int off, ch;
+
+      /* For tr_TR.UTF-8 [[:islower:]] there is
+	 [[: CAPITAL LETTER I WITH DOT lower:]] in mbs.  Skip
+	 in that case the whole multi-byte character and return
+	 the original letter.  On the other side, with
+	 [[: DOTLESS SMALL LETTER I return [[:I, as doing
+	 anything else would complicate things too much.  */
+
+      if (!re_string_first_byte (pstr, pstr->cur_idx))
+	return re_string_fetch_byte (pstr);
+
+      off = pstr->offsets[pstr->cur_idx];
+      ch = pstr->raw_mbs[pstr->raw_mbs_idx + off];
+
+      if (! isascii (ch))
+	return re_string_fetch_byte (pstr);
+
+      re_string_skip_bytes (pstr,
+			    re_string_char_size_at (pstr, pstr->cur_idx));
+      return ch;
+    }
+#endif
+
+  return pstr->raw_mbs[pstr->raw_mbs_idx + pstr->cur_idx++];
+}
+
+static void
+internal_function
+re_string_destruct (re_string_t *pstr)
+{
+#ifdef RE_ENABLE_I18N
+  re_free (pstr->wcs);
+  re_free (pstr->offsets);
+#endif /* RE_ENABLE_I18N  */
+  if (pstr->mbs_allocated)
+    re_free (pstr->mbs);
+}
+
+/* Return the context at IDX in INPUT.  */
+
+static unsigned int
+internal_function
+re_string_context_at (const re_string_t *input, int idx, int eflags)
+{
+  int c;
+  if (BE (idx < 0, 0))
+    /* In this case, we use the value stored in input->tip_context,
+       since we can't know the character in input->mbs[-1] here.  */
+    return input->tip_context;
+  if (BE (idx == input->len, 0))
+    return ((eflags & REG_NOTEOL) ? CONTEXT_ENDBUF
+	    : CONTEXT_NEWLINE | CONTEXT_ENDBUF);
+#ifdef RE_ENABLE_I18N
+  if (input->mb_cur_max > 1)
+    {
+      wint_t wc;
+      int wc_idx = idx;
+      while(input->wcs[wc_idx] == WEOF)
+	{
+#ifdef DEBUG
+	  /* It must not happen.  */
+	  assert (wc_idx >= 0);
+#endif
+	  --wc_idx;
+	  if (wc_idx < 0)
+	    return input->tip_context;
+	}
+      wc = input->wcs[wc_idx];
+      if (BE (input->word_ops_used != 0, 0) && IS_WIDE_WORD_CHAR (wc))
+	return CONTEXT_WORD;
+      return (IS_WIDE_NEWLINE (wc) && input->newline_anchor
+	      ? CONTEXT_NEWLINE : 0);
+    }
+  else
+#endif
+    {
+      c = re_string_byte_at (input, idx);
+      if (bitset_contain (input->word_char, c))
+	return CONTEXT_WORD;
+      return IS_NEWLINE (c) && input->newline_anchor ? CONTEXT_NEWLINE : 0;
+    }
+}
+
+/* Functions for set operation.  */
+
+static reg_errcode_t
+internal_function
+re_node_set_alloc (re_node_set *set, int size)
+{
+  /*
+   * ADR: valgrind says size can be 0, which then doesn't
+   * free the block of size 0.  Harumph. This seems
+   * to work ok, though.
+   */
+  if (size == 0)
+    {
+       memset(set, 0, sizeof(*set));
+       return REG_NOERROR;
+    }
+  set->alloc = size;
+  set->nelem = 0;
+  set->elems = re_malloc (int, size);
+  if (BE (set->elems == NULL, 0))
+    return REG_ESPACE;
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+re_node_set_init_1 (re_node_set *set, int elem)
+{
+  set->alloc = 1;
+  set->nelem = 1;
+  set->elems = re_malloc (int, 1);
+  if (BE (set->elems == NULL, 0))
+    {
+      set->alloc = set->nelem = 0;
+      return REG_ESPACE;
+    }
+  set->elems[0] = elem;
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+re_node_set_init_2 (re_node_set *set, int elem1, int elem2)
+{
+  set->alloc = 2;
+  set->elems = re_malloc (int, 2);
+  if (BE (set->elems == NULL, 0))
+    return REG_ESPACE;
+  if (elem1 == elem2)
+    {
+      set->nelem = 1;
+      set->elems[0] = elem1;
+    }
+  else
+    {
+      set->nelem = 2;
+      if (elem1 < elem2)
+	{
+	  set->elems[0] = elem1;
+	  set->elems[1] = elem2;
+	}
+      else
+	{
+	  set->elems[0] = elem2;
+	  set->elems[1] = elem1;
+	}
+    }
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+re_node_set_init_copy (re_node_set *dest, const re_node_set *src)
+{
+  dest->nelem = src->nelem;
+  if (src->nelem > 0)
+    {
+      dest->alloc = dest->nelem;
+      dest->elems = re_malloc (int, dest->alloc);
+      if (BE (dest->elems == NULL, 0))
+	{
+	  dest->alloc = dest->nelem = 0;
+	  return REG_ESPACE;
+	}
+      memcpy (dest->elems, src->elems, src->nelem * sizeof (int));
+    }
+  else
+    re_node_set_init_empty (dest);
+  return REG_NOERROR;
+}
+
+/* Calculate the intersection of the sets SRC1 and SRC2. And merge it to
+   DEST. Return value indicate the error code or REG_NOERROR if succeeded.
+   Note: We assume dest->elems is NULL, when dest->alloc is 0.  */
+
+static reg_errcode_t
+internal_function
+re_node_set_add_intersect (re_node_set *dest, const re_node_set *src1,
+			   const re_node_set *src2)
+{
+  int i1, i2, is, id, delta, sbase;
+  if (src1->nelem == 0 || src2->nelem == 0)
+    return REG_NOERROR;
+
+  /* We need dest->nelem + 2 * elems_in_intersection; this is a
+     conservative estimate.  */
+  if (src1->nelem + src2->nelem + dest->nelem > dest->alloc)
+    {
+      int new_alloc = src1->nelem + src2->nelem + dest->alloc;
+      int *new_elems = re_realloc (dest->elems, int, new_alloc);
+      if (BE (new_elems == NULL, 0))
+	return REG_ESPACE;
+      dest->elems = new_elems;
+      dest->alloc = new_alloc;
+    }
+
+  /* Find the items in the intersection of SRC1 and SRC2, and copy
+     into the top of DEST those that are not already in DEST itself.  */
+  sbase = dest->nelem + src1->nelem + src2->nelem;
+  i1 = src1->nelem - 1;
+  i2 = src2->nelem - 1;
+  id = dest->nelem - 1;
+  for (;;)
+    {
+      if (src1->elems[i1] == src2->elems[i2])
+	{
+	  /* Try to find the item in DEST.  Maybe we could binary search?  */
+	  while (id >= 0 && dest->elems[id] > src1->elems[i1])
+	    --id;
+
+	  if (id < 0 || dest->elems[id] != src1->elems[i1])
+	    dest->elems[--sbase] = src1->elems[i1];
+
+	  if (--i1 < 0 || --i2 < 0)
+	    break;
+	}
+
+      /* Lower the highest of the two items.  */
+      else if (src1->elems[i1] < src2->elems[i2])
+	{
+	  if (--i2 < 0)
+	    break;
+	}
+      else
+	{
+	  if (--i1 < 0)
+	    break;
+	}
+    }
+
+  id = dest->nelem - 1;
+  is = dest->nelem + src1->nelem + src2->nelem - 1;
+  delta = is - sbase + 1;
+
+  /* Now copy.  When DELTA becomes zero, the remaining
+     DEST elements are already in place; this is more or
+     less the same loop that is in re_node_set_merge.  */
+  dest->nelem += delta;
+  if (delta > 0 && id >= 0)
+    for (;;)
+      {
+	if (dest->elems[is] > dest->elems[id])
+	  {
+	    /* Copy from the top.  */
+	    dest->elems[id + delta--] = dest->elems[is--];
+	    if (delta == 0)
+	      break;
+	  }
+	else
+	  {
+	    /* Slide from the bottom.  */
+	    dest->elems[id + delta] = dest->elems[id];
+	    if (--id < 0)
+	      break;
+	  }
+      }
+
+  /* Copy remaining SRC elements.  */
+  memcpy (dest->elems, dest->elems + sbase, delta * sizeof (int));
+
+  return REG_NOERROR;
+}
+
+/* Calculate the union set of the sets SRC1 and SRC2. And store it to
+   DEST. Return value indicate the error code or REG_NOERROR if succeeded.  */
+
+static reg_errcode_t
+internal_function
+re_node_set_init_union (re_node_set *dest, const re_node_set *src1,
+			const re_node_set *src2)
+{
+  int i1, i2, id;
+  if (src1 != NULL && src1->nelem > 0 && src2 != NULL && src2->nelem > 0)
+    {
+      dest->alloc = src1->nelem + src2->nelem;
+      dest->elems = re_malloc (int, dest->alloc);
+      if (BE (dest->elems == NULL, 0))
+	return REG_ESPACE;
+    }
+  else
+    {
+      if (src1 != NULL && src1->nelem > 0)
+	return re_node_set_init_copy (dest, src1);
+      else if (src2 != NULL && src2->nelem > 0)
+	return re_node_set_init_copy (dest, src2);
+      else
+	re_node_set_init_empty (dest);
+      return REG_NOERROR;
+    }
+  for (i1 = i2 = id = 0 ; i1 < src1->nelem && i2 < src2->nelem ;)
+    {
+      if (src1->elems[i1] > src2->elems[i2])
+	{
+	  dest->elems[id++] = src2->elems[i2++];
+	  continue;
+	}
+      if (src1->elems[i1] == src2->elems[i2])
+	++i2;
+      dest->elems[id++] = src1->elems[i1++];
+    }
+  if (i1 < src1->nelem)
+    {
+      memcpy (dest->elems + id, src1->elems + i1,
+	     (src1->nelem - i1) * sizeof (int));
+      id += src1->nelem - i1;
+    }
+  else if (i2 < src2->nelem)
+    {
+      memcpy (dest->elems + id, src2->elems + i2,
+	     (src2->nelem - i2) * sizeof (int));
+      id += src2->nelem - i2;
+    }
+  dest->nelem = id;
+  return REG_NOERROR;
+}
+
+/* Calculate the union set of the sets DEST and SRC. And store it to
+   DEST. Return value indicate the error code or REG_NOERROR if succeeded.  */
+
+static reg_errcode_t
+internal_function
+re_node_set_merge (re_node_set *dest, const re_node_set *src)
+{
+  int is, id, sbase, delta;
+  if (src == NULL || src->nelem == 0)
+    return REG_NOERROR;
+  if (dest->alloc < 2 * src->nelem + dest->nelem)
+    {
+      int new_alloc = 2 * (src->nelem + dest->alloc);
+      int *new_buffer = re_realloc (dest->elems, int, new_alloc);
+      if (BE (new_buffer == NULL, 0))
+	return REG_ESPACE;
+      dest->elems = new_buffer;
+      dest->alloc = new_alloc;
+    }
+
+  if (BE (dest->nelem == 0, 0))
+    {
+      dest->nelem = src->nelem;
+      memcpy (dest->elems, src->elems, src->nelem * sizeof (int));
+      return REG_NOERROR;
+    }
+
+  /* Copy into the top of DEST the items of SRC that are not
+     found in DEST.  Maybe we could binary search in DEST?  */
+  for (sbase = dest->nelem + 2 * src->nelem,
+       is = src->nelem - 1, id = dest->nelem - 1; is >= 0 && id >= 0; )
+    {
+      if (dest->elems[id] == src->elems[is])
+	is--, id--;
+      else if (dest->elems[id] < src->elems[is])
+	dest->elems[--sbase] = src->elems[is--];
+      else /* if (dest->elems[id] > src->elems[is]) */
+	--id;
+    }
+
+  if (is >= 0)
+    {
+      /* If DEST is exhausted, the remaining items of SRC must be unique.  */
+      sbase -= is + 1;
+      memcpy (dest->elems + sbase, src->elems, (is + 1) * sizeof (int));
+    }
+
+  id = dest->nelem - 1;
+  is = dest->nelem + 2 * src->nelem - 1;
+  delta = is - sbase + 1;
+  if (delta == 0)
+    return REG_NOERROR;
+
+  /* Now copy.  When DELTA becomes zero, the remaining
+     DEST elements are already in place.  */
+  dest->nelem += delta;
+  for (;;)
+    {
+      if (dest->elems[is] > dest->elems[id])
+	{
+	  /* Copy from the top.  */
+	  dest->elems[id + delta--] = dest->elems[is--];
+	  if (delta == 0)
+	    break;
+	}
+      else
+	{
+	  /* Slide from the bottom.  */
+	  dest->elems[id + delta] = dest->elems[id];
+	  if (--id < 0)
+	    {
+	      /* Copy remaining SRC elements.  */
+	      memcpy (dest->elems, dest->elems + sbase,
+		      delta * sizeof (int));
+	      break;
+	    }
+	}
+    }
+
+  return REG_NOERROR;
+}
+
+/* Insert the new element ELEM to the re_node_set* SET.
+   SET should not already have ELEM.
+   return -1 if an error has occurred, return 1 otherwise.  */
+
+static int
+internal_function
+re_node_set_insert (re_node_set *set, int elem)
+{
+  int idx;
+  /* In case the set is empty.  */
+  if (set->alloc == 0)
+    {
+      if (BE (re_node_set_init_1 (set, elem) == REG_NOERROR, 1))
+	return 1;
+      else
+	return -1;
+    }
+
+  if (BE (set->nelem, 0) == 0)
+    {
+      /* We already guaranteed above that set->alloc != 0.  */
+      set->elems[0] = elem;
+      ++set->nelem;
+      return 1;
+    }
+
+  /* Realloc if we need.  */
+  if (set->alloc == set->nelem)
+    {
+      int *new_elems;
+      set->alloc = set->alloc * 2;
+      new_elems = re_realloc (set->elems, int, set->alloc);
+      if (BE (new_elems == NULL, 0))
+	return -1;
+      set->elems = new_elems;
+    }
+
+  /* Move the elements which follows the new element.  Test the
+     first element separately to skip a check in the inner loop.  */
+  if (elem < set->elems[0])
+    {
+      idx = 0;
+      for (idx = set->nelem; idx > 0; idx--)
+	set->elems[idx] = set->elems[idx - 1];
+    }
+  else
+    {
+      for (idx = set->nelem; set->elems[idx - 1] > elem; idx--)
+	set->elems[idx] = set->elems[idx - 1];
+    }
+
+  /* Insert the new element.  */
+  set->elems[idx] = elem;
+  ++set->nelem;
+  return 1;
+}
+
+/* Insert the new element ELEM to the re_node_set* SET.
+   SET should not already have any element greater than or equal to ELEM.
+   Return -1 if an error has occurred, return 1 otherwise.  */
+
+static int
+internal_function
+re_node_set_insert_last (re_node_set *set, int elem)
+{
+  /* Realloc if we need.  */
+  if (set->alloc == set->nelem)
+    {
+      int *new_elems;
+      set->alloc = (set->alloc + 1) * 2;
+      new_elems = re_realloc (set->elems, int, set->alloc);
+      if (BE (new_elems == NULL, 0))
+	return -1;
+      set->elems = new_elems;
+    }
+
+  /* Insert the new element.  */
+  set->elems[set->nelem++] = elem;
+  return 1;
+}
+
+/* Compare two node sets SET1 and SET2.
+   return 1 if SET1 and SET2 are equivalent, return 0 otherwise.  */
+
+static int
+internal_function __attribute ((pure))
+re_node_set_compare (const re_node_set *set1, const re_node_set *set2)
+{
+  int i;
+  if (set1 == NULL || set2 == NULL || set1->nelem != set2->nelem)
+    return 0;
+  for (i = set1->nelem ; --i >= 0 ; )
+    if (set1->elems[i] != set2->elems[i])
+      return 0;
+  return 1;
+}
+
+/* Return (idx + 1) if SET contains the element ELEM, return 0 otherwise.  */
+
+static int
+internal_function __attribute ((pure))
+re_node_set_contains (const re_node_set *set, int elem)
+{
+  unsigned int idx, right, mid;
+  if (set->nelem <= 0)
+    return 0;
+
+  /* Binary search the element.  */
+  idx = 0;
+  right = set->nelem - 1;
+  while (idx < right)
+    {
+      mid = idx + (right - idx) / 2;
+      if (set->elems[mid] < elem)
+	idx = mid + 1;
+      else
+	right = mid;
+    }
+  return set->elems[idx] == elem ? idx + 1 : 0;
+}
+
+static void
+internal_function
+re_node_set_remove_at (re_node_set *set, int idx)
+{
+  if (idx < 0 || idx >= set->nelem)
+    return;
+  --set->nelem;
+  for (; idx < set->nelem; idx++)
+    set->elems[idx] = set->elems[idx + 1];
+}
+
+
+/* Add the token TOKEN to dfa->nodes, and return the index of the token.
+   Or return -1, if an error has occurred.  */
+
+static int
+internal_function
+re_dfa_add_node (re_dfa_t *dfa, re_token_t token)
+{
+  if (BE (dfa->nodes_len >= dfa->nodes_alloc, 0))
+    {
+      size_t new_nodes_alloc = dfa->nodes_alloc * 2;
+      int *new_nexts, *new_indices;
+      re_node_set *new_edests, *new_eclosures;
+      re_token_t *new_nodes;
+
+      /* Avoid overflows in realloc.  */
+      const size_t max_object_size = MAX (sizeof (re_token_t),
+					  MAX (sizeof (re_node_set),
+					       sizeof (int)));
+      if (BE (SIZE_MAX / max_object_size < new_nodes_alloc, 0))
+	return -1;
+
+      new_nodes = re_realloc (dfa->nodes, re_token_t, new_nodes_alloc);
+      if (BE (new_nodes == NULL, 0))
+	return -1;
+      dfa->nodes = new_nodes;
+      new_nexts = re_realloc (dfa->nexts, int, new_nodes_alloc);
+      new_indices = re_realloc (dfa->org_indices, int, new_nodes_alloc);
+      new_edests = re_realloc (dfa->edests, re_node_set, new_nodes_alloc);
+      new_eclosures = re_realloc (dfa->eclosures, re_node_set, new_nodes_alloc);
+      if (BE (new_nexts == NULL || new_indices == NULL
+	      || new_edests == NULL || new_eclosures == NULL, 0))
+	return -1;
+      dfa->nexts = new_nexts;
+      dfa->org_indices = new_indices;
+      dfa->edests = new_edests;
+      dfa->eclosures = new_eclosures;
+      dfa->nodes_alloc = new_nodes_alloc;
+    }
+  dfa->nodes[dfa->nodes_len] = token;
+  dfa->nodes[dfa->nodes_len].constraint = 0;
+#ifdef RE_ENABLE_I18N
+  dfa->nodes[dfa->nodes_len].accept_mb =
+    (token.type == OP_PERIOD && dfa->mb_cur_max > 1) || token.type == COMPLEX_BRACKET;
+#endif
+  dfa->nexts[dfa->nodes_len] = -1;
+  re_node_set_init_empty (dfa->edests + dfa->nodes_len);
+  re_node_set_init_empty (dfa->eclosures + dfa->nodes_len);
+  return dfa->nodes_len++;
+}
+
+static inline unsigned int
+internal_function
+calc_state_hash (const re_node_set *nodes, unsigned int context)
+{
+  unsigned int hash = nodes->nelem + context;
+  int i;
+  for (i = 0 ; i < nodes->nelem ; i++)
+    hash += nodes->elems[i];
+  return hash;
+}
+
+/* Search for the state whose node_set is equivalent to NODES.
+   Return the pointer to the state, if we found it in the DFA.
+   Otherwise create the new one and return it.  In case of an error
+   return NULL and set the error code in ERR.
+   Note: - We assume NULL as the invalid state, then it is possible that
+	   return value is NULL and ERR is REG_NOERROR.
+	 - We never return non-NULL value in case of any errors, it is for
+	   optimization.  */
+
+static re_dfastate_t *
+internal_function
+re_acquire_state (reg_errcode_t *err, const re_dfa_t *dfa,
+		  const re_node_set *nodes)
+{
+  unsigned int hash;
+  re_dfastate_t *new_state;
+  struct re_state_table_entry *spot;
+  int i;
+  if (BE (nodes->nelem == 0, 0))
+    {
+      *err = REG_NOERROR;
+      return NULL;
+    }
+  hash = calc_state_hash (nodes, 0);
+  spot = dfa->state_table + (hash & dfa->state_hash_mask);
+
+  for (i = 0 ; i < spot->num ; i++)
+    {
+      re_dfastate_t *state = spot->array[i];
+      if (hash != state->hash)
+	continue;
+      if (re_node_set_compare (&state->nodes, nodes))
+	return state;
+    }
+
+  /* There are no appropriate state in the dfa, create the new one.  */
+  new_state = create_ci_newstate (dfa, nodes, hash);
+  if (BE (new_state == NULL, 0))
+    *err = REG_ESPACE;
+
+  return new_state;
+}
+
+/* Search for the state whose node_set is equivalent to NODES and
+   whose context is equivalent to CONTEXT.
+   Return the pointer to the state, if we found it in the DFA.
+   Otherwise create the new one and return it.  In case of an error
+   return NULL and set the error code in ERR.
+   Note: - We assume NULL as the invalid state, then it is possible that
+	   return value is NULL and ERR is REG_NOERROR.
+	 - We never return non-NULL value in case of any errors, it is for
+	   optimization.  */
+
+static re_dfastate_t *
+internal_function
+re_acquire_state_context (reg_errcode_t *err, const re_dfa_t *dfa,
+			  const re_node_set *nodes, unsigned int context)
+{
+  unsigned int hash;
+  re_dfastate_t *new_state;
+  struct re_state_table_entry *spot;
+  int i;
+  if (nodes->nelem == 0)
+    {
+      *err = REG_NOERROR;
+      return NULL;
+    }
+  hash = calc_state_hash (nodes, context);
+  spot = dfa->state_table + (hash & dfa->state_hash_mask);
+
+  for (i = 0 ; i < spot->num ; i++)
+    {
+      re_dfastate_t *state = spot->array[i];
+      if (state->hash == hash
+	  && state->context == context
+	  && re_node_set_compare (state->entrance_nodes, nodes))
+	return state;
+    }
+  /* There are no appropriate state in `dfa', create the new one.  */
+  new_state = create_cd_newstate (dfa, nodes, context, hash);
+  if (BE (new_state == NULL, 0))
+    *err = REG_ESPACE;
+
+  return new_state;
+}
+
+/* Finish initialization of the new state NEWSTATE, and using its hash value
+   HASH put in the appropriate bucket of DFA's state table.  Return value
+   indicates the error code if failed.  */
+
+static reg_errcode_t
+register_state (const re_dfa_t *dfa, re_dfastate_t *newstate,
+		unsigned int hash)
+{
+  struct re_state_table_entry *spot;
+  reg_errcode_t err;
+  int i;
+
+  newstate->hash = hash;
+  err = re_node_set_alloc (&newstate->non_eps_nodes, newstate->nodes.nelem);
+  if (BE (err != REG_NOERROR, 0))
+    return REG_ESPACE;
+  for (i = 0; i < newstate->nodes.nelem; i++)
+    {
+      int elem = newstate->nodes.elems[i];
+      if (!IS_EPSILON_NODE (dfa->nodes[elem].type))
+	if (re_node_set_insert_last (&newstate->non_eps_nodes, elem) < 0)
+	  return REG_ESPACE;
+    }
+
+  spot = dfa->state_table + (hash & dfa->state_hash_mask);
+  if (BE (spot->alloc <= spot->num, 0))
+    {
+      int new_alloc = 2 * spot->num + 2;
+      re_dfastate_t **new_array = re_realloc (spot->array, re_dfastate_t *,
+					      new_alloc);
+      if (BE (new_array == NULL, 0))
+	return REG_ESPACE;
+      spot->array = new_array;
+      spot->alloc = new_alloc;
+    }
+  spot->array[spot->num++] = newstate;
+  return REG_NOERROR;
+}
+
+static void
+free_state (re_dfastate_t *state)
+{
+  re_node_set_free (&state->non_eps_nodes);
+  re_node_set_free (&state->inveclosure);
+  if (state->entrance_nodes != &state->nodes)
+    {
+      re_node_set_free (state->entrance_nodes);
+      re_free (state->entrance_nodes);
+    }
+  re_node_set_free (&state->nodes);
+  re_free (state->word_trtable);
+  re_free (state->trtable);
+  re_free (state);
+}
+
+/* Create the new state which is independ of contexts.
+   Return the new state if succeeded, otherwise return NULL.  */
+
+static re_dfastate_t *
+internal_function
+create_ci_newstate (const re_dfa_t *dfa, const re_node_set *nodes,
+		    unsigned int hash)
+{
+  int i;
+  reg_errcode_t err;
+  re_dfastate_t *newstate;
+
+  newstate = (re_dfastate_t *) calloc (sizeof (re_dfastate_t), 1);
+  if (BE (newstate == NULL, 0))
+    return NULL;
+  err = re_node_set_init_copy (&newstate->nodes, nodes);
+  if (BE (err != REG_NOERROR, 0))
+    {
+      re_free (newstate);
+      return NULL;
+    }
+
+  newstate->entrance_nodes = &newstate->nodes;
+  for (i = 0 ; i < nodes->nelem ; i++)
+    {
+      re_token_t *node = dfa->nodes + nodes->elems[i];
+      re_token_type_t type = node->type;
+      if (type == CHARACTER && !node->constraint)
+	continue;
+#ifdef RE_ENABLE_I18N
+      newstate->accept_mb |= node->accept_mb;
+#endif /* RE_ENABLE_I18N */
+
+      /* If the state has the halt node, the state is a halt state.  */
+      if (type == END_OF_RE)
+	newstate->halt = 1;
+      else if (type == OP_BACK_REF)
+	newstate->has_backref = 1;
+      else if (type == ANCHOR || node->constraint)
+	newstate->has_constraint = 1;
+    }
+  err = register_state (dfa, newstate, hash);
+  if (BE (err != REG_NOERROR, 0))
+    {
+      free_state (newstate);
+      newstate = NULL;
+    }
+  return newstate;
+}
+
+/* Create the new state which is depend on the context CONTEXT.
+   Return the new state if succeeded, otherwise return NULL.  */
+
+static re_dfastate_t *
+internal_function
+create_cd_newstate (const re_dfa_t *dfa, const re_node_set *nodes,
+		    unsigned int context, unsigned int hash)
+{
+  int i, nctx_nodes = 0;
+  reg_errcode_t err;
+  re_dfastate_t *newstate;
+
+  newstate = (re_dfastate_t *) calloc (sizeof (re_dfastate_t), 1);
+  if (BE (newstate == NULL, 0))
+    return NULL;
+  err = re_node_set_init_copy (&newstate->nodes, nodes);
+  if (BE (err != REG_NOERROR, 0))
+    {
+      re_free (newstate);
+      return NULL;
+    }
+
+  newstate->context = context;
+  newstate->entrance_nodes = &newstate->nodes;
+
+  for (i = 0 ; i < nodes->nelem ; i++)
+    {
+      re_token_t *node = dfa->nodes + nodes->elems[i];
+      re_token_type_t type = node->type;
+      unsigned int constraint = node->constraint;
+
+      if (type == CHARACTER && !constraint)
+	continue;
+#ifdef RE_ENABLE_I18N
+      newstate->accept_mb |= node->accept_mb;
+#endif /* RE_ENABLE_I18N */
+
+      /* If the state has the halt node, the state is a halt state.  */
+      if (type == END_OF_RE)
+	newstate->halt = 1;
+      else if (type == OP_BACK_REF)
+	newstate->has_backref = 1;
+
+      if (constraint)
+	{
+	  if (newstate->entrance_nodes == &newstate->nodes)
+	    {
+	      newstate->entrance_nodes = re_malloc (re_node_set, 1);
+	      if (BE (newstate->entrance_nodes == NULL, 0))
+		{
+		  free_state (newstate);
+		  return NULL;
+		}
+	      if (re_node_set_init_copy (newstate->entrance_nodes, nodes)
+		  != REG_NOERROR)
+		return NULL;
+	      nctx_nodes = 0;
+	      newstate->has_constraint = 1;
+	    }
+
+	  if (NOT_SATISFY_PREV_CONSTRAINT (constraint,context))
+	    {
+	      re_node_set_remove_at (&newstate->nodes, i - nctx_nodes);
+	      ++nctx_nodes;
+	    }
+	}
+    }
+  err = register_state (dfa, newstate, hash);
+  if (BE (err != REG_NOERROR, 0))
+    {
+      free_state (newstate);
+      newstate = NULL;
+    }
+  return  newstate;
+}
diff --git a/compat/regex/regex_internal.h b/compat/regex/regex_internal.h
new file mode 100644
index 000000000000..3ee8aae59d79
--- /dev/null
+++ b/compat/regex/regex_internal.h
@@ -0,0 +1,809 @@
+/* Extended regular expression matching and search library.
+   Copyright (C) 2002-2005, 2007, 2008, 2010 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _REGEX_INTERNAL_H
+#define _REGEX_INTERNAL_H 1
+
+#include <assert.h>
+#include <ctype.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#if defined HAVE_LANGINFO_H || defined HAVE_LANGINFO_CODESET || defined _LIBC
+# include <langinfo.h>
+#endif
+#if defined HAVE_LOCALE_H || defined _LIBC
+# include <locale.h>
+#endif
+#if defined HAVE_WCHAR_H || defined _LIBC
+# include <wchar.h>
+#endif /* HAVE_WCHAR_H || _LIBC */
+#if defined HAVE_WCTYPE_H || defined _LIBC
+# include <wctype.h>
+#endif /* HAVE_WCTYPE_H || _LIBC */
+#if defined HAVE_STDBOOL_H || defined _LIBC
+# include <stdbool.h>
+#endif /* HAVE_STDBOOL_H || _LIBC */
+#if !defined(ZOS_USS)
+#if defined HAVE_STDINT_H || defined _LIBC
+# include <stdint.h>
+#endif /* HAVE_STDINT_H || _LIBC */
+#endif /* !ZOS_USS */
+#if defined _LIBC
+# include <bits/libc-lock.h>
+#else
+# define __libc_lock_define(CLASS,NAME)
+# define __libc_lock_init(NAME) do { } while (0)
+# define __libc_lock_lock(NAME) do { } while (0)
+# define __libc_lock_unlock(NAME) do { } while (0)
+#endif
+
+#ifndef GAWK
+/* In case that the system doesn't have isblank().  */
+#if !defined _LIBC && !defined HAVE_ISBLANK && !defined isblank
+# define isblank(ch) ((ch) == ' ' || (ch) == '\t')
+#endif
+#else /* GAWK */
+/*
+ * This is a freaking mess. On glibc systems you have to define
+ * a magic constant to get isblank() out of <ctype.h>, since it's
+ * a C99 function.  To heck with all that and borrow a page from
+ * dfa.c's book.
+ */
+
+static int
+is_blank (int c)
+{
+   return (c == ' ' || c == '\t');
+}
+#endif /* GAWK */
+
+#ifdef _LIBC
+# ifndef _RE_DEFINE_LOCALE_FUNCTIONS
+#  define _RE_DEFINE_LOCALE_FUNCTIONS 1
+#   include <locale/localeinfo.h>
+#   include <locale/elem-hash.h>
+#   include <locale/coll-lookup.h>
+# endif
+#endif
+
+/* This is for other GNU distributions with internationalized messages.  */
+#if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
+# include <libintl.h>
+# ifdef _LIBC
+#  undef gettext
+#  define gettext(msgid) \
+  INTUSE(__dcgettext) (_libc_intl_domainname, msgid, LC_MESSAGES)
+# endif
+#else
+# define gettext(msgid) (msgid)
+#endif
+
+#ifndef gettext_noop
+/* This define is so xgettext can find the internationalizable
+   strings.  */
+# define gettext_noop(String) String
+#endif
+
+/* For loser systems without the definition.  */
+#ifndef SIZE_MAX
+# define SIZE_MAX ((size_t) -1)
+#endif
+
+#ifndef NO_MBSUPPORT
+#include "mbsupport.h" /* gawk */
+#endif
+#ifndef MB_CUR_MAX
+#define MB_CUR_MAX 1
+#endif
+
+#if (defined MBS_SUPPORT) || _LIBC
+# define RE_ENABLE_I18N
+#endif
+
+#if __GNUC__ >= 3
+# define BE(expr, val) __builtin_expect (expr, val)
+#else
+# define BE(expr, val) (expr)
+# ifdef inline
+# undef inline
+# endif
+# define inline
+#endif
+
+/* Number of single byte character.  */
+#define SBC_MAX 256
+
+#define COLL_ELEM_LEN_MAX 8
+
+/* The character which represents newline.  */
+#define NEWLINE_CHAR '\n'
+#define WIDE_NEWLINE_CHAR L'\n'
+
+/* Rename to standard API for using out of glibc.  */
+#ifndef _LIBC
+# ifdef __wctype
+# undef __wctype
+# endif
+# define __wctype wctype
+# ifdef __iswctype
+# undef __iswctype
+# endif
+# define __iswctype iswctype
+# define __btowc btowc
+# define __mbrtowc mbrtowc
+#undef __mempcpy	/* GAWK */
+# define __mempcpy mempcpy
+# define __wcrtomb wcrtomb
+# define __regfree regfree
+# define attribute_hidden
+#endif /* not _LIBC */
+
+#ifdef __GNUC__
+# define __attribute(arg) __attribute__ (arg)
+#else
+# define __attribute(arg)
+#endif
+
+extern const char __re_error_msgid[] attribute_hidden;
+extern const size_t __re_error_msgid_idx[] attribute_hidden;
+
+/* An integer used to represent a set of bits.  It must be unsigned,
+   and must be at least as wide as unsigned int.  */
+typedef unsigned long int bitset_word_t;
+/* All bits set in a bitset_word_t.  */
+#define BITSET_WORD_MAX ULONG_MAX
+/* Number of bits in a bitset_word_t.  */
+#define BITSET_WORD_BITS (sizeof (bitset_word_t) * CHAR_BIT)
+/* Number of bitset_word_t in a bit_set.  */
+#define BITSET_WORDS (SBC_MAX / BITSET_WORD_BITS)
+typedef bitset_word_t bitset_t[BITSET_WORDS];
+typedef bitset_word_t *re_bitset_ptr_t;
+typedef const bitset_word_t *re_const_bitset_ptr_t;
+
+#define bitset_set(set,i) \
+  (set[i / BITSET_WORD_BITS] |= (bitset_word_t) 1 << i % BITSET_WORD_BITS)
+#define bitset_clear(set,i) \
+  (set[i / BITSET_WORD_BITS] &= ~((bitset_word_t) 1 << i % BITSET_WORD_BITS))
+#define bitset_contain(set,i) \
+  (set[i / BITSET_WORD_BITS] & ((bitset_word_t) 1 << i % BITSET_WORD_BITS))
+#define bitset_empty(set) memset (set, '\0', sizeof (bitset_t))
+#define bitset_set_all(set) memset (set, '\xff', sizeof (bitset_t))
+#define bitset_copy(dest,src) memcpy (dest, src, sizeof (bitset_t))
+
+#define PREV_WORD_CONSTRAINT 0x0001
+#define PREV_NOTWORD_CONSTRAINT 0x0002
+#define NEXT_WORD_CONSTRAINT 0x0004
+#define NEXT_NOTWORD_CONSTRAINT 0x0008
+#define PREV_NEWLINE_CONSTRAINT 0x0010
+#define NEXT_NEWLINE_CONSTRAINT 0x0020
+#define PREV_BEGBUF_CONSTRAINT 0x0040
+#define NEXT_ENDBUF_CONSTRAINT 0x0080
+#define WORD_DELIM_CONSTRAINT 0x0100
+#define NOT_WORD_DELIM_CONSTRAINT 0x0200
+
+typedef enum
+{
+  INSIDE_WORD = PREV_WORD_CONSTRAINT | NEXT_WORD_CONSTRAINT,
+  WORD_FIRST = PREV_NOTWORD_CONSTRAINT | NEXT_WORD_CONSTRAINT,
+  WORD_LAST = PREV_WORD_CONSTRAINT | NEXT_NOTWORD_CONSTRAINT,
+  INSIDE_NOTWORD = PREV_NOTWORD_CONSTRAINT | NEXT_NOTWORD_CONSTRAINT,
+  LINE_FIRST = PREV_NEWLINE_CONSTRAINT,
+  LINE_LAST = NEXT_NEWLINE_CONSTRAINT,
+  BUF_FIRST = PREV_BEGBUF_CONSTRAINT,
+  BUF_LAST = NEXT_ENDBUF_CONSTRAINT,
+  WORD_DELIM = WORD_DELIM_CONSTRAINT,
+  NOT_WORD_DELIM = NOT_WORD_DELIM_CONSTRAINT
+} re_context_type;
+
+typedef struct
+{
+  int alloc;
+  int nelem;
+  int *elems;
+} re_node_set;
+
+typedef enum
+{
+  NON_TYPE = 0,
+
+  /* Node type, These are used by token, node, tree.  */
+  CHARACTER = 1,
+  END_OF_RE = 2,
+  SIMPLE_BRACKET = 3,
+  OP_BACK_REF = 4,
+  OP_PERIOD = 5,
+#ifdef RE_ENABLE_I18N
+  COMPLEX_BRACKET = 6,
+  OP_UTF8_PERIOD = 7,
+#endif /* RE_ENABLE_I18N */
+
+  /* We define EPSILON_BIT as a macro so that OP_OPEN_SUBEXP is used
+     when the debugger shows values of this enum type.  */
+#define EPSILON_BIT 8
+  OP_OPEN_SUBEXP = EPSILON_BIT | 0,
+  OP_CLOSE_SUBEXP = EPSILON_BIT | 1,
+  OP_ALT = EPSILON_BIT | 2,
+  OP_DUP_ASTERISK = EPSILON_BIT | 3,
+  ANCHOR = EPSILON_BIT | 4,
+
+  /* Tree type, these are used only by tree. */
+  CONCAT = 16,
+  SUBEXP = 17,
+
+  /* Token type, these are used only by token.  */
+  OP_DUP_PLUS = 18,
+  OP_DUP_QUESTION,
+  OP_OPEN_BRACKET,
+  OP_CLOSE_BRACKET,
+  OP_CHARSET_RANGE,
+  OP_OPEN_DUP_NUM,
+  OP_CLOSE_DUP_NUM,
+  OP_NON_MATCH_LIST,
+  OP_OPEN_COLL_ELEM,
+  OP_CLOSE_COLL_ELEM,
+  OP_OPEN_EQUIV_CLASS,
+  OP_CLOSE_EQUIV_CLASS,
+  OP_OPEN_CHAR_CLASS,
+  OP_CLOSE_CHAR_CLASS,
+  OP_WORD,
+  OP_NOTWORD,
+  OP_SPACE,
+  OP_NOTSPACE,
+  BACK_SLASH
+
+} re_token_type_t;
+
+#ifdef RE_ENABLE_I18N
+typedef struct
+{
+  /* Multibyte characters.  */
+  wchar_t *mbchars;
+
+  /* Collating symbols.  */
+# ifdef _LIBC
+  int32_t *coll_syms;
+# endif
+
+  /* Equivalence classes. */
+# ifdef _LIBC
+  int32_t *equiv_classes;
+# endif
+
+  /* Range expressions. */
+# ifdef _LIBC
+  uint32_t *range_starts;
+  uint32_t *range_ends;
+# else /* not _LIBC */
+  wchar_t *range_starts;
+  wchar_t *range_ends;
+# endif /* not _LIBC */
+
+  /* Character classes. */
+  wctype_t *char_classes;
+
+  /* If this character set is the non-matching list.  */
+  unsigned int non_match : 1;
+
+  /* # of multibyte characters.  */
+  int nmbchars;
+
+  /* # of collating symbols.  */
+  int ncoll_syms;
+
+  /* # of equivalence classes. */
+  int nequiv_classes;
+
+  /* # of range expressions. */
+  int nranges;
+
+  /* # of character classes. */
+  int nchar_classes;
+} re_charset_t;
+#endif /* RE_ENABLE_I18N */
+
+typedef struct
+{
+  union
+  {
+    unsigned char c;		/* for CHARACTER */
+    re_bitset_ptr_t sbcset;	/* for SIMPLE_BRACKET */
+#ifdef RE_ENABLE_I18N
+    re_charset_t *mbcset;	/* for COMPLEX_BRACKET */
+#endif /* RE_ENABLE_I18N */
+    int idx;			/* for BACK_REF */
+    re_context_type ctx_type;	/* for ANCHOR */
+  } opr;
+#if __GNUC__ >= 2
+  re_token_type_t type : 8;
+#else
+  re_token_type_t type;
+#endif
+  unsigned int constraint : 10;	/* context constraint */
+  unsigned int duplicated : 1;
+  unsigned int opt_subexp : 1;
+#ifdef RE_ENABLE_I18N
+  unsigned int accept_mb : 1;
+  /* These 2 bits can be moved into the union if needed (e.g. if running out
+     of bits; move opr.c to opr.c.c and move the flags to opr.c.flags).  */
+  unsigned int mb_partial : 1;
+#endif
+  unsigned int word_char : 1;
+} re_token_t;
+
+#define IS_EPSILON_NODE(type) ((type) & EPSILON_BIT)
+
+struct re_string_t
+{
+  /* Indicate the raw buffer which is the original string passed as an
+     argument of regexec(), re_search(), etc..  */
+  const unsigned char *raw_mbs;
+  /* Store the multibyte string.  In case of "case insensitive mode" like
+     REG_ICASE, upper cases of the string are stored, otherwise MBS points
+     the same address that RAW_MBS points.  */
+  unsigned char *mbs;
+#ifdef RE_ENABLE_I18N
+  /* Store the wide character string which is corresponding to MBS.  */
+  wint_t *wcs;
+  int *offsets;
+  mbstate_t cur_state;
+#endif
+  /* Index in RAW_MBS.  Each character mbs[i] corresponds to
+     raw_mbs[raw_mbs_idx + i].  */
+  int raw_mbs_idx;
+  /* The length of the valid characters in the buffers.  */
+  int valid_len;
+  /* The corresponding number of bytes in raw_mbs array.  */
+  int valid_raw_len;
+  /* The length of the buffers MBS and WCS.  */
+  int bufs_len;
+  /* The index in MBS, which is updated by re_string_fetch_byte.  */
+  int cur_idx;
+  /* length of RAW_MBS array.  */
+  int raw_len;
+  /* This is RAW_LEN - RAW_MBS_IDX + VALID_LEN - VALID_RAW_LEN.  */
+  int len;
+  /* End of the buffer may be shorter than its length in the cases such
+     as re_match_2, re_search_2.  Then, we use STOP for end of the buffer
+     instead of LEN.  */
+  int raw_stop;
+  /* This is RAW_STOP - RAW_MBS_IDX adjusted through OFFSETS.  */
+  int stop;
+
+  /* The context of mbs[0].  We store the context independently, since
+     the context of mbs[0] may be different from raw_mbs[0], which is
+     the beginning of the input string.  */
+  unsigned int tip_context;
+  /* The translation passed as a part of an argument of re_compile_pattern.  */
+  RE_TRANSLATE_TYPE trans;
+  /* Copy of re_dfa_t's word_char.  */
+  re_const_bitset_ptr_t word_char;
+  /* 1 if REG_ICASE.  */
+  unsigned char icase;
+  unsigned char is_utf8;
+  unsigned char map_notascii;
+  unsigned char mbs_allocated;
+  unsigned char offsets_needed;
+  unsigned char newline_anchor;
+  unsigned char word_ops_used;
+  int mb_cur_max;
+};
+typedef struct re_string_t re_string_t;
+
+
+struct re_dfa_t;
+typedef struct re_dfa_t re_dfa_t;
+
+#ifndef _LIBC
+# ifdef __i386__
+#  define internal_function   __attribute ((regparm (3), stdcall))
+# else
+#  define internal_function
+# endif
+#endif
+
+#ifndef NOT_IN_libc
+static reg_errcode_t re_string_realloc_buffers (re_string_t *pstr,
+						int new_buf_len)
+     internal_function;
+# ifdef RE_ENABLE_I18N
+static void build_wcs_buffer (re_string_t *pstr) internal_function;
+static reg_errcode_t build_wcs_upper_buffer (re_string_t *pstr)
+  internal_function;
+# endif /* RE_ENABLE_I18N */
+static void build_upper_buffer (re_string_t *pstr) internal_function;
+static void re_string_translate_buffer (re_string_t *pstr) internal_function;
+static unsigned int re_string_context_at (const re_string_t *input, int idx,
+					  int eflags)
+     internal_function __attribute ((pure));
+#endif
+#define re_string_peek_byte(pstr, offset) \
+  ((pstr)->mbs[(pstr)->cur_idx + offset])
+#define re_string_fetch_byte(pstr) \
+  ((pstr)->mbs[(pstr)->cur_idx++])
+#define re_string_first_byte(pstr, idx) \
+  ((idx) == (pstr)->valid_len || (pstr)->wcs[idx] != WEOF)
+#define re_string_is_single_byte_char(pstr, idx) \
+  ((pstr)->wcs[idx] != WEOF && ((pstr)->valid_len == (idx) + 1 \
+				|| (pstr)->wcs[(idx) + 1] != WEOF))
+#define re_string_eoi(pstr) ((pstr)->stop <= (pstr)->cur_idx)
+#define re_string_cur_idx(pstr) ((pstr)->cur_idx)
+#define re_string_get_buffer(pstr) ((pstr)->mbs)
+#define re_string_length(pstr) ((pstr)->len)
+#define re_string_byte_at(pstr,idx) ((pstr)->mbs[idx])
+#define re_string_skip_bytes(pstr,idx) ((pstr)->cur_idx += (idx))
+#define re_string_set_index(pstr,idx) ((pstr)->cur_idx = (idx))
+
+#ifndef _LIBC
+# if HAVE_ALLOCA
+#  if (_MSC_VER)
+#   include <malloc.h>
+#   define __libc_use_alloca(n) 0
+#  else
+#   include <alloca.h>
+/* The OS usually guarantees only one guard page at the bottom of the stack,
+   and a page size can be as small as 4096 bytes.  So we cannot safely
+   allocate anything larger than 4096 bytes.  Also care for the possibility
+   of a few compiler-allocated temporary stack slots.  */
+#  define __libc_use_alloca(n) ((n) < 4032)
+#  endif
+# else
+/* alloca is implemented with malloc, so just use malloc.  */
+#  define __libc_use_alloca(n) 0
+# endif
+#endif
+
+#define re_malloc(t,n) ((t *) malloc ((n) * sizeof (t)))
+/* SunOS 4.1.x realloc doesn't accept null pointers: pre-Standard C. Sigh. */
+#define re_realloc(p,t,n) ((p != NULL) ? (t *) realloc (p,(n)*sizeof(t)) : (t *) calloc(n,sizeof(t)))
+#define re_free(p) free (p)
+
+struct bin_tree_t
+{
+  struct bin_tree_t *parent;
+  struct bin_tree_t *left;
+  struct bin_tree_t *right;
+  struct bin_tree_t *first;
+  struct bin_tree_t *next;
+
+  re_token_t token;
+
+  /* `node_idx' is the index in dfa->nodes, if `type' == 0.
+     Otherwise `type' indicate the type of this node.  */
+  int node_idx;
+};
+typedef struct bin_tree_t bin_tree_t;
+
+#define BIN_TREE_STORAGE_SIZE \
+  ((1024 - sizeof (void *)) / sizeof (bin_tree_t))
+
+struct bin_tree_storage_t
+{
+  struct bin_tree_storage_t *next;
+  bin_tree_t data[BIN_TREE_STORAGE_SIZE];
+};
+typedef struct bin_tree_storage_t bin_tree_storage_t;
+
+#define CONTEXT_WORD 1
+#define CONTEXT_NEWLINE (CONTEXT_WORD << 1)
+#define CONTEXT_BEGBUF (CONTEXT_NEWLINE << 1)
+#define CONTEXT_ENDBUF (CONTEXT_BEGBUF << 1)
+
+#define IS_WORD_CONTEXT(c) ((c) & CONTEXT_WORD)
+#define IS_NEWLINE_CONTEXT(c) ((c) & CONTEXT_NEWLINE)
+#define IS_BEGBUF_CONTEXT(c) ((c) & CONTEXT_BEGBUF)
+#define IS_ENDBUF_CONTEXT(c) ((c) & CONTEXT_ENDBUF)
+#define IS_ORDINARY_CONTEXT(c) ((c) == 0)
+
+#define IS_WORD_CHAR(ch) (isalnum (ch) || (ch) == '_')
+#define IS_NEWLINE(ch) ((ch) == NEWLINE_CHAR)
+#define IS_WIDE_WORD_CHAR(ch) (iswalnum (ch) || (ch) == L'_')
+#define IS_WIDE_NEWLINE(ch) ((ch) == WIDE_NEWLINE_CHAR)
+
+#define NOT_SATISFY_PREV_CONSTRAINT(constraint,context) \
+ ((((constraint) & PREV_WORD_CONSTRAINT) && !IS_WORD_CONTEXT (context)) \
+  || ((constraint & PREV_NOTWORD_CONSTRAINT) && IS_WORD_CONTEXT (context)) \
+  || ((constraint & PREV_NEWLINE_CONSTRAINT) && !IS_NEWLINE_CONTEXT (context))\
+  || ((constraint & PREV_BEGBUF_CONSTRAINT) && !IS_BEGBUF_CONTEXT (context)))
+
+#define NOT_SATISFY_NEXT_CONSTRAINT(constraint,context) \
+ ((((constraint) & NEXT_WORD_CONSTRAINT) && !IS_WORD_CONTEXT (context)) \
+  || (((constraint) & NEXT_NOTWORD_CONSTRAINT) && IS_WORD_CONTEXT (context)) \
+  || (((constraint) & NEXT_NEWLINE_CONSTRAINT) && !IS_NEWLINE_CONTEXT (context)) \
+  || (((constraint) & NEXT_ENDBUF_CONSTRAINT) && !IS_ENDBUF_CONTEXT (context)))
+
+struct re_dfastate_t
+{
+  unsigned int hash;
+  re_node_set nodes;
+  re_node_set non_eps_nodes;
+  re_node_set inveclosure;
+  re_node_set *entrance_nodes;
+  struct re_dfastate_t **trtable, **word_trtable;
+  unsigned int context : 4;
+  unsigned int halt : 1;
+  /* If this state can accept `multi byte'.
+     Note that we refer to multibyte characters, and multi character
+     collating elements as `multi byte'.  */
+  unsigned int accept_mb : 1;
+  /* If this state has backreference node(s).  */
+  unsigned int has_backref : 1;
+  unsigned int has_constraint : 1;
+};
+typedef struct re_dfastate_t re_dfastate_t;
+
+struct re_state_table_entry
+{
+  int num;
+  int alloc;
+  re_dfastate_t **array;
+};
+
+/* Array type used in re_sub_match_last_t and re_sub_match_top_t.  */
+
+typedef struct
+{
+  int next_idx;
+  int alloc;
+  re_dfastate_t **array;
+} state_array_t;
+
+/* Store information about the node NODE whose type is OP_CLOSE_SUBEXP.  */
+
+typedef struct
+{
+  int node;
+  int str_idx; /* The position NODE match at.  */
+  state_array_t path;
+} re_sub_match_last_t;
+
+/* Store information about the node NODE whose type is OP_OPEN_SUBEXP.
+   And information about the node, whose type is OP_CLOSE_SUBEXP,
+   corresponding to NODE is stored in LASTS.  */
+
+typedef struct
+{
+  int str_idx;
+  int node;
+  state_array_t *path;
+  int alasts; /* Allocation size of LASTS.  */
+  int nlasts; /* The number of LASTS.  */
+  re_sub_match_last_t **lasts;
+} re_sub_match_top_t;
+
+struct re_backref_cache_entry
+{
+  int node;
+  int str_idx;
+  int subexp_from;
+  int subexp_to;
+  char more;
+  char unused;
+  unsigned short int eps_reachable_subexps_map;
+};
+
+typedef struct
+{
+  /* The string object corresponding to the input string.  */
+  re_string_t input;
+#if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
+  const re_dfa_t *const dfa;
+#else
+  const re_dfa_t *dfa;
+#endif
+  /* EFLAGS of the argument of regexec.  */
+  int eflags;
+  /* Where the matching ends.  */
+  int match_last;
+  int last_node;
+  /* The state log used by the matcher.  */
+  re_dfastate_t **state_log;
+  int state_log_top;
+  /* Back reference cache.  */
+  int nbkref_ents;
+  int abkref_ents;
+  struct re_backref_cache_entry *bkref_ents;
+  int max_mb_elem_len;
+  int nsub_tops;
+  int asub_tops;
+  re_sub_match_top_t **sub_tops;
+} re_match_context_t;
+
+typedef struct
+{
+  re_dfastate_t **sifted_states;
+  re_dfastate_t **limited_states;
+  int last_node;
+  int last_str_idx;
+  re_node_set limits;
+} re_sift_context_t;
+
+struct re_fail_stack_ent_t
+{
+  int idx;
+  int node;
+  regmatch_t *regs;
+  re_node_set eps_via_nodes;
+};
+
+struct re_fail_stack_t
+{
+  int num;
+  int alloc;
+  struct re_fail_stack_ent_t *stack;
+};
+
+struct re_dfa_t
+{
+  re_token_t *nodes;
+  size_t nodes_alloc;
+  size_t nodes_len;
+  int *nexts;
+  int *org_indices;
+  re_node_set *edests;
+  re_node_set *eclosures;
+  re_node_set *inveclosures;
+  struct re_state_table_entry *state_table;
+  re_dfastate_t *init_state;
+  re_dfastate_t *init_state_word;
+  re_dfastate_t *init_state_nl;
+  re_dfastate_t *init_state_begbuf;
+  bin_tree_t *str_tree;
+  bin_tree_storage_t *str_tree_storage;
+  re_bitset_ptr_t sb_char;
+  int str_tree_storage_idx;
+
+  /* number of subexpressions `re_nsub' is in regex_t.  */
+  unsigned int state_hash_mask;
+  int init_node;
+  int nbackref; /* The number of backreference in this dfa.  */
+
+  /* Bitmap expressing which backreference is used.  */
+  bitset_word_t used_bkref_map;
+  bitset_word_t completed_bkref_map;
+
+  unsigned int has_plural_match : 1;
+  /* If this dfa has "multibyte node", which is a backreference or
+     a node which can accept multibyte character or multi character
+     collating element.  */
+  unsigned int has_mb_node : 1;
+  unsigned int is_utf8 : 1;
+  unsigned int map_notascii : 1;
+  unsigned int word_ops_used : 1;
+  int mb_cur_max;
+  bitset_t word_char;
+  reg_syntax_t syntax;
+  int *subexp_map;
+#ifdef DEBUG
+  char* re_str;
+#endif
+#if defined _LIBC
+  __libc_lock_define (, lock)
+#endif
+};
+
+#define re_node_set_init_empty(set) memset (set, '\0', sizeof (re_node_set))
+#define re_node_set_remove(set,id) \
+  (re_node_set_remove_at (set, re_node_set_contains (set, id) - 1))
+#define re_node_set_empty(p) ((p)->nelem = 0)
+#define re_node_set_free(set) re_free ((set)->elems)
+
+
+typedef enum
+{
+  SB_CHAR,
+  MB_CHAR,
+  EQUIV_CLASS,
+  COLL_SYM,
+  CHAR_CLASS
+} bracket_elem_type;
+
+typedef struct
+{
+  bracket_elem_type type;
+  union
+  {
+    unsigned char ch;
+    unsigned char *name;
+    wchar_t wch;
+  } opr;
+} bracket_elem_t;
+
+
+/* Inline functions for bitset operation.  */
+static inline void
+bitset_not (bitset_t set)
+{
+  int bitset_i;
+  for (bitset_i = 0; bitset_i < BITSET_WORDS; ++bitset_i)
+    set[bitset_i] = ~set[bitset_i];
+}
+
+static inline void
+bitset_merge (bitset_t dest, const bitset_t src)
+{
+  int bitset_i;
+  for (bitset_i = 0; bitset_i < BITSET_WORDS; ++bitset_i)
+    dest[bitset_i] |= src[bitset_i];
+}
+
+static inline void
+bitset_mask (bitset_t dest, const bitset_t src)
+{
+  int bitset_i;
+  for (bitset_i = 0; bitset_i < BITSET_WORDS; ++bitset_i)
+    dest[bitset_i] &= src[bitset_i];
+}
+
+#ifdef RE_ENABLE_I18N
+/* Inline functions for re_string.  */
+static inline int
+internal_function __attribute ((pure))
+re_string_char_size_at (const re_string_t *pstr, int idx)
+{
+  int byte_idx;
+  if (pstr->mb_cur_max == 1)
+    return 1;
+  for (byte_idx = 1; idx + byte_idx < pstr->valid_len; ++byte_idx)
+    if (pstr->wcs[idx + byte_idx] != WEOF)
+      break;
+  return byte_idx;
+}
+
+static inline wint_t
+internal_function __attribute ((pure))
+re_string_wchar_at (const re_string_t *pstr, int idx)
+{
+  if (pstr->mb_cur_max == 1)
+    return (wint_t) pstr->mbs[idx];
+  return (wint_t) pstr->wcs[idx];
+}
+
+# ifndef NOT_IN_libc
+static int
+internal_function __attribute ((pure))
+re_string_elem_size_at (const re_string_t *pstr, int idx)
+{
+#  ifdef _LIBC
+  const unsigned char *p, *extra;
+  const int32_t *table, *indirect;
+  int32_t tmp;
+#   include <locale/weight.h>
+  uint_fast32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
+
+  if (nrules != 0)
+    {
+      table = (const int32_t *) _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
+      extra = (const unsigned char *)
+	_NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
+      indirect = (const int32_t *) _NL_CURRENT (LC_COLLATE,
+						_NL_COLLATE_INDIRECTMB);
+      p = pstr->mbs + idx;
+      tmp = findidx (&p);
+      return p - pstr->mbs - idx;
+    }
+  else
+#  endif /* _LIBC */
+    return 1;
+}
+# endif
+#endif /* RE_ENABLE_I18N */
+
+#endif /*  _REGEX_INTERNAL_H */
diff --git a/compat/regex/regexec.c b/compat/regex/regexec.c
new file mode 100644
index 000000000000..1b5d89fd5ed1
--- /dev/null
+++ b/compat/regex/regexec.c
@@ -0,0 +1,4368 @@
+/* Extended regular expression matching and search library.
+   Copyright (C) 2002-2005, 2007, 2009, 2010 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Isamu Hasegawa <isamu@yamato.ibm.com>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+static reg_errcode_t match_ctx_init (re_match_context_t *cache, int eflags,
+				     int n) internal_function;
+static void match_ctx_clean (re_match_context_t *mctx) internal_function;
+static void match_ctx_free (re_match_context_t *cache) internal_function;
+static reg_errcode_t match_ctx_add_entry (re_match_context_t *cache, int node,
+					  int str_idx, int from, int to)
+     internal_function;
+static int search_cur_bkref_entry (const re_match_context_t *mctx, int str_idx)
+     internal_function;
+static reg_errcode_t match_ctx_add_subtop (re_match_context_t *mctx, int node,
+					   int str_idx) internal_function;
+static re_sub_match_last_t * match_ctx_add_sublast (re_sub_match_top_t *subtop,
+						   int node, int str_idx)
+     internal_function;
+static void sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
+			   re_dfastate_t **limited_sts, int last_node,
+			   int last_str_idx)
+     internal_function;
+static reg_errcode_t re_search_internal (const regex_t *preg,
+					 const char *string, int length,
+					 int start, int range, int stop,
+					 size_t nmatch, regmatch_t pmatch[],
+					 int eflags);
+static int re_search_2_stub (struct re_pattern_buffer *bufp,
+			     const char *string1, int length1,
+			     const char *string2, int length2,
+			     int start, int range, struct re_registers *regs,
+			     int stop, int ret_len);
+static int re_search_stub (struct re_pattern_buffer *bufp,
+			   const char *string, int length, int start,
+			   int range, int stop, struct re_registers *regs,
+			   int ret_len);
+static unsigned re_copy_regs (struct re_registers *regs, regmatch_t *pmatch,
+			      int nregs, int regs_allocated);
+static reg_errcode_t prune_impossible_nodes (re_match_context_t *mctx);
+static int check_matching (re_match_context_t *mctx, int fl_longest_match,
+			   int *p_match_first) internal_function;
+static int check_halt_state_context (const re_match_context_t *mctx,
+				     const re_dfastate_t *state, int idx)
+     internal_function;
+static void update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
+			 regmatch_t *prev_idx_match, int cur_node,
+			 int cur_idx, int nmatch) internal_function;
+static reg_errcode_t push_fail_stack (struct re_fail_stack_t *fs,
+				      int str_idx, int dest_node, int nregs,
+				      regmatch_t *regs,
+				      re_node_set *eps_via_nodes)
+     internal_function;
+static reg_errcode_t set_regs (const regex_t *preg,
+			       const re_match_context_t *mctx,
+			       size_t nmatch, regmatch_t *pmatch,
+			       int fl_backtrack) internal_function;
+static reg_errcode_t free_fail_stack_return (struct re_fail_stack_t *fs)
+     internal_function;
+
+#ifdef RE_ENABLE_I18N
+static int sift_states_iter_mb (const re_match_context_t *mctx,
+				re_sift_context_t *sctx,
+				int node_idx, int str_idx, int max_str_idx)
+     internal_function;
+#endif /* RE_ENABLE_I18N */
+static reg_errcode_t sift_states_backward (const re_match_context_t *mctx,
+					   re_sift_context_t *sctx)
+     internal_function;
+static reg_errcode_t build_sifted_states (const re_match_context_t *mctx,
+					  re_sift_context_t *sctx, int str_idx,
+					  re_node_set *cur_dest)
+     internal_function;
+static reg_errcode_t update_cur_sifted_state (const re_match_context_t *mctx,
+					      re_sift_context_t *sctx,
+					      int str_idx,
+					      re_node_set *dest_nodes)
+     internal_function;
+static reg_errcode_t add_epsilon_src_nodes (const re_dfa_t *dfa,
+					    re_node_set *dest_nodes,
+					    const re_node_set *candidates)
+     internal_function;
+static int check_dst_limits (const re_match_context_t *mctx,
+			     re_node_set *limits,
+			     int dst_node, int dst_idx, int src_node,
+			     int src_idx) internal_function;
+static int check_dst_limits_calc_pos_1 (const re_match_context_t *mctx,
+					int boundaries, int subexp_idx,
+					int from_node, int bkref_idx)
+     internal_function;
+static int check_dst_limits_calc_pos (const re_match_context_t *mctx,
+				      int limit, int subexp_idx,
+				      int node, int str_idx,
+				      int bkref_idx) internal_function;
+static reg_errcode_t check_subexp_limits (const re_dfa_t *dfa,
+					  re_node_set *dest_nodes,
+					  const re_node_set *candidates,
+					  re_node_set *limits,
+					  struct re_backref_cache_entry *bkref_ents,
+					  int str_idx) internal_function;
+static reg_errcode_t sift_states_bkref (const re_match_context_t *mctx,
+					re_sift_context_t *sctx,
+					int str_idx, const re_node_set *candidates)
+     internal_function;
+static reg_errcode_t merge_state_array (const re_dfa_t *dfa,
+					re_dfastate_t **dst,
+					re_dfastate_t **src, int num)
+     internal_function;
+static re_dfastate_t *find_recover_state (reg_errcode_t *err,
+					 re_match_context_t *mctx) internal_function;
+static re_dfastate_t *transit_state (reg_errcode_t *err,
+				     re_match_context_t *mctx,
+				     re_dfastate_t *state) internal_function;
+static re_dfastate_t *merge_state_with_log (reg_errcode_t *err,
+					    re_match_context_t *mctx,
+					    re_dfastate_t *next_state)
+     internal_function;
+static reg_errcode_t check_subexp_matching_top (re_match_context_t *mctx,
+						re_node_set *cur_nodes,
+						int str_idx) internal_function;
+#if 0
+static re_dfastate_t *transit_state_sb (reg_errcode_t *err,
+					re_match_context_t *mctx,
+					re_dfastate_t *pstate)
+     internal_function;
+#endif
+#ifdef RE_ENABLE_I18N
+static reg_errcode_t transit_state_mb (re_match_context_t *mctx,
+				       re_dfastate_t *pstate)
+     internal_function;
+#endif /* RE_ENABLE_I18N */
+static reg_errcode_t transit_state_bkref (re_match_context_t *mctx,
+					  const re_node_set *nodes)
+     internal_function;
+static reg_errcode_t get_subexp (re_match_context_t *mctx,
+				 int bkref_node, int bkref_str_idx)
+     internal_function;
+static reg_errcode_t get_subexp_sub (re_match_context_t *mctx,
+				     const re_sub_match_top_t *sub_top,
+				     re_sub_match_last_t *sub_last,
+				     int bkref_node, int bkref_str)
+     internal_function;
+static int find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
+			     int subexp_idx, int type) internal_function;
+static reg_errcode_t check_arrival (re_match_context_t *mctx,
+				    state_array_t *path, int top_node,
+				    int top_str, int last_node, int last_str,
+				    int type) internal_function;
+static reg_errcode_t check_arrival_add_next_nodes (re_match_context_t *mctx,
+						   int str_idx,
+						   re_node_set *cur_nodes,
+						   re_node_set *next_nodes)
+     internal_function;
+static reg_errcode_t check_arrival_expand_ecl (const re_dfa_t *dfa,
+					       re_node_set *cur_nodes,
+					       int ex_subexp, int type)
+     internal_function;
+static reg_errcode_t check_arrival_expand_ecl_sub (const re_dfa_t *dfa,
+						   re_node_set *dst_nodes,
+						   int target, int ex_subexp,
+						   int type) internal_function;
+static reg_errcode_t expand_bkref_cache (re_match_context_t *mctx,
+					 re_node_set *cur_nodes, int cur_str,
+					 int subexp_num, int type)
+     internal_function;
+static int build_trtable (const re_dfa_t *dfa,
+			  re_dfastate_t *state) internal_function;
+#ifdef RE_ENABLE_I18N
+static int check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
+				    const re_string_t *input, int idx)
+     internal_function;
+# ifdef _LIBC
+static unsigned int find_collation_sequence_value (const unsigned char *mbs,
+						   size_t name_len)
+     internal_function;
+# endif /* _LIBC */
+#endif /* RE_ENABLE_I18N */
+static int group_nodes_into_DFAstates (const re_dfa_t *dfa,
+				       const re_dfastate_t *state,
+				       re_node_set *states_node,
+				       bitset_t *states_ch) internal_function;
+static int check_node_accept (const re_match_context_t *mctx,
+			      const re_token_t *node, int idx)
+     internal_function;
+static reg_errcode_t extend_buffers (re_match_context_t *mctx)
+     internal_function;
+
+/* Entry point for POSIX code.  */
+
+/* regexec searches for a given pattern, specified by PREG, in the
+   string STRING.
+
+   If NMATCH is zero or REG_NOSUB was set in the cflags argument to
+   `regcomp', we ignore PMATCH.  Otherwise, we assume PMATCH has at
+   least NMATCH elements, and we set them to the offsets of the
+   corresponding matched substrings.
+
+   EFLAGS specifies `execution flags' which affect matching: if
+   REG_NOTBOL is set, then ^ does not match at the beginning of the
+   string; if REG_NOTEOL is set, then $ does not match at the end.
+
+   We return 0 if we find a match and REG_NOMATCH if not.  */
+
+int
+regexec (
+	const regex_t *__restrict preg,
+	const char *__restrict string,
+	size_t nmatch,
+	regmatch_t pmatch[],
+	int eflags)
+{
+  reg_errcode_t err;
+  int start, length;
+
+  if (eflags & ~(REG_NOTBOL | REG_NOTEOL | REG_STARTEND))
+    return REG_BADPAT;
+
+  if (eflags & REG_STARTEND)
+    {
+      start = pmatch[0].rm_so;
+      length = pmatch[0].rm_eo;
+    }
+  else
+    {
+      start = 0;
+      length = strlen (string);
+    }
+
+  __libc_lock_lock (dfa->lock);
+  if (preg->no_sub)
+    err = re_search_internal (preg, string, length, start, length - start,
+			      length, 0, NULL, eflags);
+  else
+    err = re_search_internal (preg, string, length, start, length - start,
+			      length, nmatch, pmatch, eflags);
+  __libc_lock_unlock (dfa->lock);
+  return err != REG_NOERROR;
+}
+
+#ifdef _LIBC
+# include <shlib-compat.h>
+versioned_symbol (libc, __regexec, regexec, GLIBC_2_3_4);
+
+# if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
+__typeof__ (__regexec) __compat_regexec;
+
+int
+attribute_compat_text_section
+__compat_regexec (const regex_t *__restrict preg,
+		  const char *__restrict string, size_t nmatch,
+		  regmatch_t pmatch[], int eflags)
+{
+  return regexec (preg, string, nmatch, pmatch,
+		  eflags & (REG_NOTBOL | REG_NOTEOL));
+}
+compat_symbol (libc, __compat_regexec, regexec, GLIBC_2_0);
+# endif
+#endif
+
+/* Entry points for GNU code.  */
+
+/* re_match, re_search, re_match_2, re_search_2
+
+   The former two functions operate on STRING with length LENGTH,
+   while the later two operate on concatenation of STRING1 and STRING2
+   with lengths LENGTH1 and LENGTH2, respectively.
+
+   re_match() matches the compiled pattern in BUFP against the string,
+   starting at index START.
+
+   re_search() first tries matching at index START, then it tries to match
+   starting from index START + 1, and so on.  The last start position tried
+   is START + RANGE.  (Thus RANGE = 0 forces re_search to operate the same
+   way as re_match().)
+
+   The parameter STOP of re_{match,search}_2 specifies that no match exceeding
+   the first STOP characters of the concatenation of the strings should be
+   concerned.
+
+   If REGS is not NULL, and BUFP->no_sub is not set, the offsets of the match
+   and all groups is stroed in REGS.  (For the "_2" variants, the offsets are
+   computed relative to the concatenation, not relative to the individual
+   strings.)
+
+   On success, re_match* functions return the length of the match, re_search*
+   return the position of the start of the match.  Return value -1 means no
+   match was found and -2 indicates an internal error.  */
+
+int
+re_match (struct re_pattern_buffer *bufp,
+	  const char *string,
+	  int length,
+	  int start,
+	  struct re_registers *regs)
+{
+  return re_search_stub (bufp, string, length, start, 0, length, regs, 1);
+}
+#ifdef _LIBC
+weak_alias (__re_match, re_match)
+#endif
+
+int
+re_search (struct re_pattern_buffer *bufp,
+	   const char *string,
+	   int length, int start, int range,
+	   struct re_registers *regs)
+{
+  return re_search_stub (bufp, string, length, start, range, length, regs, 0);
+}
+#ifdef _LIBC
+weak_alias (__re_search, re_search)
+#endif
+
+int
+re_match_2 (struct re_pattern_buffer *bufp,
+	    const char *string1, int length1,
+	    const char *string2, int length2, int start,
+	    struct re_registers *regs, int stop)
+{
+  return re_search_2_stub (bufp, string1, length1, string2, length2,
+			   start, 0, regs, stop, 1);
+}
+#ifdef _LIBC
+weak_alias (__re_match_2, re_match_2)
+#endif
+
+int
+re_search_2 (struct re_pattern_buffer *bufp,
+	     const char *string1, int length1,
+	     const char *string2, int length2, int start,
+	     int range, struct re_registers *regs,  int stop)
+{
+  return re_search_2_stub (bufp, string1, length1, string2, length2,
+			   start, range, regs, stop, 0);
+}
+#ifdef _LIBC
+weak_alias (__re_search_2, re_search_2)
+#endif
+
+static int
+re_search_2_stub (struct re_pattern_buffer *bufp,
+		  const char *string1, int length1,
+		  const char *string2, int length2, int start,
+		  int range, struct re_registers *regs,
+		  int stop, int ret_len)
+{
+  const char *str;
+  int rval;
+  int len = length1 + length2;
+  int free_str = 0;
+
+  if (BE (length1 < 0 || length2 < 0 || stop < 0, 0))
+    return -2;
+
+  /* Concatenate the strings.  */
+  if (length2 > 0)
+    if (length1 > 0)
+      {
+	char *s = re_malloc (char, len);
+
+	if (BE (s == NULL, 0))
+	  return -2;
+	memcpy (s, string1, length1);
+	memcpy (s + length1, string2, length2);
+	str = s;
+	free_str = 1;
+      }
+    else
+      str = string2;
+  else
+    str = string1;
+
+  rval = re_search_stub (bufp, str, len, start, range, stop, regs, ret_len);
+  if (free_str)
+    re_free ((char *) str);
+  return rval;
+}
+
+/* The parameters have the same meaning as those of re_search.
+   Additional parameters:
+   If RET_LEN is nonzero the length of the match is returned (re_match style);
+   otherwise the position of the match is returned.  */
+
+static int
+re_search_stub (struct re_pattern_buffer *bufp,
+		const char *string, int length, int start,
+		int range, int stop,
+		struct re_registers *regs, int ret_len)
+{
+  reg_errcode_t result;
+  regmatch_t *pmatch;
+  int nregs, rval;
+  int eflags = 0;
+
+  /* Check for out-of-range.  */
+  if (BE (start < 0 || start > length, 0))
+    return -1;
+  if (BE (start + range > length, 0))
+    range = length - start;
+  else if (BE (start + range < 0, 0))
+    range = -start;
+
+  __libc_lock_lock (dfa->lock);
+
+  eflags |= (bufp->not_bol) ? REG_NOTBOL : 0;
+  eflags |= (bufp->not_eol) ? REG_NOTEOL : 0;
+
+  /* Compile fastmap if we haven't yet.  */
+  if (range > 0 && bufp->fastmap != NULL && !bufp->fastmap_accurate)
+    re_compile_fastmap (bufp);
+
+  if (BE (bufp->no_sub, 0))
+    regs = NULL;
+
+  /* We need at least 1 register.  */
+  if (regs == NULL)
+    nregs = 1;
+  else if (BE (bufp->regs_allocated == REGS_FIXED &&
+	       regs->num_regs < bufp->re_nsub + 1, 0))
+    {
+      nregs = regs->num_regs;
+      if (BE (nregs < 1, 0))
+	{
+	  /* Nothing can be copied to regs.  */
+	  regs = NULL;
+	  nregs = 1;
+	}
+    }
+  else
+    nregs = bufp->re_nsub + 1;
+  pmatch = re_malloc (regmatch_t, nregs);
+  if (BE (pmatch == NULL, 0))
+    {
+      rval = -2;
+      goto out;
+    }
+
+  result = re_search_internal (bufp, string, length, start, range, stop,
+			       nregs, pmatch, eflags);
+
+  rval = 0;
+
+  /* I hope we needn't fill their regs with -1's when no match was found.  */
+  if (result != REG_NOERROR)
+    rval = -1;
+  else if (regs != NULL)
+    {
+      /* If caller wants register contents data back, copy them.  */
+      bufp->regs_allocated = re_copy_regs (regs, pmatch, nregs,
+					   bufp->regs_allocated);
+      if (BE (bufp->regs_allocated == REGS_UNALLOCATED, 0))
+	rval = -2;
+    }
+
+  if (BE (rval == 0, 1))
+    {
+      if (ret_len)
+	{
+	  assert (pmatch[0].rm_so == start);
+	  rval = pmatch[0].rm_eo - start;
+	}
+      else
+	rval = pmatch[0].rm_so;
+    }
+  re_free (pmatch);
+ out:
+  __libc_lock_unlock (dfa->lock);
+  return rval;
+}
+
+static unsigned
+re_copy_regs (struct re_registers *regs,
+	      regmatch_t *pmatch,
+	      int nregs, int regs_allocated)
+{
+  int rval = REGS_REALLOCATE;
+  int i;
+  int need_regs = nregs + 1;
+  /* We need one extra element beyond `num_regs' for the `-1' marker GNU code
+     uses.  */
+
+  /* Have the register data arrays been allocated?  */
+  if (regs_allocated == REGS_UNALLOCATED)
+    { /* No.  So allocate them with malloc.  */
+      regs->start = re_malloc (regoff_t, need_regs);
+      if (BE (regs->start == NULL, 0))
+	return REGS_UNALLOCATED;
+      regs->end = re_malloc (regoff_t, need_regs);
+      if (BE (regs->end == NULL, 0))
+	{
+	  re_free (regs->start);
+	  return REGS_UNALLOCATED;
+	}
+      regs->num_regs = need_regs;
+    }
+  else if (regs_allocated == REGS_REALLOCATE)
+    { /* Yes.  If we need more elements than were already
+	 allocated, reallocate them.  If we need fewer, just
+	 leave it alone.  */
+      if (BE (need_regs > regs->num_regs, 0))
+	{
+	  regoff_t *new_start = re_realloc (regs->start, regoff_t, need_regs);
+	  regoff_t *new_end;
+	  if (BE (new_start == NULL, 0))
+	    return REGS_UNALLOCATED;
+	  new_end = re_realloc (regs->end, regoff_t, need_regs);
+	  if (BE (new_end == NULL, 0))
+	    {
+	      re_free (new_start);
+	      return REGS_UNALLOCATED;
+	    }
+	  regs->start = new_start;
+	  regs->end = new_end;
+	  regs->num_regs = need_regs;
+	}
+    }
+  else
+    {
+      assert (regs_allocated == REGS_FIXED);
+      /* This function may not be called with REGS_FIXED and nregs too big.  */
+      assert (regs->num_regs >= nregs);
+      rval = REGS_FIXED;
+    }
+
+  /* Copy the regs.  */
+  for (i = 0; i < nregs; ++i)
+    {
+      regs->start[i] = pmatch[i].rm_so;
+      regs->end[i] = pmatch[i].rm_eo;
+    }
+  for ( ; i < regs->num_regs; ++i)
+    regs->start[i] = regs->end[i] = -1;
+
+  return rval;
+}
+
+/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
+   ENDS.  Subsequent matches using PATTERN_BUFFER and REGS will use
+   this memory for recording register information.  STARTS and ENDS
+   must be allocated using the malloc library routine, and must each
+   be at least NUM_REGS * sizeof (regoff_t) bytes long.
+
+   If NUM_REGS == 0, then subsequent matches should allocate their own
+   register data.
+
+   Unless this function is called, the first search or match using
+   PATTERN_BUFFER will allocate its own register data, without
+   freeing the old data.  */
+
+void
+re_set_registers (struct re_pattern_buffer *bufp,
+		  struct re_registers *regs,
+		  unsigned num_regs,
+		  regoff_t *starts,
+		  regoff_t *ends)
+{
+  if (num_regs)
+    {
+      bufp->regs_allocated = REGS_REALLOCATE;
+      regs->num_regs = num_regs;
+      regs->start = starts;
+      regs->end = ends;
+    }
+  else
+    {
+      bufp->regs_allocated = REGS_UNALLOCATED;
+      regs->num_regs = 0;
+      regs->start = regs->end = (regoff_t *) 0;
+    }
+}
+#ifdef _LIBC
+weak_alias (__re_set_registers, re_set_registers)
+#endif
+
+/* Entry points compatible with 4.2 BSD regex library.  We don't define
+   them unless specifically requested.  */
+
+#if defined _REGEX_RE_COMP || defined _LIBC
+int
+# ifdef _LIBC
+weak_function
+# endif
+re_exec (s)
+     const char *s;
+{
+  return 0 == regexec (&re_comp_buf, s, 0, NULL, 0);
+}
+#endif /* _REGEX_RE_COMP */
+
+/* Internal entry point.  */
+
+/* Searches for a compiled pattern PREG in the string STRING, whose
+   length is LENGTH.  NMATCH, PMATCH, and EFLAGS have the same
+   mingings with regexec.  START, and RANGE have the same meanings
+   with re_search.
+   Return REG_NOERROR if we find a match, and REG_NOMATCH if not,
+   otherwise return the error code.
+   Note: We assume front end functions already check ranges.
+   (START + RANGE >= 0 && START + RANGE <= LENGTH)  */
+
+static reg_errcode_t
+re_search_internal (const regex_t *preg,
+		    const char *string,
+		    int length, int start, int range, int stop,
+		    size_t nmatch, regmatch_t pmatch[],
+		    int eflags)
+{
+  reg_errcode_t err;
+  const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer;
+  int left_lim, right_lim, incr;
+  int fl_longest_match, match_first, match_kind, match_last = -1;
+  int extra_nmatch;
+  int sb, ch;
+#if defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
+  re_match_context_t mctx = { .dfa = dfa };
+#else
+  re_match_context_t mctx;
+#endif
+  char *fastmap = (preg->fastmap != NULL && preg->fastmap_accurate
+		   && range && !preg->can_be_null) ? preg->fastmap : NULL;
+  RE_TRANSLATE_TYPE t = preg->translate;
+
+#if !(defined _LIBC || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L))
+  memset (&mctx, '\0', sizeof (re_match_context_t));
+  mctx.dfa = dfa;
+#endif
+
+  extra_nmatch = (nmatch > preg->re_nsub) ? nmatch - (preg->re_nsub + 1) : 0;
+  nmatch -= extra_nmatch;
+
+  /* Check if the DFA haven't been compiled.  */
+  if (BE (preg->used == 0 || dfa->init_state == NULL
+	  || dfa->init_state_word == NULL || dfa->init_state_nl == NULL
+	  || dfa->init_state_begbuf == NULL, 0))
+    return REG_NOMATCH;
+
+#ifdef DEBUG
+  /* We assume front-end functions already check them.  */
+  assert (start + range >= 0 && start + range <= length);
+#endif
+
+  /* If initial states with non-begbuf contexts have no elements,
+     the regex must be anchored.  If preg->newline_anchor is set,
+     we'll never use init_state_nl, so do not check it.  */
+  if (dfa->init_state->nodes.nelem == 0
+      && dfa->init_state_word->nodes.nelem == 0
+      && (dfa->init_state_nl->nodes.nelem == 0
+	  || !preg->newline_anchor))
+    {
+      if (start != 0 && start + range != 0)
+	return REG_NOMATCH;
+      start = range = 0;
+    }
+
+  /* We must check the longest matching, if nmatch > 0.  */
+  fl_longest_match = (nmatch != 0 || dfa->nbackref);
+
+  err = re_string_allocate (&mctx.input, string, length, dfa->nodes_len + 1,
+			    preg->translate, preg->syntax & RE_ICASE, dfa);
+  if (BE (err != REG_NOERROR, 0))
+    goto free_return;
+  mctx.input.stop = stop;
+  mctx.input.raw_stop = stop;
+  mctx.input.newline_anchor = preg->newline_anchor;
+
+  err = match_ctx_init (&mctx, eflags, dfa->nbackref * 2);
+  if (BE (err != REG_NOERROR, 0))
+    goto free_return;
+
+  /* We will log all the DFA states through which the dfa pass,
+     if nmatch > 1, or this dfa has "multibyte node", which is a
+     back-reference or a node which can accept multibyte character or
+     multi character collating element.  */
+  if (nmatch > 1 || dfa->has_mb_node)
+    {
+      /* Avoid overflow.  */
+      if (BE (SIZE_MAX / sizeof (re_dfastate_t *) <= mctx.input.bufs_len, 0))
+	{
+	  err = REG_ESPACE;
+	  goto free_return;
+	}
+
+      mctx.state_log = re_malloc (re_dfastate_t *, mctx.input.bufs_len + 1);
+      if (BE (mctx.state_log == NULL, 0))
+	{
+	  err = REG_ESPACE;
+	  goto free_return;
+	}
+    }
+  else
+    mctx.state_log = NULL;
+
+  match_first = start;
+  mctx.input.tip_context = (eflags & REG_NOTBOL) ? CONTEXT_BEGBUF
+			   : CONTEXT_NEWLINE | CONTEXT_BEGBUF;
+
+  /* Check incrementally whether of not the input string match.  */
+  incr = (range < 0) ? -1 : 1;
+  left_lim = (range < 0) ? start + range : start;
+  right_lim = (range < 0) ? start : start + range;
+  sb = dfa->mb_cur_max == 1;
+  match_kind =
+    (fastmap
+     ? ((sb || !(preg->syntax & RE_ICASE || t) ? 4 : 0)
+	| (range >= 0 ? 2 : 0)
+	| (t != NULL ? 1 : 0))
+     : 8);
+
+  for (;; match_first += incr)
+    {
+      err = REG_NOMATCH;
+      if (match_first < left_lim || right_lim < match_first)
+	goto free_return;
+
+      /* Advance as rapidly as possible through the string, until we
+	 find a plausible place to start matching.  This may be done
+	 with varying efficiency, so there are various possibilities:
+	 only the most common of them are specialized, in order to
+	 save on code size.  We use a switch statement for speed.  */
+      switch (match_kind)
+	{
+	case 8:
+	  /* No fastmap.  */
+	  break;
+
+	case 7:
+	  /* Fastmap with single-byte translation, match forward.  */
+	  while (BE (match_first < right_lim, 1)
+		 && !fastmap[t[(unsigned char) string[match_first]]])
+	    ++match_first;
+	  goto forward_match_found_start_or_reached_end;
+
+	case 6:
+	  /* Fastmap without translation, match forward.  */
+	  while (BE (match_first < right_lim, 1)
+		 && !fastmap[(unsigned char) string[match_first]])
+	    ++match_first;
+
+	forward_match_found_start_or_reached_end:
+	  if (BE (match_first == right_lim, 0))
+	    {
+	      ch = match_first >= length
+		       ? 0 : (unsigned char) string[match_first];
+	      if (!fastmap[t ? t[ch] : ch])
+		goto free_return;
+	    }
+	  break;
+
+	case 4:
+	case 5:
+	  /* Fastmap without multi-byte translation, match backwards.  */
+	  while (match_first >= left_lim)
+	    {
+	      ch = match_first >= length
+		       ? 0 : (unsigned char) string[match_first];
+	      if (fastmap[t ? t[ch] : ch])
+		break;
+	      --match_first;
+	    }
+	  if (match_first < left_lim)
+	    goto free_return;
+	  break;
+
+	default:
+	  /* In this case, we can't determine easily the current byte,
+	     since it might be a component byte of a multibyte
+	     character.  Then we use the constructed buffer instead.  */
+	  for (;;)
+	    {
+	      /* If MATCH_FIRST is out of the valid range, reconstruct the
+		 buffers.  */
+	      unsigned int offset = match_first - mctx.input.raw_mbs_idx;
+	      if (BE (offset >= (unsigned int) mctx.input.valid_raw_len, 0))
+		{
+		  err = re_string_reconstruct (&mctx.input, match_first,
+					       eflags);
+		  if (BE (err != REG_NOERROR, 0))
+		    goto free_return;
+
+		  offset = match_first - mctx.input.raw_mbs_idx;
+		}
+	      /* If MATCH_FIRST is out of the buffer, leave it as '\0'.
+		 Note that MATCH_FIRST must not be smaller than 0.  */
+	      ch = (match_first >= length
+		    ? 0 : re_string_byte_at (&mctx.input, offset));
+	      if (fastmap[ch])
+		break;
+	      match_first += incr;
+	      if (match_first < left_lim || match_first > right_lim)
+		{
+		  err = REG_NOMATCH;
+		  goto free_return;
+		}
+	    }
+	  break;
+	}
+
+      /* Reconstruct the buffers so that the matcher can assume that
+	 the matching starts from the beginning of the buffer.  */
+      err = re_string_reconstruct (&mctx.input, match_first, eflags);
+      if (BE (err != REG_NOERROR, 0))
+	goto free_return;
+
+#ifdef RE_ENABLE_I18N
+     /* Don't consider this char as a possible match start if it part,
+	yet isn't the head, of a multibyte character.  */
+      if (!sb && !re_string_first_byte (&mctx.input, 0))
+	continue;
+#endif
+
+      /* It seems to be appropriate one, then use the matcher.  */
+      /* We assume that the matching starts from 0.  */
+      mctx.state_log_top = mctx.nbkref_ents = mctx.max_mb_elem_len = 0;
+      match_last = check_matching (&mctx, fl_longest_match,
+				   range >= 0 ? &match_first : NULL);
+      if (match_last != -1)
+	{
+	  if (BE (match_last == -2, 0))
+	    {
+	      err = REG_ESPACE;
+	      goto free_return;
+	    }
+	  else
+	    {
+	      mctx.match_last = match_last;
+	      if ((!preg->no_sub && nmatch > 1) || dfa->nbackref)
+		{
+		  re_dfastate_t *pstate = mctx.state_log[match_last];
+		  mctx.last_node = check_halt_state_context (&mctx, pstate,
+							     match_last);
+		}
+	      if ((!preg->no_sub && nmatch > 1 && dfa->has_plural_match)
+		  || dfa->nbackref)
+		{
+		  err = prune_impossible_nodes (&mctx);
+		  if (err == REG_NOERROR)
+		    break;
+		  if (BE (err != REG_NOMATCH, 0))
+		    goto free_return;
+		  match_last = -1;
+		}
+	      else
+		break; /* We found a match.  */
+	    }
+	}
+
+      match_ctx_clean (&mctx);
+    }
+
+#ifdef DEBUG
+  assert (match_last != -1);
+  assert (err == REG_NOERROR);
+#endif
+
+  /* Set pmatch[] if we need.  */
+  if (nmatch > 0)
+    {
+      int reg_idx;
+
+      /* Initialize registers.  */
+      for (reg_idx = 1; reg_idx < nmatch; ++reg_idx)
+	pmatch[reg_idx].rm_so = pmatch[reg_idx].rm_eo = -1;
+
+      /* Set the points where matching start/end.  */
+      pmatch[0].rm_so = 0;
+      pmatch[0].rm_eo = mctx.match_last;
+
+      if (!preg->no_sub && nmatch > 1)
+	{
+	  err = set_regs (preg, &mctx, nmatch, pmatch,
+			  dfa->has_plural_match && dfa->nbackref > 0);
+	  if (BE (err != REG_NOERROR, 0))
+	    goto free_return;
+	}
+
+      /* At last, add the offset to the each registers, since we slided
+	 the buffers so that we could assume that the matching starts
+	 from 0.  */
+      for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
+	if (pmatch[reg_idx].rm_so != -1)
+	  {
+#ifdef RE_ENABLE_I18N
+	    if (BE (mctx.input.offsets_needed != 0, 0))
+	      {
+		pmatch[reg_idx].rm_so =
+		  (pmatch[reg_idx].rm_so == mctx.input.valid_len
+		   ? mctx.input.valid_raw_len
+		   : mctx.input.offsets[pmatch[reg_idx].rm_so]);
+		pmatch[reg_idx].rm_eo =
+		  (pmatch[reg_idx].rm_eo == mctx.input.valid_len
+		   ? mctx.input.valid_raw_len
+		   : mctx.input.offsets[pmatch[reg_idx].rm_eo]);
+	      }
+#else
+	    assert (mctx.input.offsets_needed == 0);
+#endif
+	    pmatch[reg_idx].rm_so += match_first;
+	    pmatch[reg_idx].rm_eo += match_first;
+	  }
+      for (reg_idx = 0; reg_idx < extra_nmatch; ++reg_idx)
+	{
+	  pmatch[nmatch + reg_idx].rm_so = -1;
+	  pmatch[nmatch + reg_idx].rm_eo = -1;
+	}
+
+      if (dfa->subexp_map)
+	for (reg_idx = 0; reg_idx + 1 < nmatch; reg_idx++)
+	  if (dfa->subexp_map[reg_idx] != reg_idx)
+	    {
+	      pmatch[reg_idx + 1].rm_so
+		= pmatch[dfa->subexp_map[reg_idx] + 1].rm_so;
+	      pmatch[reg_idx + 1].rm_eo
+		= pmatch[dfa->subexp_map[reg_idx] + 1].rm_eo;
+	    }
+    }
+
+ free_return:
+  re_free (mctx.state_log);
+  if (dfa->nbackref)
+    match_ctx_free (&mctx);
+  re_string_destruct (&mctx.input);
+  return err;
+}
+
+static reg_errcode_t
+prune_impossible_nodes (re_match_context_t *mctx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int halt_node, match_last;
+  reg_errcode_t ret;
+  re_dfastate_t **sifted_states;
+  re_dfastate_t **lim_states = NULL;
+  re_sift_context_t sctx;
+#ifdef DEBUG
+  assert (mctx->state_log != NULL);
+#endif
+  match_last = mctx->match_last;
+  halt_node = mctx->last_node;
+
+  /* Avoid overflow.  */
+  if (BE (SIZE_MAX / sizeof (re_dfastate_t *) <= match_last, 0))
+    return REG_ESPACE;
+
+  sifted_states = re_malloc (re_dfastate_t *, match_last + 1);
+  if (BE (sifted_states == NULL, 0))
+    {
+      ret = REG_ESPACE;
+      goto free_return;
+    }
+  if (dfa->nbackref)
+    {
+      lim_states = re_malloc (re_dfastate_t *, match_last + 1);
+      if (BE (lim_states == NULL, 0))
+	{
+	  ret = REG_ESPACE;
+	  goto free_return;
+	}
+      while (1)
+	{
+	  memset (lim_states, '\0',
+		  sizeof (re_dfastate_t *) * (match_last + 1));
+	  sift_ctx_init (&sctx, sifted_states, lim_states, halt_node,
+			 match_last);
+	  ret = sift_states_backward (mctx, &sctx);
+	  re_node_set_free (&sctx.limits);
+	  if (BE (ret != REG_NOERROR, 0))
+	      goto free_return;
+	  if (sifted_states[0] != NULL || lim_states[0] != NULL)
+	    break;
+	  do
+	    {
+	      --match_last;
+	      if (match_last < 0)
+		{
+		  ret = REG_NOMATCH;
+		  goto free_return;
+		}
+	    } while (mctx->state_log[match_last] == NULL
+		     || !mctx->state_log[match_last]->halt);
+	  halt_node = check_halt_state_context (mctx,
+						mctx->state_log[match_last],
+						match_last);
+	}
+      ret = merge_state_array (dfa, sifted_states, lim_states,
+			       match_last + 1);
+      re_free (lim_states);
+      lim_states = NULL;
+      if (BE (ret != REG_NOERROR, 0))
+	goto free_return;
+    }
+  else
+    {
+      sift_ctx_init (&sctx, sifted_states, lim_states, halt_node, match_last);
+      ret = sift_states_backward (mctx, &sctx);
+      re_node_set_free (&sctx.limits);
+      if (BE (ret != REG_NOERROR, 0))
+	goto free_return;
+      if (sifted_states[0] == NULL)
+	{
+	  ret = REG_NOMATCH;
+	  goto free_return;
+	}
+    }
+  re_free (mctx->state_log);
+  mctx->state_log = sifted_states;
+  sifted_states = NULL;
+  mctx->last_node = halt_node;
+  mctx->match_last = match_last;
+  ret = REG_NOERROR;
+ free_return:
+  re_free (sifted_states);
+  re_free (lim_states);
+  return ret;
+}
+
+/* Acquire an initial state and return it.
+   We must select appropriate initial state depending on the context,
+   since initial states may have constraints like "\<", "^", etc..  */
+
+static inline re_dfastate_t *
+__attribute ((always_inline)) internal_function
+acquire_init_state_context (reg_errcode_t *err, const re_match_context_t *mctx,
+			    int idx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  if (dfa->init_state->has_constraint)
+    {
+      unsigned int context;
+      context = re_string_context_at (&mctx->input, idx - 1, mctx->eflags);
+      if (IS_WORD_CONTEXT (context))
+	return dfa->init_state_word;
+      else if (IS_ORDINARY_CONTEXT (context))
+	return dfa->init_state;
+      else if (IS_BEGBUF_CONTEXT (context) && IS_NEWLINE_CONTEXT (context))
+	return dfa->init_state_begbuf;
+      else if (IS_NEWLINE_CONTEXT (context))
+	return dfa->init_state_nl;
+      else if (IS_BEGBUF_CONTEXT (context))
+	{
+	  /* It is relatively rare case, then calculate on demand.  */
+	  return re_acquire_state_context (err, dfa,
+					   dfa->init_state->entrance_nodes,
+					   context);
+	}
+      else
+	/* Must not happen?  */
+	return dfa->init_state;
+    }
+  else
+    return dfa->init_state;
+}
+
+/* Check whether the regular expression match input string INPUT or not,
+   and return the index where the matching end, return -1 if not match,
+   or return -2 in case of an error.
+   FL_LONGEST_MATCH means we want the POSIX longest matching.
+   If P_MATCH_FIRST is not NULL, and the match fails, it is set to the
+   next place where we may want to try matching.
+   Note that the matcher assume that the matching starts from the current
+   index of the buffer.  */
+
+static int
+internal_function
+check_matching (re_match_context_t *mctx, int fl_longest_match,
+		int *p_match_first)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err;
+  int match = 0;
+  int match_last = -1;
+  int cur_str_idx = re_string_cur_idx (&mctx->input);
+  re_dfastate_t *cur_state;
+  int at_init_state = p_match_first != NULL;
+  int next_start_idx = cur_str_idx;
+
+  err = REG_NOERROR;
+  cur_state = acquire_init_state_context (&err, mctx, cur_str_idx);
+  /* An initial state must not be NULL (invalid).  */
+  if (BE (cur_state == NULL, 0))
+    {
+      assert (err == REG_ESPACE);
+      return -2;
+    }
+
+  if (mctx->state_log != NULL)
+    {
+      mctx->state_log[cur_str_idx] = cur_state;
+
+      /* Check OP_OPEN_SUBEXP in the initial state in case that we use them
+	 later.  E.g. Processing back references.  */
+      if (BE (dfa->nbackref, 0))
+	{
+	  at_init_state = 0;
+	  err = check_subexp_matching_top (mctx, &cur_state->nodes, 0);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+
+	  if (cur_state->has_backref)
+	    {
+	      err = transit_state_bkref (mctx, &cur_state->nodes);
+	      if (BE (err != REG_NOERROR, 0))
+		return err;
+	    }
+	}
+    }
+
+  /* If the RE accepts NULL string.  */
+  if (BE (cur_state->halt, 0))
+    {
+      if (!cur_state->has_constraint
+	  || check_halt_state_context (mctx, cur_state, cur_str_idx))
+	{
+	  if (!fl_longest_match)
+	    return cur_str_idx;
+	  else
+	    {
+	      match_last = cur_str_idx;
+	      match = 1;
+	    }
+	}
+    }
+
+  while (!re_string_eoi (&mctx->input))
+    {
+      re_dfastate_t *old_state = cur_state;
+      int next_char_idx = re_string_cur_idx (&mctx->input) + 1;
+
+      if (BE (next_char_idx >= mctx->input.bufs_len, 0)
+	  || (BE (next_char_idx >= mctx->input.valid_len, 0)
+	      && mctx->input.valid_len < mctx->input.len))
+	{
+	  err = extend_buffers (mctx);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      assert (err == REG_ESPACE);
+	      return -2;
+	    }
+	}
+
+      cur_state = transit_state (&err, mctx, cur_state);
+      if (mctx->state_log != NULL)
+	cur_state = merge_state_with_log (&err, mctx, cur_state);
+
+      if (cur_state == NULL)
+	{
+	  /* Reached the invalid state or an error.  Try to recover a valid
+	     state using the state log, if available and if we have not
+	     already found a valid (even if not the longest) match.  */
+	  if (BE (err != REG_NOERROR, 0))
+	    return -2;
+
+	  if (mctx->state_log == NULL
+	      || (match && !fl_longest_match)
+	      || (cur_state = find_recover_state (&err, mctx)) == NULL)
+	    break;
+	}
+
+      if (BE (at_init_state, 0))
+	{
+	  if (old_state == cur_state)
+	    next_start_idx = next_char_idx;
+	  else
+	    at_init_state = 0;
+	}
+
+      if (cur_state->halt)
+	{
+	  /* Reached a halt state.
+	     Check the halt state can satisfy the current context.  */
+	  if (!cur_state->has_constraint
+	      || check_halt_state_context (mctx, cur_state,
+					   re_string_cur_idx (&mctx->input)))
+	    {
+	      /* We found an appropriate halt state.  */
+	      match_last = re_string_cur_idx (&mctx->input);
+	      match = 1;
+
+	      /* We found a match, do not modify match_first below.  */
+	      p_match_first = NULL;
+	      if (!fl_longest_match)
+		break;
+	    }
+	}
+    }
+
+  if (p_match_first)
+    *p_match_first += next_start_idx;
+
+  return match_last;
+}
+
+/* Check NODE match the current context.  */
+
+static int
+internal_function
+check_halt_node_context (const re_dfa_t *dfa, int node, unsigned int context)
+{
+  re_token_type_t type = dfa->nodes[node].type;
+  unsigned int constraint = dfa->nodes[node].constraint;
+  if (type != END_OF_RE)
+    return 0;
+  if (!constraint)
+    return 1;
+  if (NOT_SATISFY_NEXT_CONSTRAINT (constraint, context))
+    return 0;
+  return 1;
+}
+
+/* Check the halt state STATE match the current context.
+   Return 0 if not match, if the node, STATE has, is a halt node and
+   match the context, return the node.  */
+
+static int
+internal_function
+check_halt_state_context (const re_match_context_t *mctx,
+			  const re_dfastate_t *state, int idx)
+{
+  int i;
+  unsigned int context;
+#ifdef DEBUG
+  assert (state->halt);
+#endif
+  context = re_string_context_at (&mctx->input, idx, mctx->eflags);
+  for (i = 0; i < state->nodes.nelem; ++i)
+    if (check_halt_node_context (mctx->dfa, state->nodes.elems[i], context))
+      return state->nodes.elems[i];
+  return 0;
+}
+
+/* Compute the next node to which "NFA" transit from NODE("NFA" is a NFA
+   corresponding to the DFA).
+   Return the destination node, and update EPS_VIA_NODES, return -1 in case
+   of errors.  */
+
+static int
+internal_function
+proceed_next_node (const re_match_context_t *mctx, int nregs, regmatch_t *regs,
+		   int *pidx, int node, re_node_set *eps_via_nodes,
+		   struct re_fail_stack_t *fs)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int i, err;
+  if (IS_EPSILON_NODE (dfa->nodes[node].type))
+    {
+      re_node_set *cur_nodes = &mctx->state_log[*pidx]->nodes;
+      re_node_set *edests = &dfa->edests[node];
+      int dest_node;
+      err = re_node_set_insert (eps_via_nodes, node);
+      if (BE (err < 0, 0))
+	return -2;
+      /* Pick up a valid destination, or return -1 if none is found.  */
+      for (dest_node = -1, i = 0; i < edests->nelem; ++i)
+	{
+	  int candidate = edests->elems[i];
+	  if (!re_node_set_contains (cur_nodes, candidate))
+	    continue;
+	  if (dest_node == -1)
+	    dest_node = candidate;
+
+	  else
+	    {
+	      /* In order to avoid infinite loop like "(a*)*", return the second
+		 epsilon-transition if the first was already considered.  */
+	      if (re_node_set_contains (eps_via_nodes, dest_node))
+		return candidate;
+
+	      /* Otherwise, push the second epsilon-transition on the fail stack.  */
+	      else if (fs != NULL
+		       && push_fail_stack (fs, *pidx, candidate, nregs, regs,
+					   eps_via_nodes))
+		return -2;
+
+	      /* We know we are going to exit.  */
+	      break;
+	    }
+	}
+      return dest_node;
+    }
+  else
+    {
+      int naccepted = 0;
+      re_token_type_t type = dfa->nodes[node].type;
+
+#ifdef RE_ENABLE_I18N
+      if (dfa->nodes[node].accept_mb)
+	naccepted = check_node_accept_bytes (dfa, node, &mctx->input, *pidx);
+      else
+#endif /* RE_ENABLE_I18N */
+      if (type == OP_BACK_REF)
+	{
+	  int subexp_idx = dfa->nodes[node].opr.idx + 1;
+	  naccepted = regs[subexp_idx].rm_eo - regs[subexp_idx].rm_so;
+	  if (fs != NULL)
+	    {
+	      if (regs[subexp_idx].rm_so == -1 || regs[subexp_idx].rm_eo == -1)
+		return -1;
+	      else if (naccepted)
+		{
+		  char *buf = (char *) re_string_get_buffer (&mctx->input);
+		  if (memcmp (buf + regs[subexp_idx].rm_so, buf + *pidx,
+			      naccepted) != 0)
+		    return -1;
+		}
+	    }
+
+	  if (naccepted == 0)
+	    {
+	      int dest_node;
+	      err = re_node_set_insert (eps_via_nodes, node);
+	      if (BE (err < 0, 0))
+		return -2;
+	      dest_node = dfa->edests[node].elems[0];
+	      if (re_node_set_contains (&mctx->state_log[*pidx]->nodes,
+					dest_node))
+		return dest_node;
+	    }
+	}
+
+      if (naccepted != 0
+	  || check_node_accept (mctx, dfa->nodes + node, *pidx))
+	{
+	  int dest_node = dfa->nexts[node];
+	  *pidx = (naccepted == 0) ? *pidx + 1 : *pidx + naccepted;
+	  if (fs && (*pidx > mctx->match_last || mctx->state_log[*pidx] == NULL
+		     || !re_node_set_contains (&mctx->state_log[*pidx]->nodes,
+					       dest_node)))
+	    return -1;
+	  re_node_set_empty (eps_via_nodes);
+	  return dest_node;
+	}
+    }
+  return -1;
+}
+
+static reg_errcode_t
+internal_function
+push_fail_stack (struct re_fail_stack_t *fs, int str_idx, int dest_node,
+		 int nregs, regmatch_t *regs, re_node_set *eps_via_nodes)
+{
+  reg_errcode_t err;
+  int num = fs->num++;
+  if (fs->num == fs->alloc)
+    {
+      struct re_fail_stack_ent_t *new_array;
+      new_array = realloc (fs->stack, (sizeof (struct re_fail_stack_ent_t)
+				       * fs->alloc * 2));
+      if (new_array == NULL)
+	return REG_ESPACE;
+      fs->alloc *= 2;
+      fs->stack = new_array;
+    }
+  fs->stack[num].idx = str_idx;
+  fs->stack[num].node = dest_node;
+  fs->stack[num].regs = re_malloc (regmatch_t, nregs);
+  if (fs->stack[num].regs == NULL)
+    return REG_ESPACE;
+  memcpy (fs->stack[num].regs, regs, sizeof (regmatch_t) * nregs);
+  err = re_node_set_init_copy (&fs->stack[num].eps_via_nodes, eps_via_nodes);
+  return err;
+}
+
+static int
+internal_function
+pop_fail_stack (struct re_fail_stack_t *fs, int *pidx, int nregs,
+		regmatch_t *regs, re_node_set *eps_via_nodes)
+{
+  int num = --fs->num;
+  assert (num >= 0);
+  *pidx = fs->stack[num].idx;
+  memcpy (regs, fs->stack[num].regs, sizeof (regmatch_t) * nregs);
+  re_node_set_free (eps_via_nodes);
+  re_free (fs->stack[num].regs);
+  *eps_via_nodes = fs->stack[num].eps_via_nodes;
+  return fs->stack[num].node;
+}
+
+/* Set the positions where the subexpressions are starts/ends to registers
+   PMATCH.
+   Note: We assume that pmatch[0] is already set, and
+   pmatch[i].rm_so == pmatch[i].rm_eo == -1 for 0 < i < nmatch.  */
+
+static reg_errcode_t
+internal_function
+set_regs (const regex_t *preg, const re_match_context_t *mctx, size_t nmatch,
+	  regmatch_t *pmatch, int fl_backtrack)
+{
+  const re_dfa_t *dfa = (const re_dfa_t *) preg->buffer;
+  int idx, cur_node;
+  re_node_set eps_via_nodes;
+  struct re_fail_stack_t *fs;
+  struct re_fail_stack_t fs_body = { 0, 2, NULL };
+  regmatch_t *prev_idx_match;
+  int prev_idx_match_malloced = 0;
+
+#ifdef DEBUG
+  assert (nmatch > 1);
+  assert (mctx->state_log != NULL);
+#endif
+  if (fl_backtrack)
+    {
+      fs = &fs_body;
+      fs->stack = re_malloc (struct re_fail_stack_ent_t, fs->alloc);
+      if (fs->stack == NULL)
+	return REG_ESPACE;
+    }
+  else
+    fs = NULL;
+
+  cur_node = dfa->init_node;
+  re_node_set_init_empty (&eps_via_nodes);
+
+#ifdef HAVE_ALLOCA
+  if (__libc_use_alloca (nmatch * sizeof (regmatch_t)))
+    prev_idx_match = (regmatch_t *) alloca (nmatch * sizeof (regmatch_t));
+  else
+#endif
+    {
+      prev_idx_match = re_malloc (regmatch_t, nmatch);
+      if (prev_idx_match == NULL)
+	{
+	  free_fail_stack_return (fs);
+	  return REG_ESPACE;
+	}
+      prev_idx_match_malloced = 1;
+    }
+  memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
+
+  for (idx = pmatch[0].rm_so; idx <= pmatch[0].rm_eo ;)
+    {
+      update_regs (dfa, pmatch, prev_idx_match, cur_node, idx, nmatch);
+
+      if (idx == pmatch[0].rm_eo && cur_node == mctx->last_node)
+	{
+	  int reg_idx;
+	  if (fs)
+	    {
+	      for (reg_idx = 0; reg_idx < nmatch; ++reg_idx)
+		if (pmatch[reg_idx].rm_so > -1 && pmatch[reg_idx].rm_eo == -1)
+		  break;
+	      if (reg_idx == nmatch)
+		{
+		  re_node_set_free (&eps_via_nodes);
+		  if (prev_idx_match_malloced)
+		    re_free (prev_idx_match);
+		  return free_fail_stack_return (fs);
+		}
+	      cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
+					 &eps_via_nodes);
+	    }
+	  else
+	    {
+	      re_node_set_free (&eps_via_nodes);
+	      if (prev_idx_match_malloced)
+		re_free (prev_idx_match);
+	      return REG_NOERROR;
+	    }
+	}
+
+      /* Proceed to next node.  */
+      cur_node = proceed_next_node (mctx, nmatch, pmatch, &idx, cur_node,
+				    &eps_via_nodes, fs);
+
+      if (BE (cur_node < 0, 0))
+	{
+	  if (BE (cur_node == -2, 0))
+	    {
+	      re_node_set_free (&eps_via_nodes);
+	      if (prev_idx_match_malloced)
+		re_free (prev_idx_match);
+	      free_fail_stack_return (fs);
+	      return REG_ESPACE;
+	    }
+	  if (fs)
+	    cur_node = pop_fail_stack (fs, &idx, nmatch, pmatch,
+				       &eps_via_nodes);
+	  else
+	    {
+	      re_node_set_free (&eps_via_nodes);
+	      if (prev_idx_match_malloced)
+		re_free (prev_idx_match);
+	      return REG_NOMATCH;
+	    }
+	}
+    }
+  re_node_set_free (&eps_via_nodes);
+  if (prev_idx_match_malloced)
+    re_free (prev_idx_match);
+  return free_fail_stack_return (fs);
+}
+
+static reg_errcode_t
+internal_function
+free_fail_stack_return (struct re_fail_stack_t *fs)
+{
+  if (fs)
+    {
+      int fs_idx;
+      for (fs_idx = 0; fs_idx < fs->num; ++fs_idx)
+	{
+	  re_node_set_free (&fs->stack[fs_idx].eps_via_nodes);
+	  re_free (fs->stack[fs_idx].regs);
+	}
+      re_free (fs->stack);
+    }
+  return REG_NOERROR;
+}
+
+static void
+internal_function
+update_regs (const re_dfa_t *dfa, regmatch_t *pmatch,
+	     regmatch_t *prev_idx_match, int cur_node, int cur_idx, int nmatch)
+{
+  int type = dfa->nodes[cur_node].type;
+  if (type == OP_OPEN_SUBEXP)
+    {
+      int reg_num = dfa->nodes[cur_node].opr.idx + 1;
+
+      /* We are at the first node of this sub expression.  */
+      if (reg_num < nmatch)
+	{
+	  pmatch[reg_num].rm_so = cur_idx;
+	  pmatch[reg_num].rm_eo = -1;
+	}
+    }
+  else if (type == OP_CLOSE_SUBEXP)
+    {
+      int reg_num = dfa->nodes[cur_node].opr.idx + 1;
+      if (reg_num < nmatch)
+	{
+	  /* We are at the last node of this sub expression.  */
+	  if (pmatch[reg_num].rm_so < cur_idx)
+	    {
+	      pmatch[reg_num].rm_eo = cur_idx;
+	      /* This is a non-empty match or we are not inside an optional
+		 subexpression.  Accept this right away.  */
+	      memcpy (prev_idx_match, pmatch, sizeof (regmatch_t) * nmatch);
+	    }
+	  else
+	    {
+	      if (dfa->nodes[cur_node].opt_subexp
+		  && prev_idx_match[reg_num].rm_so != -1)
+		/* We transited through an empty match for an optional
+		   subexpression, like (a?)*, and this is not the subexp's
+		   first match.  Copy back the old content of the registers
+		   so that matches of an inner subexpression are undone as
+		   well, like in ((a?))*.  */
+		memcpy (pmatch, prev_idx_match, sizeof (regmatch_t) * nmatch);
+	      else
+		/* We completed a subexpression, but it may be part of
+		   an optional one, so do not update PREV_IDX_MATCH.  */
+		pmatch[reg_num].rm_eo = cur_idx;
+	    }
+	}
+    }
+}
+
+/* This function checks the STATE_LOG from the SCTX->last_str_idx to 0
+   and sift the nodes in each states according to the following rules.
+   Updated state_log will be wrote to STATE_LOG.
+
+   Rules: We throw away the Node `a' in the STATE_LOG[STR_IDX] if...
+     1. When STR_IDX == MATCH_LAST(the last index in the state_log):
+	If `a' isn't the LAST_NODE and `a' can't epsilon transit to
+	the LAST_NODE, we throw away the node `a'.
+     2. When 0 <= STR_IDX < MATCH_LAST and `a' accepts
+	string `s' and transit to `b':
+	i. If 'b' isn't in the STATE_LOG[STR_IDX+strlen('s')], we throw
+	   away the node `a'.
+	ii. If 'b' is in the STATE_LOG[STR_IDX+strlen('s')] but 'b' is
+	    thrown away, we throw away the node `a'.
+     3. When 0 <= STR_IDX < MATCH_LAST and 'a' epsilon transit to 'b':
+	i. If 'b' isn't in the STATE_LOG[STR_IDX], we throw away the
+	   node `a'.
+	ii. If 'b' is in the STATE_LOG[STR_IDX] but 'b' is thrown away,
+	    we throw away the node `a'.  */
+
+#define STATE_NODE_CONTAINS(state,node) \
+  ((state) != NULL && re_node_set_contains (&(state)->nodes, node))
+
+static reg_errcode_t
+internal_function
+sift_states_backward (const re_match_context_t *mctx, re_sift_context_t *sctx)
+{
+  reg_errcode_t err;
+  int null_cnt = 0;
+  int str_idx = sctx->last_str_idx;
+  re_node_set cur_dest;
+
+#ifdef DEBUG
+  assert (mctx->state_log != NULL && mctx->state_log[str_idx] != NULL);
+#endif
+
+  /* Build sifted state_log[str_idx].  It has the nodes which can epsilon
+     transit to the last_node and the last_node itself.  */
+  err = re_node_set_init_1 (&cur_dest, sctx->last_node);
+  if (BE (err != REG_NOERROR, 0))
+    return err;
+  err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
+  if (BE (err != REG_NOERROR, 0))
+    goto free_return;
+
+  /* Then check each states in the state_log.  */
+  while (str_idx > 0)
+    {
+      /* Update counters.  */
+      null_cnt = (sctx->sifted_states[str_idx] == NULL) ? null_cnt + 1 : 0;
+      if (null_cnt > mctx->max_mb_elem_len)
+	{
+	  memset (sctx->sifted_states, '\0',
+		  sizeof (re_dfastate_t *) * str_idx);
+	  re_node_set_free (&cur_dest);
+	  return REG_NOERROR;
+	}
+      re_node_set_empty (&cur_dest);
+      --str_idx;
+
+      if (mctx->state_log[str_idx])
+	{
+	  err = build_sifted_states (mctx, sctx, str_idx, &cur_dest);
+	  if (BE (err != REG_NOERROR, 0))
+	    goto free_return;
+	}
+
+      /* Add all the nodes which satisfy the following conditions:
+	 - It can epsilon transit to a node in CUR_DEST.
+	 - It is in CUR_SRC.
+	 And update state_log.  */
+      err = update_cur_sifted_state (mctx, sctx, str_idx, &cur_dest);
+      if (BE (err != REG_NOERROR, 0))
+	goto free_return;
+    }
+  err = REG_NOERROR;
+ free_return:
+  re_node_set_free (&cur_dest);
+  return err;
+}
+
+static reg_errcode_t
+internal_function
+build_sifted_states (const re_match_context_t *mctx, re_sift_context_t *sctx,
+		     int str_idx, re_node_set *cur_dest)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  const re_node_set *cur_src = &mctx->state_log[str_idx]->non_eps_nodes;
+  int i;
+
+  /* Then build the next sifted state.
+     We build the next sifted state on `cur_dest', and update
+     `sifted_states[str_idx]' with `cur_dest'.
+     Note:
+     `cur_dest' is the sifted state from `state_log[str_idx + 1]'.
+     `cur_src' points the node_set of the old `state_log[str_idx]'
+     (with the epsilon nodes pre-filtered out).  */
+  for (i = 0; i < cur_src->nelem; i++)
+    {
+      int prev_node = cur_src->elems[i];
+      int naccepted = 0;
+      int ret;
+
+#ifdef DEBUG
+      re_token_type_t type = dfa->nodes[prev_node].type;
+      assert (!IS_EPSILON_NODE (type));
+#endif
+#ifdef RE_ENABLE_I18N
+      /* If the node may accept `multi byte'.  */
+      if (dfa->nodes[prev_node].accept_mb)
+	naccepted = sift_states_iter_mb (mctx, sctx, prev_node,
+					 str_idx, sctx->last_str_idx);
+#endif /* RE_ENABLE_I18N */
+
+      /* We don't check backreferences here.
+	 See update_cur_sifted_state().  */
+      if (!naccepted
+	  && check_node_accept (mctx, dfa->nodes + prev_node, str_idx)
+	  && STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + 1],
+				  dfa->nexts[prev_node]))
+	naccepted = 1;
+
+      if (naccepted == 0)
+	continue;
+
+      if (sctx->limits.nelem)
+	{
+	  int to_idx = str_idx + naccepted;
+	  if (check_dst_limits (mctx, &sctx->limits,
+				dfa->nexts[prev_node], to_idx,
+				prev_node, str_idx))
+	    continue;
+	}
+      ret = re_node_set_insert (cur_dest, prev_node);
+      if (BE (ret == -1, 0))
+	return REG_ESPACE;
+    }
+
+  return REG_NOERROR;
+}
+
+/* Helper functions.  */
+
+static reg_errcode_t
+internal_function
+clean_state_log_if_needed (re_match_context_t *mctx, int next_state_log_idx)
+{
+  int top = mctx->state_log_top;
+
+  if (next_state_log_idx >= mctx->input.bufs_len
+      || (next_state_log_idx >= mctx->input.valid_len
+	  && mctx->input.valid_len < mctx->input.len))
+    {
+      reg_errcode_t err;
+      err = extend_buffers (mctx);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+    }
+
+  if (top < next_state_log_idx)
+    {
+      memset (mctx->state_log + top + 1, '\0',
+	      sizeof (re_dfastate_t *) * (next_state_log_idx - top));
+      mctx->state_log_top = next_state_log_idx;
+    }
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+merge_state_array (const re_dfa_t *dfa, re_dfastate_t **dst,
+		   re_dfastate_t **src, int num)
+{
+  int st_idx;
+  reg_errcode_t err;
+  for (st_idx = 0; st_idx < num; ++st_idx)
+    {
+      if (dst[st_idx] == NULL)
+	dst[st_idx] = src[st_idx];
+      else if (src[st_idx] != NULL)
+	{
+	  re_node_set merged_set;
+	  err = re_node_set_init_union (&merged_set, &dst[st_idx]->nodes,
+					&src[st_idx]->nodes);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	  dst[st_idx] = re_acquire_state (&err, dfa, &merged_set);
+	  re_node_set_free (&merged_set);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	}
+    }
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+update_cur_sifted_state (const re_match_context_t *mctx,
+			 re_sift_context_t *sctx, int str_idx,
+			 re_node_set *dest_nodes)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err = REG_NOERROR;
+  const re_node_set *candidates;
+  candidates = ((mctx->state_log[str_idx] == NULL) ? NULL
+		: &mctx->state_log[str_idx]->nodes);
+
+  if (dest_nodes->nelem == 0)
+    sctx->sifted_states[str_idx] = NULL;
+  else
+    {
+      if (candidates)
+	{
+	  /* At first, add the nodes which can epsilon transit to a node in
+	     DEST_NODE.  */
+	  err = add_epsilon_src_nodes (dfa, dest_nodes, candidates);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+
+	  /* Then, check the limitations in the current sift_context.  */
+	  if (sctx->limits.nelem)
+	    {
+	      err = check_subexp_limits (dfa, dest_nodes, candidates, &sctx->limits,
+					 mctx->bkref_ents, str_idx);
+	      if (BE (err != REG_NOERROR, 0))
+		return err;
+	    }
+	}
+
+      sctx->sifted_states[str_idx] = re_acquire_state (&err, dfa, dest_nodes);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+    }
+
+  if (candidates && mctx->state_log[str_idx]->has_backref)
+    {
+      err = sift_states_bkref (mctx, sctx, str_idx, candidates);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+    }
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+add_epsilon_src_nodes (const re_dfa_t *dfa, re_node_set *dest_nodes,
+		       const re_node_set *candidates)
+{
+  reg_errcode_t err = REG_NOERROR;
+  int i;
+
+  re_dfastate_t *state = re_acquire_state (&err, dfa, dest_nodes);
+  if (BE (err != REG_NOERROR, 0))
+    return err;
+
+  if (!state->inveclosure.alloc)
+    {
+      err = re_node_set_alloc (&state->inveclosure, dest_nodes->nelem);
+      if (BE (err != REG_NOERROR, 0))
+	return REG_ESPACE;
+      for (i = 0; i < dest_nodes->nelem; i++)
+	{
+	  err = re_node_set_merge (&state->inveclosure,
+				   dfa->inveclosures + dest_nodes->elems[i]);
+	  if (BE (err != REG_NOERROR, 0))
+	    return REG_ESPACE;
+	}
+    }
+  return re_node_set_add_intersect (dest_nodes, candidates,
+				    &state->inveclosure);
+}
+
+static reg_errcode_t
+internal_function
+sub_epsilon_src_nodes (const re_dfa_t *dfa, int node, re_node_set *dest_nodes,
+		       const re_node_set *candidates)
+{
+    int ecl_idx;
+    reg_errcode_t err;
+    re_node_set *inv_eclosure = dfa->inveclosures + node;
+    re_node_set except_nodes;
+    re_node_set_init_empty (&except_nodes);
+    for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
+      {
+	int cur_node = inv_eclosure->elems[ecl_idx];
+	if (cur_node == node)
+	  continue;
+	if (IS_EPSILON_NODE (dfa->nodes[cur_node].type))
+	  {
+	    int edst1 = dfa->edests[cur_node].elems[0];
+	    int edst2 = ((dfa->edests[cur_node].nelem > 1)
+			 ? dfa->edests[cur_node].elems[1] : -1);
+	    if ((!re_node_set_contains (inv_eclosure, edst1)
+		 && re_node_set_contains (dest_nodes, edst1))
+		|| (edst2 > 0
+		    && !re_node_set_contains (inv_eclosure, edst2)
+		    && re_node_set_contains (dest_nodes, edst2)))
+	      {
+		err = re_node_set_add_intersect (&except_nodes, candidates,
+						 dfa->inveclosures + cur_node);
+		if (BE (err != REG_NOERROR, 0))
+		  {
+		    re_node_set_free (&except_nodes);
+		    return err;
+		  }
+	      }
+	  }
+      }
+    for (ecl_idx = 0; ecl_idx < inv_eclosure->nelem; ++ecl_idx)
+      {
+	int cur_node = inv_eclosure->elems[ecl_idx];
+	if (!re_node_set_contains (&except_nodes, cur_node))
+	  {
+	    int idx = re_node_set_contains (dest_nodes, cur_node) - 1;
+	    re_node_set_remove_at (dest_nodes, idx);
+	  }
+      }
+    re_node_set_free (&except_nodes);
+    return REG_NOERROR;
+}
+
+static int
+internal_function
+check_dst_limits (const re_match_context_t *mctx, re_node_set *limits,
+		  int dst_node, int dst_idx, int src_node, int src_idx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int lim_idx, src_pos, dst_pos;
+
+  int dst_bkref_idx = search_cur_bkref_entry (mctx, dst_idx);
+  int src_bkref_idx = search_cur_bkref_entry (mctx, src_idx);
+  for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
+    {
+      int subexp_idx;
+      struct re_backref_cache_entry *ent;
+      ent = mctx->bkref_ents + limits->elems[lim_idx];
+      subexp_idx = dfa->nodes[ent->node].opr.idx;
+
+      dst_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
+					   subexp_idx, dst_node, dst_idx,
+					   dst_bkref_idx);
+      src_pos = check_dst_limits_calc_pos (mctx, limits->elems[lim_idx],
+					   subexp_idx, src_node, src_idx,
+					   src_bkref_idx);
+
+      /* In case of:
+	 <src> <dst> ( <subexp> )
+	 ( <subexp> ) <src> <dst>
+	 ( <subexp1> <src> <subexp2> <dst> <subexp3> )  */
+      if (src_pos == dst_pos)
+	continue; /* This is unrelated limitation.  */
+      else
+	return 1;
+    }
+  return 0;
+}
+
+static int
+internal_function
+check_dst_limits_calc_pos_1 (const re_match_context_t *mctx, int boundaries,
+			     int subexp_idx, int from_node, int bkref_idx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  const re_node_set *eclosures = dfa->eclosures + from_node;
+  int node_idx;
+
+  /* Else, we are on the boundary: examine the nodes on the epsilon
+     closure.  */
+  for (node_idx = 0; node_idx < eclosures->nelem; ++node_idx)
+    {
+      int node = eclosures->elems[node_idx];
+      switch (dfa->nodes[node].type)
+	{
+	case OP_BACK_REF:
+	  if (bkref_idx != -1)
+	    {
+	      struct re_backref_cache_entry *ent = mctx->bkref_ents + bkref_idx;
+	      do
+		{
+		  int dst, cpos;
+
+		  if (ent->node != node)
+		    continue;
+
+		  if (subexp_idx < BITSET_WORD_BITS
+		      && !(ent->eps_reachable_subexps_map
+			   & ((bitset_word_t) 1 << subexp_idx)))
+		    continue;
+
+		  /* Recurse trying to reach the OP_OPEN_SUBEXP and
+		     OP_CLOSE_SUBEXP cases below.  But, if the
+		     destination node is the same node as the source
+		     node, don't recurse because it would cause an
+		     infinite loop: a regex that exhibits this behavior
+		     is ()\1*\1*  */
+		  dst = dfa->edests[node].elems[0];
+		  if (dst == from_node)
+		    {
+		      if (boundaries & 1)
+			return -1;
+		      else /* if (boundaries & 2) */
+			return 0;
+		    }
+
+		  cpos =
+		    check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
+						 dst, bkref_idx);
+		  if (cpos == -1 /* && (boundaries & 1) */)
+		    return -1;
+		  if (cpos == 0 && (boundaries & 2))
+		    return 0;
+
+		  if (subexp_idx < BITSET_WORD_BITS)
+		    ent->eps_reachable_subexps_map
+		      &= ~((bitset_word_t) 1 << subexp_idx);
+		}
+	      while (ent++->more);
+	    }
+	  break;
+
+	case OP_OPEN_SUBEXP:
+	  if ((boundaries & 1) && subexp_idx == dfa->nodes[node].opr.idx)
+	    return -1;
+	  break;
+
+	case OP_CLOSE_SUBEXP:
+	  if ((boundaries & 2) && subexp_idx == dfa->nodes[node].opr.idx)
+	    return 0;
+	  break;
+
+	default:
+	    break;
+	}
+    }
+
+  return (boundaries & 2) ? 1 : 0;
+}
+
+static int
+internal_function
+check_dst_limits_calc_pos (const re_match_context_t *mctx, int limit,
+			   int subexp_idx, int from_node, int str_idx,
+			   int bkref_idx)
+{
+  struct re_backref_cache_entry *lim = mctx->bkref_ents + limit;
+  int boundaries;
+
+  /* If we are outside the range of the subexpression, return -1 or 1.  */
+  if (str_idx < lim->subexp_from)
+    return -1;
+
+  if (lim->subexp_to < str_idx)
+    return 1;
+
+  /* If we are within the subexpression, return 0.  */
+  boundaries = (str_idx == lim->subexp_from);
+  boundaries |= (str_idx == lim->subexp_to) << 1;
+  if (boundaries == 0)
+    return 0;
+
+  /* Else, examine epsilon closure.  */
+  return check_dst_limits_calc_pos_1 (mctx, boundaries, subexp_idx,
+				      from_node, bkref_idx);
+}
+
+/* Check the limitations of sub expressions LIMITS, and remove the nodes
+   which are against limitations from DEST_NODES. */
+
+static reg_errcode_t
+internal_function
+check_subexp_limits (const re_dfa_t *dfa, re_node_set *dest_nodes,
+		     const re_node_set *candidates, re_node_set *limits,
+		     struct re_backref_cache_entry *bkref_ents, int str_idx)
+{
+  reg_errcode_t err;
+  int node_idx, lim_idx;
+
+  for (lim_idx = 0; lim_idx < limits->nelem; ++lim_idx)
+    {
+      int subexp_idx;
+      struct re_backref_cache_entry *ent;
+      ent = bkref_ents + limits->elems[lim_idx];
+
+      if (str_idx <= ent->subexp_from || ent->str_idx < str_idx)
+	continue; /* This is unrelated limitation.  */
+
+      subexp_idx = dfa->nodes[ent->node].opr.idx;
+      if (ent->subexp_to == str_idx)
+	{
+	  int ops_node = -1;
+	  int cls_node = -1;
+	  for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
+	    {
+	      int node = dest_nodes->elems[node_idx];
+	      re_token_type_t type = dfa->nodes[node].type;
+	      if (type == OP_OPEN_SUBEXP
+		  && subexp_idx == dfa->nodes[node].opr.idx)
+		ops_node = node;
+	      else if (type == OP_CLOSE_SUBEXP
+		       && subexp_idx == dfa->nodes[node].opr.idx)
+		cls_node = node;
+	    }
+
+	  /* Check the limitation of the open subexpression.  */
+	  /* Note that (ent->subexp_to = str_idx != ent->subexp_from).  */
+	  if (ops_node >= 0)
+	    {
+	      err = sub_epsilon_src_nodes (dfa, ops_node, dest_nodes,
+					   candidates);
+	      if (BE (err != REG_NOERROR, 0))
+		return err;
+	    }
+
+	  /* Check the limitation of the close subexpression.  */
+	  if (cls_node >= 0)
+	    for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
+	      {
+		int node = dest_nodes->elems[node_idx];
+		if (!re_node_set_contains (dfa->inveclosures + node,
+					   cls_node)
+		    && !re_node_set_contains (dfa->eclosures + node,
+					      cls_node))
+		  {
+		    /* It is against this limitation.
+		       Remove it form the current sifted state.  */
+		    err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
+						 candidates);
+		    if (BE (err != REG_NOERROR, 0))
+		      return err;
+		    --node_idx;
+		  }
+	      }
+	}
+      else /* (ent->subexp_to != str_idx)  */
+	{
+	  for (node_idx = 0; node_idx < dest_nodes->nelem; ++node_idx)
+	    {
+	      int node = dest_nodes->elems[node_idx];
+	      re_token_type_t type = dfa->nodes[node].type;
+	      if (type == OP_CLOSE_SUBEXP || type == OP_OPEN_SUBEXP)
+		{
+		  if (subexp_idx != dfa->nodes[node].opr.idx)
+		    continue;
+		  /* It is against this limitation.
+		     Remove it form the current sifted state.  */
+		  err = sub_epsilon_src_nodes (dfa, node, dest_nodes,
+					       candidates);
+		  if (BE (err != REG_NOERROR, 0))
+		    return err;
+		}
+	    }
+	}
+    }
+  return REG_NOERROR;
+}
+
+static reg_errcode_t
+internal_function
+sift_states_bkref (const re_match_context_t *mctx, re_sift_context_t *sctx,
+		   int str_idx, const re_node_set *candidates)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err;
+  int node_idx, node;
+  re_sift_context_t local_sctx;
+  int first_idx = search_cur_bkref_entry (mctx, str_idx);
+
+  if (first_idx == -1)
+    return REG_NOERROR;
+
+  local_sctx.sifted_states = NULL; /* Mark that it hasn't been initialized.  */
+
+  for (node_idx = 0; node_idx < candidates->nelem; ++node_idx)
+    {
+      int enabled_idx;
+      re_token_type_t type;
+      struct re_backref_cache_entry *entry;
+      node = candidates->elems[node_idx];
+      type = dfa->nodes[node].type;
+      /* Avoid infinite loop for the REs like "()\1+".  */
+      if (node == sctx->last_node && str_idx == sctx->last_str_idx)
+	continue;
+      if (type != OP_BACK_REF)
+	continue;
+
+      entry = mctx->bkref_ents + first_idx;
+      enabled_idx = first_idx;
+      do
+	{
+	  int subexp_len;
+	  int to_idx;
+	  int dst_node;
+	  int ret;
+	  re_dfastate_t *cur_state;
+
+	  if (entry->node != node)
+	    continue;
+	  subexp_len = entry->subexp_to - entry->subexp_from;
+	  to_idx = str_idx + subexp_len;
+	  dst_node = (subexp_len ? dfa->nexts[node]
+		      : dfa->edests[node].elems[0]);
+
+	  if (to_idx > sctx->last_str_idx
+	      || sctx->sifted_states[to_idx] == NULL
+	      || !STATE_NODE_CONTAINS (sctx->sifted_states[to_idx], dst_node)
+	      || check_dst_limits (mctx, &sctx->limits, node,
+				   str_idx, dst_node, to_idx))
+	    continue;
+
+	  if (local_sctx.sifted_states == NULL)
+	    {
+	      local_sctx = *sctx;
+	      err = re_node_set_init_copy (&local_sctx.limits, &sctx->limits);
+	      if (BE (err != REG_NOERROR, 0))
+		goto free_return;
+	    }
+	  local_sctx.last_node = node;
+	  local_sctx.last_str_idx = str_idx;
+	  ret = re_node_set_insert (&local_sctx.limits, enabled_idx);
+	  if (BE (ret < 0, 0))
+	    {
+	      err = REG_ESPACE;
+	      goto free_return;
+	    }
+	  cur_state = local_sctx.sifted_states[str_idx];
+	  err = sift_states_backward (mctx, &local_sctx);
+	  if (BE (err != REG_NOERROR, 0))
+	    goto free_return;
+	  if (sctx->limited_states != NULL)
+	    {
+	      err = merge_state_array (dfa, sctx->limited_states,
+				       local_sctx.sifted_states,
+				       str_idx + 1);
+	      if (BE (err != REG_NOERROR, 0))
+		goto free_return;
+	    }
+	  local_sctx.sifted_states[str_idx] = cur_state;
+	  re_node_set_remove (&local_sctx.limits, enabled_idx);
+
+	  /* mctx->bkref_ents may have changed, reload the pointer.  */
+	  entry = mctx->bkref_ents + enabled_idx;
+	}
+      while (enabled_idx++, entry++->more);
+    }
+  err = REG_NOERROR;
+ free_return:
+  if (local_sctx.sifted_states != NULL)
+    {
+      re_node_set_free (&local_sctx.limits);
+    }
+
+  return err;
+}
+
+
+#ifdef RE_ENABLE_I18N
+static int
+internal_function
+sift_states_iter_mb (const re_match_context_t *mctx, re_sift_context_t *sctx,
+		     int node_idx, int str_idx, int max_str_idx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int naccepted;
+  /* Check the node can accept `multi byte'.  */
+  naccepted = check_node_accept_bytes (dfa, node_idx, &mctx->input, str_idx);
+  if (naccepted > 0 && str_idx + naccepted <= max_str_idx &&
+      !STATE_NODE_CONTAINS (sctx->sifted_states[str_idx + naccepted],
+			    dfa->nexts[node_idx]))
+    /* The node can't accept the `multi byte', or the
+       destination was already thrown away, then the node
+       couldn't accept the current input `multi byte'.   */
+    naccepted = 0;
+  /* Otherwise, it is sure that the node could accept
+     `naccepted' bytes input.  */
+  return naccepted;
+}
+#endif /* RE_ENABLE_I18N */
+
+
+/* Functions for state transition.  */
+
+/* Return the next state to which the current state STATE will transit by
+   accepting the current input byte, and update STATE_LOG if necessary.
+   If STATE can accept a multibyte char/collating element/back reference
+   update the destination of STATE_LOG.  */
+
+static re_dfastate_t *
+internal_function
+transit_state (reg_errcode_t *err, re_match_context_t *mctx,
+	       re_dfastate_t *state)
+{
+  re_dfastate_t **trtable;
+  unsigned char ch;
+
+#ifdef RE_ENABLE_I18N
+  /* If the current state can accept multibyte.  */
+  if (BE (state->accept_mb, 0))
+    {
+      *err = transit_state_mb (mctx, state);
+      if (BE (*err != REG_NOERROR, 0))
+	return NULL;
+    }
+#endif /* RE_ENABLE_I18N */
+
+  /* Then decide the next state with the single byte.  */
+#if 0
+  if (0)
+    /* don't use transition table  */
+    return transit_state_sb (err, mctx, state);
+#endif
+
+  /* Use transition table  */
+  ch = re_string_fetch_byte (&mctx->input);
+  for (;;)
+    {
+      trtable = state->trtable;
+      if (BE (trtable != NULL, 1))
+	return trtable[ch];
+
+      trtable = state->word_trtable;
+      if (BE (trtable != NULL, 1))
+	{
+	  unsigned int context;
+	  context
+	    = re_string_context_at (&mctx->input,
+				    re_string_cur_idx (&mctx->input) - 1,
+				    mctx->eflags);
+	  if (IS_WORD_CONTEXT (context))
+	    return trtable[ch + SBC_MAX];
+	  else
+	    return trtable[ch];
+	}
+
+      if (!build_trtable (mctx->dfa, state))
+	{
+	  *err = REG_ESPACE;
+	  return NULL;
+	}
+
+      /* Retry, we now have a transition table.  */
+    }
+}
+
+/* Update the state_log if we need */
+static re_dfastate_t *
+internal_function
+merge_state_with_log (reg_errcode_t *err, re_match_context_t *mctx,
+		      re_dfastate_t *next_state)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int cur_idx = re_string_cur_idx (&mctx->input);
+
+  if (cur_idx > mctx->state_log_top)
+    {
+      mctx->state_log[cur_idx] = next_state;
+      mctx->state_log_top = cur_idx;
+    }
+  else if (mctx->state_log[cur_idx] == NULL)
+    {
+      mctx->state_log[cur_idx] = next_state;
+    }
+  else
+    {
+      re_dfastate_t *pstate;
+      unsigned int context;
+      re_node_set next_nodes, *log_nodes, *table_nodes = NULL;
+      /* If (state_log[cur_idx] != 0), it implies that cur_idx is
+	 the destination of a multibyte char/collating element/
+	 back reference.  Then the next state is the union set of
+	 these destinations and the results of the transition table.  */
+      pstate = mctx->state_log[cur_idx];
+      log_nodes = pstate->entrance_nodes;
+      if (next_state != NULL)
+	{
+	  table_nodes = next_state->entrance_nodes;
+	  *err = re_node_set_init_union (&next_nodes, table_nodes,
+					     log_nodes);
+	  if (BE (*err != REG_NOERROR, 0))
+	    return NULL;
+	}
+      else
+	next_nodes = *log_nodes;
+      /* Note: We already add the nodes of the initial state,
+	 then we don't need to add them here.  */
+
+      context = re_string_context_at (&mctx->input,
+				      re_string_cur_idx (&mctx->input) - 1,
+				      mctx->eflags);
+      next_state = mctx->state_log[cur_idx]
+	= re_acquire_state_context (err, dfa, &next_nodes, context);
+      /* We don't need to check errors here, since the return value of
+	 this function is next_state and ERR is already set.  */
+
+      if (table_nodes != NULL)
+	re_node_set_free (&next_nodes);
+    }
+
+  if (BE (dfa->nbackref, 0) && next_state != NULL)
+    {
+      /* Check OP_OPEN_SUBEXP in the current state in case that we use them
+	 later.  We must check them here, since the back references in the
+	 next state might use them.  */
+      *err = check_subexp_matching_top (mctx, &next_state->nodes,
+					cur_idx);
+      if (BE (*err != REG_NOERROR, 0))
+	return NULL;
+
+      /* If the next state has back references.  */
+      if (next_state->has_backref)
+	{
+	  *err = transit_state_bkref (mctx, &next_state->nodes);
+	  if (BE (*err != REG_NOERROR, 0))
+	    return NULL;
+	  next_state = mctx->state_log[cur_idx];
+	}
+    }
+
+  return next_state;
+}
+
+/* Skip bytes in the input that correspond to part of a
+   multi-byte match, then look in the log for a state
+   from which to restart matching.  */
+static re_dfastate_t *
+internal_function
+find_recover_state (reg_errcode_t *err, re_match_context_t *mctx)
+{
+  re_dfastate_t *cur_state;
+  do
+    {
+      int max = mctx->state_log_top;
+      int cur_str_idx = re_string_cur_idx (&mctx->input);
+
+      do
+	{
+	  if (++cur_str_idx > max)
+	    return NULL;
+	  re_string_skip_bytes (&mctx->input, 1);
+	}
+      while (mctx->state_log[cur_str_idx] == NULL);
+
+      cur_state = merge_state_with_log (err, mctx, NULL);
+    }
+  while (*err == REG_NOERROR && cur_state == NULL);
+  return cur_state;
+}
+
+/* Helper functions for transit_state.  */
+
+/* From the node set CUR_NODES, pick up the nodes whose types are
+   OP_OPEN_SUBEXP and which have corresponding back references in the regular
+   expression. And register them to use them later for evaluating the
+   correspoding back references.  */
+
+static reg_errcode_t
+internal_function
+check_subexp_matching_top (re_match_context_t *mctx, re_node_set *cur_nodes,
+			   int str_idx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int node_idx;
+  reg_errcode_t err;
+
+  /* TODO: This isn't efficient.
+	   Because there might be more than one nodes whose types are
+	   OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
+	   nodes.
+	   E.g. RE: (a){2}  */
+  for (node_idx = 0; node_idx < cur_nodes->nelem; ++node_idx)
+    {
+      int node = cur_nodes->elems[node_idx];
+      if (dfa->nodes[node].type == OP_OPEN_SUBEXP
+	  && dfa->nodes[node].opr.idx < BITSET_WORD_BITS
+	  && (dfa->used_bkref_map
+	      & ((bitset_word_t) 1 << dfa->nodes[node].opr.idx)))
+	{
+	  err = match_ctx_add_subtop (mctx, node, str_idx);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	}
+    }
+  return REG_NOERROR;
+}
+
+#if 0
+/* Return the next state to which the current state STATE will transit by
+   accepting the current input byte.  */
+
+static re_dfastate_t *
+transit_state_sb (reg_errcode_t *err, re_match_context_t *mctx,
+		  re_dfastate_t *state)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  re_node_set next_nodes;
+  re_dfastate_t *next_state;
+  int node_cnt, cur_str_idx = re_string_cur_idx (&mctx->input);
+  unsigned int context;
+
+  *err = re_node_set_alloc (&next_nodes, state->nodes.nelem + 1);
+  if (BE (*err != REG_NOERROR, 0))
+    return NULL;
+  for (node_cnt = 0; node_cnt < state->nodes.nelem; ++node_cnt)
+    {
+      int cur_node = state->nodes.elems[node_cnt];
+      if (check_node_accept (mctx, dfa->nodes + cur_node, cur_str_idx))
+	{
+	  *err = re_node_set_merge (&next_nodes,
+				    dfa->eclosures + dfa->nexts[cur_node]);
+	  if (BE (*err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&next_nodes);
+	      return NULL;
+	    }
+	}
+    }
+  context = re_string_context_at (&mctx->input, cur_str_idx, mctx->eflags);
+  next_state = re_acquire_state_context (err, dfa, &next_nodes, context);
+  /* We don't need to check errors here, since the return value of
+     this function is next_state and ERR is already set.  */
+
+  re_node_set_free (&next_nodes);
+  re_string_skip_bytes (&mctx->input, 1);
+  return next_state;
+}
+#endif
+
+#ifdef RE_ENABLE_I18N
+static reg_errcode_t
+internal_function
+transit_state_mb (re_match_context_t *mctx, re_dfastate_t *pstate)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err;
+  int i;
+
+  for (i = 0; i < pstate->nodes.nelem; ++i)
+    {
+      re_node_set dest_nodes, *new_nodes;
+      int cur_node_idx = pstate->nodes.elems[i];
+      int naccepted, dest_idx;
+      unsigned int context;
+      re_dfastate_t *dest_state;
+
+      if (!dfa->nodes[cur_node_idx].accept_mb)
+	continue;
+
+      if (dfa->nodes[cur_node_idx].constraint)
+	{
+	  context = re_string_context_at (&mctx->input,
+					  re_string_cur_idx (&mctx->input),
+					  mctx->eflags);
+	  if (NOT_SATISFY_NEXT_CONSTRAINT (dfa->nodes[cur_node_idx].constraint,
+					   context))
+	    continue;
+	}
+
+      /* How many bytes the node can accept?  */
+      naccepted = check_node_accept_bytes (dfa, cur_node_idx, &mctx->input,
+					   re_string_cur_idx (&mctx->input));
+      if (naccepted == 0)
+	continue;
+
+      /* The node can accepts `naccepted' bytes.  */
+      dest_idx = re_string_cur_idx (&mctx->input) + naccepted;
+      mctx->max_mb_elem_len = ((mctx->max_mb_elem_len < naccepted) ? naccepted
+			       : mctx->max_mb_elem_len);
+      err = clean_state_log_if_needed (mctx, dest_idx);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+#ifdef DEBUG
+      assert (dfa->nexts[cur_node_idx] != -1);
+#endif
+      new_nodes = dfa->eclosures + dfa->nexts[cur_node_idx];
+
+      dest_state = mctx->state_log[dest_idx];
+      if (dest_state == NULL)
+	dest_nodes = *new_nodes;
+      else
+	{
+	  err = re_node_set_init_union (&dest_nodes,
+					dest_state->entrance_nodes, new_nodes);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	}
+      context = re_string_context_at (&mctx->input, dest_idx - 1,
+				      mctx->eflags);
+      mctx->state_log[dest_idx]
+	= re_acquire_state_context (&err, dfa, &dest_nodes, context);
+      if (dest_state != NULL)
+	re_node_set_free (&dest_nodes);
+      if (BE (mctx->state_log[dest_idx] == NULL && err != REG_NOERROR, 0))
+	return err;
+    }
+  return REG_NOERROR;
+}
+#endif /* RE_ENABLE_I18N */
+
+static reg_errcode_t
+internal_function
+transit_state_bkref (re_match_context_t *mctx, const re_node_set *nodes)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err;
+  int i;
+  int cur_str_idx = re_string_cur_idx (&mctx->input);
+
+  for (i = 0; i < nodes->nelem; ++i)
+    {
+      int dest_str_idx, prev_nelem, bkc_idx;
+      int node_idx = nodes->elems[i];
+      unsigned int context;
+      const re_token_t *node = dfa->nodes + node_idx;
+      re_node_set *new_dest_nodes;
+
+      /* Check whether `node' is a backreference or not.  */
+      if (node->type != OP_BACK_REF)
+	continue;
+
+      if (node->constraint)
+	{
+	  context = re_string_context_at (&mctx->input, cur_str_idx,
+					  mctx->eflags);
+	  if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
+	    continue;
+	}
+
+      /* `node' is a backreference.
+	 Check the substring which the substring matched.  */
+      bkc_idx = mctx->nbkref_ents;
+      err = get_subexp (mctx, node_idx, cur_str_idx);
+      if (BE (err != REG_NOERROR, 0))
+	goto free_return;
+
+      /* And add the epsilon closures (which is `new_dest_nodes') of
+	 the backreference to appropriate state_log.  */
+#ifdef DEBUG
+      assert (dfa->nexts[node_idx] != -1);
+#endif
+      for (; bkc_idx < mctx->nbkref_ents; ++bkc_idx)
+	{
+	  int subexp_len;
+	  re_dfastate_t *dest_state;
+	  struct re_backref_cache_entry *bkref_ent;
+	  bkref_ent = mctx->bkref_ents + bkc_idx;
+	  if (bkref_ent->node != node_idx || bkref_ent->str_idx != cur_str_idx)
+	    continue;
+	  subexp_len = bkref_ent->subexp_to - bkref_ent->subexp_from;
+	  new_dest_nodes = (subexp_len == 0
+			    ? dfa->eclosures + dfa->edests[node_idx].elems[0]
+			    : dfa->eclosures + dfa->nexts[node_idx]);
+	  dest_str_idx = (cur_str_idx + bkref_ent->subexp_to
+			  - bkref_ent->subexp_from);
+	  context = re_string_context_at (&mctx->input, dest_str_idx - 1,
+					  mctx->eflags);
+	  dest_state = mctx->state_log[dest_str_idx];
+	  prev_nelem = ((mctx->state_log[cur_str_idx] == NULL) ? 0
+			: mctx->state_log[cur_str_idx]->nodes.nelem);
+	  /* Add `new_dest_node' to state_log.  */
+	  if (dest_state == NULL)
+	    {
+	      mctx->state_log[dest_str_idx]
+		= re_acquire_state_context (&err, dfa, new_dest_nodes,
+					    context);
+	      if (BE (mctx->state_log[dest_str_idx] == NULL
+		      && err != REG_NOERROR, 0))
+		goto free_return;
+	    }
+	  else
+	    {
+	      re_node_set dest_nodes;
+	      err = re_node_set_init_union (&dest_nodes,
+					    dest_state->entrance_nodes,
+					    new_dest_nodes);
+	      if (BE (err != REG_NOERROR, 0))
+		{
+		  re_node_set_free (&dest_nodes);
+		  goto free_return;
+		}
+	      mctx->state_log[dest_str_idx]
+		= re_acquire_state_context (&err, dfa, &dest_nodes, context);
+	      re_node_set_free (&dest_nodes);
+	      if (BE (mctx->state_log[dest_str_idx] == NULL
+		      && err != REG_NOERROR, 0))
+		goto free_return;
+	    }
+	  /* We need to check recursively if the backreference can epsilon
+	     transit.  */
+	  if (subexp_len == 0
+	      && mctx->state_log[cur_str_idx]->nodes.nelem > prev_nelem)
+	    {
+	      err = check_subexp_matching_top (mctx, new_dest_nodes,
+					       cur_str_idx);
+	      if (BE (err != REG_NOERROR, 0))
+		goto free_return;
+	      err = transit_state_bkref (mctx, new_dest_nodes);
+	      if (BE (err != REG_NOERROR, 0))
+		goto free_return;
+	    }
+	}
+    }
+  err = REG_NOERROR;
+ free_return:
+  return err;
+}
+
+/* Enumerate all the candidates which the backreference BKREF_NODE can match
+   at BKREF_STR_IDX, and register them by match_ctx_add_entry().
+   Note that we might collect inappropriate candidates here.
+   However, the cost of checking them strictly here is too high, then we
+   delay these checking for prune_impossible_nodes().  */
+
+static reg_errcode_t
+internal_function
+get_subexp (re_match_context_t *mctx, int bkref_node, int bkref_str_idx)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int subexp_num, sub_top_idx;
+  const char *buf = (const char *) re_string_get_buffer (&mctx->input);
+  /* Return if we have already checked BKREF_NODE at BKREF_STR_IDX.  */
+  int cache_idx = search_cur_bkref_entry (mctx, bkref_str_idx);
+  if (cache_idx != -1)
+    {
+      const struct re_backref_cache_entry *entry
+	= mctx->bkref_ents + cache_idx;
+      do
+	if (entry->node == bkref_node)
+	  return REG_NOERROR; /* We already checked it.  */
+      while (entry++->more);
+    }
+
+  subexp_num = dfa->nodes[bkref_node].opr.idx;
+
+  /* For each sub expression  */
+  for (sub_top_idx = 0; sub_top_idx < mctx->nsub_tops; ++sub_top_idx)
+    {
+      reg_errcode_t err;
+      re_sub_match_top_t *sub_top = mctx->sub_tops[sub_top_idx];
+      re_sub_match_last_t *sub_last;
+      int sub_last_idx, sl_str, bkref_str_off;
+
+      if (dfa->nodes[sub_top->node].opr.idx != subexp_num)
+	continue; /* It isn't related.  */
+
+      sl_str = sub_top->str_idx;
+      bkref_str_off = bkref_str_idx;
+      /* At first, check the last node of sub expressions we already
+	 evaluated.  */
+      for (sub_last_idx = 0; sub_last_idx < sub_top->nlasts; ++sub_last_idx)
+	{
+	  int sl_str_diff;
+	  sub_last = sub_top->lasts[sub_last_idx];
+	  sl_str_diff = sub_last->str_idx - sl_str;
+	  /* The matched string by the sub expression match with the substring
+	     at the back reference?  */
+	  if (sl_str_diff > 0)
+	    {
+	      if (BE (bkref_str_off + sl_str_diff > mctx->input.valid_len, 0))
+		{
+		  /* Not enough chars for a successful match.  */
+		  if (bkref_str_off + sl_str_diff > mctx->input.len)
+		    break;
+
+		  err = clean_state_log_if_needed (mctx,
+						   bkref_str_off
+						   + sl_str_diff);
+		  if (BE (err != REG_NOERROR, 0))
+		    return err;
+		  buf = (const char *) re_string_get_buffer (&mctx->input);
+		}
+	      if (memcmp (buf + bkref_str_off, buf + sl_str, sl_str_diff) != 0)
+		/* We don't need to search this sub expression any more.  */
+		break;
+	    }
+	  bkref_str_off += sl_str_diff;
+	  sl_str += sl_str_diff;
+	  err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
+				bkref_str_idx);
+
+	  /* Reload buf, since the preceding call might have reallocated
+	     the buffer.  */
+	  buf = (const char *) re_string_get_buffer (&mctx->input);
+
+	  if (err == REG_NOMATCH)
+	    continue;
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	}
+
+      if (sub_last_idx < sub_top->nlasts)
+	continue;
+      if (sub_last_idx > 0)
+	++sl_str;
+      /* Then, search for the other last nodes of the sub expression.  */
+      for (; sl_str <= bkref_str_idx; ++sl_str)
+	{
+	  int cls_node, sl_str_off;
+	  const re_node_set *nodes;
+	  sl_str_off = sl_str - sub_top->str_idx;
+	  /* The matched string by the sub expression match with the substring
+	     at the back reference?  */
+	  if (sl_str_off > 0)
+	    {
+	      if (BE (bkref_str_off >= mctx->input.valid_len, 0))
+		{
+		  /* If we are at the end of the input, we cannot match.  */
+		  if (bkref_str_off >= mctx->input.len)
+		    break;
+
+		  err = extend_buffers (mctx);
+		  if (BE (err != REG_NOERROR, 0))
+		    return err;
+
+		  buf = (const char *) re_string_get_buffer (&mctx->input);
+		}
+	      if (buf [bkref_str_off++] != buf[sl_str - 1])
+		break; /* We don't need to search this sub expression
+			  any more.  */
+	    }
+	  if (mctx->state_log[sl_str] == NULL)
+	    continue;
+	  /* Does this state have a ')' of the sub expression?  */
+	  nodes = &mctx->state_log[sl_str]->nodes;
+	  cls_node = find_subexp_node (dfa, nodes, subexp_num,
+				       OP_CLOSE_SUBEXP);
+	  if (cls_node == -1)
+	    continue; /* No.  */
+	  if (sub_top->path == NULL)
+	    {
+	      sub_top->path = calloc (sizeof (state_array_t),
+				      sl_str - sub_top->str_idx + 1);
+	      if (sub_top->path == NULL)
+		return REG_ESPACE;
+	    }
+	  /* Can the OP_OPEN_SUBEXP node arrive the OP_CLOSE_SUBEXP node
+	     in the current context?  */
+	  err = check_arrival (mctx, sub_top->path, sub_top->node,
+			       sub_top->str_idx, cls_node, sl_str,
+			       OP_CLOSE_SUBEXP);
+	  if (err == REG_NOMATCH)
+	      continue;
+	  if (BE (err != REG_NOERROR, 0))
+	      return err;
+	  sub_last = match_ctx_add_sublast (sub_top, cls_node, sl_str);
+	  if (BE (sub_last == NULL, 0))
+	    return REG_ESPACE;
+	  err = get_subexp_sub (mctx, sub_top, sub_last, bkref_node,
+				bkref_str_idx);
+	  if (err == REG_NOMATCH)
+	    continue;
+	}
+    }
+  return REG_NOERROR;
+}
+
+/* Helper functions for get_subexp().  */
+
+/* Check SUB_LAST can arrive to the back reference BKREF_NODE at BKREF_STR.
+   If it can arrive, register the sub expression expressed with SUB_TOP
+   and SUB_LAST.  */
+
+static reg_errcode_t
+internal_function
+get_subexp_sub (re_match_context_t *mctx, const re_sub_match_top_t *sub_top,
+		re_sub_match_last_t *sub_last, int bkref_node, int bkref_str)
+{
+  reg_errcode_t err;
+  int to_idx;
+  /* Can the subexpression arrive the back reference?  */
+  err = check_arrival (mctx, &sub_last->path, sub_last->node,
+		       sub_last->str_idx, bkref_node, bkref_str,
+		       OP_OPEN_SUBEXP);
+  if (err != REG_NOERROR)
+    return err;
+  err = match_ctx_add_entry (mctx, bkref_node, bkref_str, sub_top->str_idx,
+			     sub_last->str_idx);
+  if (BE (err != REG_NOERROR, 0))
+    return err;
+  to_idx = bkref_str + sub_last->str_idx - sub_top->str_idx;
+  return clean_state_log_if_needed (mctx, to_idx);
+}
+
+/* Find the first node which is '(' or ')' and whose index is SUBEXP_IDX.
+   Search '(' if FL_OPEN, or search ')' otherwise.
+   TODO: This function isn't efficient...
+	 Because there might be more than one nodes whose types are
+	 OP_OPEN_SUBEXP and whose index is SUBEXP_IDX, we must check all
+	 nodes.
+	 E.g. RE: (a){2}  */
+
+static int
+internal_function
+find_subexp_node (const re_dfa_t *dfa, const re_node_set *nodes,
+		  int subexp_idx, int type)
+{
+  int cls_idx;
+  for (cls_idx = 0; cls_idx < nodes->nelem; ++cls_idx)
+    {
+      int cls_node = nodes->elems[cls_idx];
+      const re_token_t *node = dfa->nodes + cls_node;
+      if (node->type == type
+	  && node->opr.idx == subexp_idx)
+	return cls_node;
+    }
+  return -1;
+}
+
+/* Check whether the node TOP_NODE at TOP_STR can arrive to the node
+   LAST_NODE at LAST_STR.  We record the path onto PATH since it will be
+   heavily reused.
+   Return REG_NOERROR if it can arrive, or REG_NOMATCH otherwise.  */
+
+static reg_errcode_t
+internal_function
+check_arrival (re_match_context_t *mctx, state_array_t *path, int top_node,
+	       int top_str, int last_node, int last_str, int type)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err = REG_NOERROR;
+  int subexp_num, backup_cur_idx, str_idx, null_cnt;
+  re_dfastate_t *cur_state = NULL;
+  re_node_set *cur_nodes, next_nodes;
+  re_dfastate_t **backup_state_log;
+  unsigned int context;
+
+  subexp_num = dfa->nodes[top_node].opr.idx;
+  /* Extend the buffer if we need.  */
+  if (BE (path->alloc < last_str + mctx->max_mb_elem_len + 1, 0))
+    {
+      re_dfastate_t **new_array;
+      int old_alloc = path->alloc;
+      path->alloc += last_str + mctx->max_mb_elem_len + 1;
+      new_array = re_realloc (path->array, re_dfastate_t *, path->alloc);
+      if (BE (new_array == NULL, 0))
+	{
+	  path->alloc = old_alloc;
+	  return REG_ESPACE;
+	}
+      path->array = new_array;
+      memset (new_array + old_alloc, '\0',
+	      sizeof (re_dfastate_t *) * (path->alloc - old_alloc));
+    }
+
+  str_idx = path->next_idx ? path->next_idx : top_str;
+
+  /* Temporary modify MCTX.  */
+  backup_state_log = mctx->state_log;
+  backup_cur_idx = mctx->input.cur_idx;
+  mctx->state_log = path->array;
+  mctx->input.cur_idx = str_idx;
+
+  /* Setup initial node set.  */
+  context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
+  if (str_idx == top_str)
+    {
+      err = re_node_set_init_1 (&next_nodes, top_node);
+      if (BE (err != REG_NOERROR, 0))
+	return err;
+      err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
+      if (BE (err != REG_NOERROR, 0))
+	{
+	  re_node_set_free (&next_nodes);
+	  return err;
+	}
+    }
+  else
+    {
+      cur_state = mctx->state_log[str_idx];
+      if (cur_state && cur_state->has_backref)
+	{
+	  err = re_node_set_init_copy (&next_nodes, &cur_state->nodes);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	}
+      else
+	re_node_set_init_empty (&next_nodes);
+    }
+  if (str_idx == top_str || (cur_state && cur_state->has_backref))
+    {
+      if (next_nodes.nelem)
+	{
+	  err = expand_bkref_cache (mctx, &next_nodes, str_idx,
+				    subexp_num, type);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&next_nodes);
+	      return err;
+	    }
+	}
+      cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
+      if (BE (cur_state == NULL && err != REG_NOERROR, 0))
+	{
+	  re_node_set_free (&next_nodes);
+	  return err;
+	}
+      mctx->state_log[str_idx] = cur_state;
+    }
+
+  for (null_cnt = 0; str_idx < last_str && null_cnt <= mctx->max_mb_elem_len;)
+    {
+      re_node_set_empty (&next_nodes);
+      if (mctx->state_log[str_idx + 1])
+	{
+	  err = re_node_set_merge (&next_nodes,
+				   &mctx->state_log[str_idx + 1]->nodes);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&next_nodes);
+	      return err;
+	    }
+	}
+      if (cur_state)
+	{
+	  err = check_arrival_add_next_nodes (mctx, str_idx,
+					      &cur_state->non_eps_nodes,
+					      &next_nodes);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&next_nodes);
+	      return err;
+	    }
+	}
+      ++str_idx;
+      if (next_nodes.nelem)
+	{
+	  err = check_arrival_expand_ecl (dfa, &next_nodes, subexp_num, type);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&next_nodes);
+	      return err;
+	    }
+	  err = expand_bkref_cache (mctx, &next_nodes, str_idx,
+				    subexp_num, type);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&next_nodes);
+	      return err;
+	    }
+	}
+      context = re_string_context_at (&mctx->input, str_idx - 1, mctx->eflags);
+      cur_state = re_acquire_state_context (&err, dfa, &next_nodes, context);
+      if (BE (cur_state == NULL && err != REG_NOERROR, 0))
+	{
+	  re_node_set_free (&next_nodes);
+	  return err;
+	}
+      mctx->state_log[str_idx] = cur_state;
+      null_cnt = cur_state == NULL ? null_cnt + 1 : 0;
+    }
+  re_node_set_free (&next_nodes);
+  cur_nodes = (mctx->state_log[last_str] == NULL ? NULL
+	       : &mctx->state_log[last_str]->nodes);
+  path->next_idx = str_idx;
+
+  /* Fix MCTX.  */
+  mctx->state_log = backup_state_log;
+  mctx->input.cur_idx = backup_cur_idx;
+
+  /* Then check the current node set has the node LAST_NODE.  */
+  if (cur_nodes != NULL && re_node_set_contains (cur_nodes, last_node))
+    return REG_NOERROR;
+
+  return REG_NOMATCH;
+}
+
+/* Helper functions for check_arrival.  */
+
+/* Calculate the destination nodes of CUR_NODES at STR_IDX, and append them
+   to NEXT_NODES.
+   TODO: This function is similar to the functions transit_state*(),
+	 however this function has many additional works.
+	 Can't we unify them?  */
+
+static reg_errcode_t
+internal_function
+check_arrival_add_next_nodes (re_match_context_t *mctx, int str_idx,
+			      re_node_set *cur_nodes, re_node_set *next_nodes)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  int result;
+  int cur_idx;
+#ifdef RE_ENABLE_I18N
+  reg_errcode_t err = REG_NOERROR;
+#endif
+  re_node_set union_set;
+  re_node_set_init_empty (&union_set);
+  for (cur_idx = 0; cur_idx < cur_nodes->nelem; ++cur_idx)
+    {
+      int naccepted = 0;
+      int cur_node = cur_nodes->elems[cur_idx];
+#ifdef DEBUG
+      re_token_type_t type = dfa->nodes[cur_node].type;
+      assert (!IS_EPSILON_NODE (type));
+#endif
+#ifdef RE_ENABLE_I18N
+      /* If the node may accept `multi byte'.  */
+      if (dfa->nodes[cur_node].accept_mb)
+	{
+	  naccepted = check_node_accept_bytes (dfa, cur_node, &mctx->input,
+					       str_idx);
+	  if (naccepted > 1)
+	    {
+	      re_dfastate_t *dest_state;
+	      int next_node = dfa->nexts[cur_node];
+	      int next_idx = str_idx + naccepted;
+	      dest_state = mctx->state_log[next_idx];
+	      re_node_set_empty (&union_set);
+	      if (dest_state)
+		{
+		  err = re_node_set_merge (&union_set, &dest_state->nodes);
+		  if (BE (err != REG_NOERROR, 0))
+		    {
+		      re_node_set_free (&union_set);
+		      return err;
+		    }
+		}
+	      result = re_node_set_insert (&union_set, next_node);
+	      if (BE (result < 0, 0))
+		{
+		  re_node_set_free (&union_set);
+		  return REG_ESPACE;
+		}
+	      mctx->state_log[next_idx] = re_acquire_state (&err, dfa,
+							    &union_set);
+	      if (BE (mctx->state_log[next_idx] == NULL
+		      && err != REG_NOERROR, 0))
+		{
+		  re_node_set_free (&union_set);
+		  return err;
+		}
+	    }
+	}
+#endif /* RE_ENABLE_I18N */
+      if (naccepted
+	  || check_node_accept (mctx, dfa->nodes + cur_node, str_idx))
+	{
+	  result = re_node_set_insert (next_nodes, dfa->nexts[cur_node]);
+	  if (BE (result < 0, 0))
+	    {
+	      re_node_set_free (&union_set);
+	      return REG_ESPACE;
+	    }
+	}
+    }
+  re_node_set_free (&union_set);
+  return REG_NOERROR;
+}
+
+/* For all the nodes in CUR_NODES, add the epsilon closures of them to
+   CUR_NODES, however exclude the nodes which are:
+    - inside the sub expression whose number is EX_SUBEXP, if FL_OPEN.
+    - out of the sub expression whose number is EX_SUBEXP, if !FL_OPEN.
+*/
+
+static reg_errcode_t
+internal_function
+check_arrival_expand_ecl (const re_dfa_t *dfa, re_node_set *cur_nodes,
+			  int ex_subexp, int type)
+{
+  reg_errcode_t err;
+  int idx, outside_node;
+  re_node_set new_nodes;
+#ifdef DEBUG
+  assert (cur_nodes->nelem);
+#endif
+  err = re_node_set_alloc (&new_nodes, cur_nodes->nelem);
+  if (BE (err != REG_NOERROR, 0))
+    return err;
+  /* Create a new node set NEW_NODES with the nodes which are epsilon
+     closures of the node in CUR_NODES.  */
+
+  for (idx = 0; idx < cur_nodes->nelem; ++idx)
+    {
+      int cur_node = cur_nodes->elems[idx];
+      const re_node_set *eclosure = dfa->eclosures + cur_node;
+      outside_node = find_subexp_node (dfa, eclosure, ex_subexp, type);
+      if (outside_node == -1)
+	{
+	  /* There are no problematic nodes, just merge them.  */
+	  err = re_node_set_merge (&new_nodes, eclosure);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&new_nodes);
+	      return err;
+	    }
+	}
+      else
+	{
+	  /* There are problematic nodes, re-calculate incrementally.  */
+	  err = check_arrival_expand_ecl_sub (dfa, &new_nodes, cur_node,
+					      ex_subexp, type);
+	  if (BE (err != REG_NOERROR, 0))
+	    {
+	      re_node_set_free (&new_nodes);
+	      return err;
+	    }
+	}
+    }
+  re_node_set_free (cur_nodes);
+  *cur_nodes = new_nodes;
+  return REG_NOERROR;
+}
+
+/* Helper function for check_arrival_expand_ecl.
+   Check incrementally the epsilon closure of TARGET, and if it isn't
+   problematic append it to DST_NODES.  */
+
+static reg_errcode_t
+internal_function
+check_arrival_expand_ecl_sub (const re_dfa_t *dfa, re_node_set *dst_nodes,
+			      int target, int ex_subexp, int type)
+{
+  int cur_node;
+  for (cur_node = target; !re_node_set_contains (dst_nodes, cur_node);)
+    {
+      int err;
+
+      if (dfa->nodes[cur_node].type == type
+	  && dfa->nodes[cur_node].opr.idx == ex_subexp)
+	{
+	  if (type == OP_CLOSE_SUBEXP)
+	    {
+	      err = re_node_set_insert (dst_nodes, cur_node);
+	      if (BE (err == -1, 0))
+		return REG_ESPACE;
+	    }
+	  break;
+	}
+      err = re_node_set_insert (dst_nodes, cur_node);
+      if (BE (err == -1, 0))
+	return REG_ESPACE;
+      if (dfa->edests[cur_node].nelem == 0)
+	break;
+      if (dfa->edests[cur_node].nelem == 2)
+	{
+	  err = check_arrival_expand_ecl_sub (dfa, dst_nodes,
+					      dfa->edests[cur_node].elems[1],
+					      ex_subexp, type);
+	  if (BE (err != REG_NOERROR, 0))
+	    return err;
+	}
+      cur_node = dfa->edests[cur_node].elems[0];
+    }
+  return REG_NOERROR;
+}
+
+
+/* For all the back references in the current state, calculate the
+   destination of the back references by the appropriate entry
+   in MCTX->BKREF_ENTS.  */
+
+static reg_errcode_t
+internal_function
+expand_bkref_cache (re_match_context_t *mctx, re_node_set *cur_nodes,
+		    int cur_str, int subexp_num, int type)
+{
+  const re_dfa_t *const dfa = mctx->dfa;
+  reg_errcode_t err;
+  int cache_idx_start = search_cur_bkref_entry (mctx, cur_str);
+  struct re_backref_cache_entry *ent;
+
+  if (cache_idx_start == -1)
+    return REG_NOERROR;
+
+ restart:
+  ent = mctx->bkref_ents + cache_idx_start;
+  do
+    {
+      int to_idx, next_node;
+
+      /* Is this entry ENT is appropriate?  */
+      if (!re_node_set_contains (cur_nodes, ent->node))
+	continue; /* No.  */
+
+      to_idx = cur_str + ent->subexp_to - ent->subexp_from;
+      /* Calculate the destination of the back reference, and append it
+	 to MCTX->STATE_LOG.  */
+      if (to_idx == cur_str)
+	{
+	  /* The backreference did epsilon transit, we must re-check all the
+	     node in the current state.  */
+	  re_node_set new_dests;
+	  reg_errcode_t err2, err3;
+	  next_node = dfa->edests[ent->node].elems[0];
+	  if (re_node_set_contains (cur_nodes, next_node))
+	    continue;
+	  err = re_node_set_init_1 (&new_dests, next_node);
+	  err2 = check_arrival_expand_ecl (dfa, &new_dests, subexp_num, type);
+	  err3 = re_node_set_merge (cur_nodes, &new_dests);
+	  re_node_set_free (&new_dests);
+	  if (BE (err != REG_NOERROR || err2 != REG_NOERROR
+		  || err3 != REG_NOERROR, 0))
+	    {
+	      err = (err != REG_NOERROR ? err
+		     : (err2 != REG_NOERROR ? err2 : err3));
+	      return err;
+	    }
+	  /* TODO: It is still inefficient...  */
+	  goto restart;
+	}
+      else
+	{
+	  re_node_set union_set;
+	  next_node = dfa->nexts[ent->node];
+	  if (mctx->state_log[to_idx])
+	    {
+	      int ret;
+	      if (re_node_set_contains (&mctx->state_log[to_idx]->nodes,
+					next_node))
+		continue;
+	      err = re_node_set_init_copy (&union_set,
+					   &mctx->state_log[to_idx]->nodes);
+	      ret = re_node_set_insert (&union_set, next_node);
+	      if (BE (err != REG_NOERROR || ret < 0, 0))
+		{
+		  re_node_set_free (&union_set);
+		  err = err != REG_NOERROR ? err : REG_ESPACE;
+		  return err;
+		}
+	    }
+	  else
+	    {
+	      err = re_node_set_init_1 (&union_set, next_node);
+	      if (BE (err != REG_NOERROR, 0))
+		return err;
+	    }
+	  mctx->state_log[to_idx] = re_acquire_state (&err, dfa, &union_set);
+	  re_node_set_free (&union_set);
+	  if (BE (mctx->state_log[to_idx] == NULL
+		  && err != REG_NOERROR, 0))
+	    return err;
+	}
+    }
+  while (ent++->more);
+  return REG_NOERROR;
+}
+
+/* Build transition table for the state.
+   Return 1 if succeeded, otherwise return NULL.  */
+
+static int
+internal_function
+build_trtable (const re_dfa_t *dfa, re_dfastate_t *state)
+{
+  reg_errcode_t err;
+  int i, j, ch, need_word_trtable = 0;
+  bitset_word_t elem, mask;
+  bool dests_node_malloced = false;
+  bool dest_states_malloced = false;
+  int ndests; /* Number of the destination states from `state'.  */
+  re_dfastate_t **trtable;
+  re_dfastate_t **dest_states = NULL, **dest_states_word, **dest_states_nl;
+  re_node_set follows, *dests_node;
+  bitset_t *dests_ch;
+  bitset_t acceptable;
+
+  struct dests_alloc
+  {
+    re_node_set dests_node[SBC_MAX];
+    bitset_t dests_ch[SBC_MAX];
+  } *dests_alloc;
+
+  /* We build DFA states which corresponds to the destination nodes
+     from `state'.  `dests_node[i]' represents the nodes which i-th
+     destination state contains, and `dests_ch[i]' represents the
+     characters which i-th destination state accepts.  */
+#ifdef HAVE_ALLOCA
+  if (__libc_use_alloca (sizeof (struct dests_alloc)))
+    dests_alloc = (struct dests_alloc *) alloca (sizeof (struct dests_alloc));
+  else
+#endif
+    {
+      dests_alloc = re_malloc (struct dests_alloc, 1);
+      if (BE (dests_alloc == NULL, 0))
+	return 0;
+      dests_node_malloced = true;
+    }
+  dests_node = dests_alloc->dests_node;
+  dests_ch = dests_alloc->dests_ch;
+
+  /* Initialize transiton table.  */
+  state->word_trtable = state->trtable = NULL;
+
+  /* At first, group all nodes belonging to `state' into several
+     destinations.  */
+  ndests = group_nodes_into_DFAstates (dfa, state, dests_node, dests_ch);
+  if (BE (ndests <= 0, 0))
+    {
+      if (dests_node_malloced)
+	free (dests_alloc);
+      /* Return 0 in case of an error, 1 otherwise.  */
+      if (ndests == 0)
+	{
+	  state->trtable = (re_dfastate_t **)
+	    calloc (sizeof (re_dfastate_t *), SBC_MAX);
+	  return 1;
+	}
+      return 0;
+    }
+
+  err = re_node_set_alloc (&follows, ndests + 1);
+  if (BE (err != REG_NOERROR, 0))
+    goto out_free;
+
+  /* Avoid arithmetic overflow in size calculation.  */
+  if (BE ((((SIZE_MAX - (sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX)
+	    / (3 * sizeof (re_dfastate_t *)))
+	   < ndests),
+	  0))
+    goto out_free;
+
+#ifdef HAVE_ALLOCA
+  if (__libc_use_alloca ((sizeof (re_node_set) + sizeof (bitset_t)) * SBC_MAX
+			 + ndests * 3 * sizeof (re_dfastate_t *)))
+    dest_states = (re_dfastate_t **)
+      alloca (ndests * 3 * sizeof (re_dfastate_t *));
+  else
+#endif
+    {
+      dest_states = (re_dfastate_t **)
+	malloc (ndests * 3 * sizeof (re_dfastate_t *));
+      if (BE (dest_states == NULL, 0))
+	{
+out_free:
+	  if (dest_states_malloced)
+	    free (dest_states);
+	  re_node_set_free (&follows);
+	  for (i = 0; i < ndests; ++i)
+	    re_node_set_free (dests_node + i);
+	  if (dests_node_malloced)
+	    free (dests_alloc);
+	  return 0;
+	}
+      dest_states_malloced = true;
+    }
+  dest_states_word = dest_states + ndests;
+  dest_states_nl = dest_states_word + ndests;
+  bitset_empty (acceptable);
+
+  /* Then build the states for all destinations.  */
+  for (i = 0; i < ndests; ++i)
+    {
+      int next_node;
+      re_node_set_empty (&follows);
+      /* Merge the follows of this destination states.  */
+      for (j = 0; j < dests_node[i].nelem; ++j)
+	{
+	  next_node = dfa->nexts[dests_node[i].elems[j]];
+	  if (next_node != -1)
+	    {
+	      err = re_node_set_merge (&follows, dfa->eclosures + next_node);
+	      if (BE (err != REG_NOERROR, 0))
+		goto out_free;
+	    }
+	}
+      dest_states[i] = re_acquire_state_context (&err, dfa, &follows, 0);
+      if (BE (dest_states[i] == NULL && err != REG_NOERROR, 0))
+	goto out_free;
+      /* If the new state has context constraint,
+	 build appropriate states for these contexts.  */
+      if (dest_states[i]->has_constraint)
+	{
+	  dest_states_word[i] = re_acquire_state_context (&err, dfa, &follows,
+							  CONTEXT_WORD);
+	  if (BE (dest_states_word[i] == NULL && err != REG_NOERROR, 0))
+	    goto out_free;
+
+	  if (dest_states[i] != dest_states_word[i] && dfa->mb_cur_max > 1)
+	    need_word_trtable = 1;
+
+	  dest_states_nl[i] = re_acquire_state_context (&err, dfa, &follows,
+							CONTEXT_NEWLINE);
+	  if (BE (dest_states_nl[i] == NULL && err != REG_NOERROR, 0))
+	    goto out_free;
+ 	}
+      else
+	{
+	  dest_states_word[i] = dest_states[i];
+	  dest_states_nl[i] = dest_states[i];
+	}
+      bitset_merge (acceptable, dests_ch[i]);
+    }
+
+  if (!BE (need_word_trtable, 0))
+    {
+      /* We don't care about whether the following character is a word
+	 character, or we are in a single-byte character set so we can
+	 discern by looking at the character code: allocate a
+	 256-entry transition table.  */
+      trtable = state->trtable =
+	(re_dfastate_t **) calloc (sizeof (re_dfastate_t *), SBC_MAX);
+      if (BE (trtable == NULL, 0))
+	goto out_free;
+
+      /* For all characters ch...:  */
+      for (i = 0; i < BITSET_WORDS; ++i)
+	for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
+	     elem;
+	     mask <<= 1, elem >>= 1, ++ch)
+	  if (BE (elem & 1, 0))
+	    {
+	      /* There must be exactly one destination which accepts
+		 character ch.  See group_nodes_into_DFAstates.  */
+	      for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
+		;
+
+	      /* j-th destination accepts the word character ch.  */
+	      if (dfa->word_char[i] & mask)
+		trtable[ch] = dest_states_word[j];
+	      else
+		trtable[ch] = dest_states[j];
+	    }
+    }
+  else
+    {
+      /* We care about whether the following character is a word
+	 character, and we are in a multi-byte character set: discern
+	 by looking at the character code: build two 256-entry
+	 transition tables, one starting at trtable[0] and one
+	 starting at trtable[SBC_MAX].  */
+      trtable = state->word_trtable =
+	(re_dfastate_t **) calloc (sizeof (re_dfastate_t *), 2 * SBC_MAX);
+      if (BE (trtable == NULL, 0))
+	goto out_free;
+
+      /* For all characters ch...:  */
+      for (i = 0; i < BITSET_WORDS; ++i)
+	for (ch = i * BITSET_WORD_BITS, elem = acceptable[i], mask = 1;
+	     elem;
+	     mask <<= 1, elem >>= 1, ++ch)
+	  if (BE (elem & 1, 0))
+	    {
+	      /* There must be exactly one destination which accepts
+		 character ch.  See group_nodes_into_DFAstates.  */
+	      for (j = 0; (dests_ch[j][i] & mask) == 0; ++j)
+		;
+
+	      /* j-th destination accepts the word character ch.  */
+	      trtable[ch] = dest_states[j];
+	      trtable[ch + SBC_MAX] = dest_states_word[j];
+	    }
+    }
+
+  /* new line */
+  if (bitset_contain (acceptable, NEWLINE_CHAR))
+    {
+      /* The current state accepts newline character.  */
+      for (j = 0; j < ndests; ++j)
+	if (bitset_contain (dests_ch[j], NEWLINE_CHAR))
+	  {
+	    /* k-th destination accepts newline character.  */
+	    trtable[NEWLINE_CHAR] = dest_states_nl[j];
+	    if (need_word_trtable)
+	      trtable[NEWLINE_CHAR + SBC_MAX] = dest_states_nl[j];
+	    /* There must be only one destination which accepts
+	       newline.  See group_nodes_into_DFAstates.  */
+	    break;
+	  }
+    }
+
+  if (dest_states_malloced)
+    free (dest_states);
+
+  re_node_set_free (&follows);
+  for (i = 0; i < ndests; ++i)
+    re_node_set_free (dests_node + i);
+
+  if (dests_node_malloced)
+    free (dests_alloc);
+
+  return 1;
+}
+
+/* Group all nodes belonging to STATE into several destinations.
+   Then for all destinations, set the nodes belonging to the destination
+   to DESTS_NODE[i] and set the characters accepted by the destination
+   to DEST_CH[i].  This function return the number of destinations.  */
+
+static int
+internal_function
+group_nodes_into_DFAstates (const re_dfa_t *dfa, const re_dfastate_t *state,
+			    re_node_set *dests_node, bitset_t *dests_ch)
+{
+  reg_errcode_t err;
+  int result;
+  int i, j, k;
+  int ndests; /* Number of the destinations from `state'.  */
+  bitset_t accepts; /* Characters a node can accept.  */
+  const re_node_set *cur_nodes = &state->nodes;
+  bitset_empty (accepts);
+  ndests = 0;
+
+  /* For all the nodes belonging to `state',  */
+  for (i = 0; i < cur_nodes->nelem; ++i)
+    {
+      re_token_t *node = &dfa->nodes[cur_nodes->elems[i]];
+      re_token_type_t type = node->type;
+      unsigned int constraint = node->constraint;
+
+      /* Enumerate all single byte character this node can accept.  */
+      if (type == CHARACTER)
+	bitset_set (accepts, node->opr.c);
+      else if (type == SIMPLE_BRACKET)
+	{
+	  bitset_merge (accepts, node->opr.sbcset);
+	}
+      else if (type == OP_PERIOD)
+	{
+#ifdef RE_ENABLE_I18N
+	  if (dfa->mb_cur_max > 1)
+	    bitset_merge (accepts, dfa->sb_char);
+	  else
+#endif
+	    bitset_set_all (accepts);
+	  if (!(dfa->syntax & RE_DOT_NEWLINE))
+	    bitset_clear (accepts, '\n');
+	  if (dfa->syntax & RE_DOT_NOT_NULL)
+	    bitset_clear (accepts, '\0');
+	}
+#ifdef RE_ENABLE_I18N
+      else if (type == OP_UTF8_PERIOD)
+	{
+	  memset (accepts, '\xff', sizeof (bitset_t) / 2);
+	  if (!(dfa->syntax & RE_DOT_NEWLINE))
+	    bitset_clear (accepts, '\n');
+	  if (dfa->syntax & RE_DOT_NOT_NULL)
+	    bitset_clear (accepts, '\0');
+	}
+#endif
+      else
+	continue;
+
+      /* Check the `accepts' and sift the characters which are not
+	 match it the context.  */
+      if (constraint)
+	{
+	  if (constraint & NEXT_NEWLINE_CONSTRAINT)
+	    {
+	      bool accepts_newline = bitset_contain (accepts, NEWLINE_CHAR);
+	      bitset_empty (accepts);
+	      if (accepts_newline)
+		bitset_set (accepts, NEWLINE_CHAR);
+	      else
+		continue;
+	    }
+	  if (constraint & NEXT_ENDBUF_CONSTRAINT)
+	    {
+	      bitset_empty (accepts);
+	      continue;
+	    }
+
+	  if (constraint & NEXT_WORD_CONSTRAINT)
+	    {
+	      bitset_word_t any_set = 0;
+	      if (type == CHARACTER && !node->word_char)
+		{
+		  bitset_empty (accepts);
+		  continue;
+		}
+#ifdef RE_ENABLE_I18N
+	      if (dfa->mb_cur_max > 1)
+		for (j = 0; j < BITSET_WORDS; ++j)
+		  any_set |= (accepts[j] &= (dfa->word_char[j] | ~dfa->sb_char[j]));
+	      else
+#endif
+		for (j = 0; j < BITSET_WORDS; ++j)
+		  any_set |= (accepts[j] &= dfa->word_char[j]);
+	      if (!any_set)
+		continue;
+	    }
+	  if (constraint & NEXT_NOTWORD_CONSTRAINT)
+	    {
+	      bitset_word_t any_set = 0;
+	      if (type == CHARACTER && node->word_char)
+		{
+		  bitset_empty (accepts);
+		  continue;
+		}
+#ifdef RE_ENABLE_I18N
+	      if (dfa->mb_cur_max > 1)
+		for (j = 0; j < BITSET_WORDS; ++j)
+		  any_set |= (accepts[j] &= ~(dfa->word_char[j] & dfa->sb_char[j]));
+	      else
+#endif
+		for (j = 0; j < BITSET_WORDS; ++j)
+		  any_set |= (accepts[j] &= ~dfa->word_char[j]);
+	      if (!any_set)
+		continue;
+	    }
+	}
+
+      /* Then divide `accepts' into DFA states, or create a new
+	 state.  Above, we make sure that accepts is not empty.  */
+      for (j = 0; j < ndests; ++j)
+	{
+	  bitset_t intersec; /* Intersection sets, see below.  */
+	  bitset_t remains;
+	  /* Flags, see below.  */
+	  bitset_word_t has_intersec, not_subset, not_consumed;
+
+	  /* Optimization, skip if this state doesn't accept the character.  */
+	  if (type == CHARACTER && !bitset_contain (dests_ch[j], node->opr.c))
+	    continue;
+
+	  /* Enumerate the intersection set of this state and `accepts'.  */
+	  has_intersec = 0;
+	  for (k = 0; k < BITSET_WORDS; ++k)
+	    has_intersec |= intersec[k] = accepts[k] & dests_ch[j][k];
+	  /* And skip if the intersection set is empty.  */
+	  if (!has_intersec)
+	    continue;
+
+	  /* Then check if this state is a subset of `accepts'.  */
+	  not_subset = not_consumed = 0;
+	  for (k = 0; k < BITSET_WORDS; ++k)
+	    {
+	      not_subset |= remains[k] = ~accepts[k] & dests_ch[j][k];
+	      not_consumed |= accepts[k] = accepts[k] & ~dests_ch[j][k];
+	    }
+
+	  /* If this state isn't a subset of `accepts', create a
+	     new group state, which has the `remains'. */
+	  if (not_subset)
+	    {
+	      bitset_copy (dests_ch[ndests], remains);
+	      bitset_copy (dests_ch[j], intersec);
+	      err = re_node_set_init_copy (dests_node + ndests, &dests_node[j]);
+	      if (BE (err != REG_NOERROR, 0))
+		goto error_return;
+	      ++ndests;
+	    }
+
+	  /* Put the position in the current group. */
+	  result = re_node_set_insert (&dests_node[j], cur_nodes->elems[i]);
+	  if (BE (result < 0, 0))
+	    goto error_return;
+
+	  /* If all characters are consumed, go to next node. */
+	  if (!not_consumed)
+	    break;
+	}
+      /* Some characters remain, create a new group. */
+      if (j == ndests)
+	{
+	  bitset_copy (dests_ch[ndests], accepts);
+	  err = re_node_set_init_1 (dests_node + ndests, cur_nodes->elems[i]);
+	  if (BE (err != REG_NOERROR, 0))
+	    goto error_return;
+	  ++ndests;
+	  bitset_empty (accepts);
+	}
+    }
+  return ndests;
+ error_return:
+  for (j = 0; j < ndests; ++j)
+    re_node_set_free (dests_node + j);
+  return -1;
+}
+
+#ifdef RE_ENABLE_I18N
+/* Check how many bytes the node `dfa->nodes[node_idx]' accepts.
+   Return the number of the bytes the node accepts.
+   STR_IDX is the current index of the input string.
+
+   This function handles the nodes which can accept one character, or
+   one collating element like '.', '[a-z]', opposite to the other nodes
+   can only accept one byte.  */
+
+static int
+internal_function
+check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
+			 const re_string_t *input, int str_idx)
+{
+  const re_token_t *node = dfa->nodes + node_idx;
+  int char_len, elem_len;
+  int i;
+  wint_t wc;
+
+  if (BE (node->type == OP_UTF8_PERIOD, 0))
+    {
+      unsigned char c = re_string_byte_at (input, str_idx), d;
+      if (BE (c < 0xc2, 1))
+	return 0;
+
+      if (str_idx + 2 > input->len)
+	return 0;
+
+      d = re_string_byte_at (input, str_idx + 1);
+      if (c < 0xe0)
+	return (d < 0x80 || d > 0xbf) ? 0 : 2;
+      else if (c < 0xf0)
+	{
+	  char_len = 3;
+	  if (c == 0xe0 && d < 0xa0)
+	    return 0;
+	}
+      else if (c < 0xf8)
+	{
+	  char_len = 4;
+	  if (c == 0xf0 && d < 0x90)
+	    return 0;
+	}
+      else if (c < 0xfc)
+	{
+	  char_len = 5;
+	  if (c == 0xf8 && d < 0x88)
+	    return 0;
+	}
+      else if (c < 0xfe)
+	{
+	  char_len = 6;
+	  if (c == 0xfc && d < 0x84)
+	    return 0;
+	}
+      else
+	return 0;
+
+      if (str_idx + char_len > input->len)
+	return 0;
+
+      for (i = 1; i < char_len; ++i)
+	{
+	  d = re_string_byte_at (input, str_idx + i);
+	  if (d < 0x80 || d > 0xbf)
+	    return 0;
+	}
+      return char_len;
+    }
+
+  char_len = re_string_char_size_at (input, str_idx);
+  if (node->type == OP_PERIOD)
+    {
+      if (char_len <= 1)
+	return 0;
+      /* FIXME: I don't think this if is needed, as both '\n'
+	 and '\0' are char_len == 1.  */
+      /* '.' accepts any one character except the following two cases.  */
+      if ((!(dfa->syntax & RE_DOT_NEWLINE) &&
+	   re_string_byte_at (input, str_idx) == '\n') ||
+	  ((dfa->syntax & RE_DOT_NOT_NULL) &&
+	   re_string_byte_at (input, str_idx) == '\0'))
+	return 0;
+      return char_len;
+    }
+
+  elem_len = re_string_elem_size_at (input, str_idx);
+  wc = __btowc(*(input->mbs+str_idx));
+  if (((elem_len <= 1 && char_len <= 1) || char_len == 0) && (wc != WEOF && wc < SBC_MAX))
+    return 0;
+
+  if (node->type == COMPLEX_BRACKET)
+    {
+      const re_charset_t *cset = node->opr.mbcset;
+# ifdef _LIBC
+      const unsigned char *pin
+	= ((const unsigned char *) re_string_get_buffer (input) + str_idx);
+      int j;
+      uint32_t nrules;
+# endif /* _LIBC */
+      int match_len = 0;
+      wchar_t wc = ((cset->nranges || cset->nchar_classes || cset->nmbchars)
+		    ? re_string_wchar_at (input, str_idx) : 0);
+
+      /* match with multibyte character?  */
+      for (i = 0; i < cset->nmbchars; ++i)
+	if (wc == cset->mbchars[i])
+	  {
+	    match_len = char_len;
+	    goto check_node_accept_bytes_match;
+	  }
+      /* match with character_class?  */
+      for (i = 0; i < cset->nchar_classes; ++i)
+	{
+	  wctype_t wt = cset->char_classes[i];
+	  if (__iswctype (wc, wt))
+	    {
+	      match_len = char_len;
+	      goto check_node_accept_bytes_match;
+	    }
+	}
+
+# ifdef _LIBC
+      nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
+      if (nrules != 0)
+	{
+	  unsigned int in_collseq = 0;
+	  const int32_t *table, *indirect;
+	  const unsigned char *weights, *extra;
+	  const char *collseqwc;
+	  /* This #include defines a local function!  */
+#  include <locale/weight.h>
+
+	  /* match with collating_symbol?  */
+	  if (cset->ncoll_syms)
+	    extra = (const unsigned char *)
+	      _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
+	  for (i = 0; i < cset->ncoll_syms; ++i)
+	    {
+	      const unsigned char *coll_sym = extra + cset->coll_syms[i];
+	      /* Compare the length of input collating element and
+		 the length of current collating element.  */
+	      if (*coll_sym != elem_len)
+		continue;
+	      /* Compare each bytes.  */
+	      for (j = 0; j < *coll_sym; j++)
+		if (pin[j] != coll_sym[1 + j])
+		  break;
+	      if (j == *coll_sym)
+		{
+		  /* Match if every bytes is equal.  */
+		  match_len = j;
+		  goto check_node_accept_bytes_match;
+		}
+	    }
+
+	  if (cset->nranges)
+	    {
+	      if (elem_len <= char_len)
+		{
+		  collseqwc = _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQWC);
+		  in_collseq = __collseq_table_lookup (collseqwc, wc);
+		}
+	      else
+		in_collseq = find_collation_sequence_value (pin, elem_len);
+	    }
+	  /* match with range expression?  */
+	  for (i = 0; i < cset->nranges; ++i)
+	    if (cset->range_starts[i] <= in_collseq
+		&& in_collseq <= cset->range_ends[i])
+	      {
+		match_len = elem_len;
+		goto check_node_accept_bytes_match;
+	      }
+
+	  /* match with equivalence_class?  */
+	  if (cset->nequiv_classes)
+	    {
+	      const unsigned char *cp = pin;
+	      table = (const int32_t *)
+		_NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
+	      weights = (const unsigned char *)
+		_NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
+	      extra = (const unsigned char *)
+		_NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
+	      indirect = (const int32_t *)
+		_NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
+	      int32_t idx = findidx (&cp);
+	      if (idx > 0)
+		for (i = 0; i < cset->nequiv_classes; ++i)
+		  {
+		    int32_t equiv_class_idx = cset->equiv_classes[i];
+		    size_t weight_len = weights[idx & 0xffffff];
+		    if (weight_len == weights[equiv_class_idx & 0xffffff]
+			&& (idx >> 24) == (equiv_class_idx >> 24))
+		      {
+			int cnt = 0;
+
+			idx &= 0xffffff;
+			equiv_class_idx &= 0xffffff;
+
+			while (cnt <= weight_len
+			       && (weights[equiv_class_idx + 1 + cnt]
+				   == weights[idx + 1 + cnt]))
+			  ++cnt;
+			if (cnt > weight_len)
+			  {
+			    match_len = elem_len;
+			    goto check_node_accept_bytes_match;
+			  }
+		      }
+		  }
+	    }
+	}
+      else
+# endif /* _LIBC */
+	{
+	  /* match with range expression?  */
+#if __GNUC__ >= 2
+	  wchar_t cmp_buf[] = {L'\0', L'\0', wc, L'\0', L'\0', L'\0'};
+#else
+	  wchar_t cmp_buf[] = {L'\0', L'\0', L'\0', L'\0', L'\0', L'\0'};
+	  cmp_buf[2] = wc;
+#endif
+	  for (i = 0; i < cset->nranges; ++i)
+	    {
+	      cmp_buf[0] = cset->range_starts[i];
+	      cmp_buf[4] = cset->range_ends[i];
+	      if (wcscoll (cmp_buf, cmp_buf + 2) <= 0
+		  && wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0)
+		{
+		  match_len = char_len;
+		  goto check_node_accept_bytes_match;
+		}
+	    }
+	}
+    check_node_accept_bytes_match:
+      if (!cset->non_match)
+	return match_len;
+      else
+	{
+	  if (match_len > 0)
+	    return 0;
+	  else
+	    return (elem_len > char_len) ? elem_len : char_len;
+	}
+    }
+  return 0;
+}
+
+# ifdef _LIBC
+static unsigned int
+internal_function
+find_collation_sequence_value (const unsigned char *mbs, size_t mbs_len)
+{
+  uint32_t nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
+  if (nrules == 0)
+    {
+      if (mbs_len == 1)
+	{
+	  /* No valid character.  Match it as a single byte character.  */
+	  const unsigned char *collseq = (const unsigned char *)
+	    _NL_CURRENT (LC_COLLATE, _NL_COLLATE_COLLSEQMB);
+	  return collseq[mbs[0]];
+	}
+      return UINT_MAX;
+    }
+  else
+    {
+      int32_t idx;
+      const unsigned char *extra = (const unsigned char *)
+	_NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
+      int32_t extrasize = (const unsigned char *)
+	_NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB + 1) - extra;
+
+      for (idx = 0; idx < extrasize;)
+	{
+	  int mbs_cnt, found = 0;
+	  int32_t elem_mbs_len;
+	  /* Skip the name of collating element name.  */
+	  idx = idx + extra[idx] + 1;
+	  elem_mbs_len = extra[idx++];
+	  if (mbs_len == elem_mbs_len)
+	    {
+	      for (mbs_cnt = 0; mbs_cnt < elem_mbs_len; ++mbs_cnt)
+		if (extra[idx + mbs_cnt] != mbs[mbs_cnt])
+		  break;
+	      if (mbs_cnt == elem_mbs_len)
+		/* Found the entry.  */
+		found = 1;
+	    }
+	  /* Skip the byte sequence of the collating element.  */
+	  idx += elem_mbs_len;
+	  /* Adjust for the alignment.  */
+	  idx = (idx + 3) & ~3;
+	  /* Skip the collation sequence value.  */
+	  idx += sizeof (uint32_t);
+	  /* Skip the wide char sequence of the collating element.  */
+	  idx = idx + sizeof (uint32_t) * (extra[idx] + 1);
+	  /* If we found the entry, return the sequence value.  */
+	  if (found)
+	    return *(uint32_t *) (extra + idx);
+	  /* Skip the collation sequence value.  */
+	  idx += sizeof (uint32_t);
+	}
+      return UINT_MAX;
+    }
+}
+# endif /* _LIBC */
+#endif /* RE_ENABLE_I18N */
+
+/* Check whether the node accepts the byte which is IDX-th
+   byte of the INPUT.  */
+
+static int
+internal_function
+check_node_accept (const re_match_context_t *mctx, const re_token_t *node,
+		   int idx)
+{
+  unsigned char ch;
+  ch = re_string_byte_at (&mctx->input, idx);
+  switch (node->type)
+    {
+    case CHARACTER:
+      if (node->opr.c != ch)
+	return 0;
+      break;
+
+    case SIMPLE_BRACKET:
+      if (!bitset_contain (node->opr.sbcset, ch))
+	return 0;
+      break;
+
+#ifdef RE_ENABLE_I18N
+    case OP_UTF8_PERIOD:
+      if (ch >= 0x80)
+	return 0;
+      /* FALLTHROUGH */
+#endif
+    case OP_PERIOD:
+      if ((ch == '\n' && !(mctx->dfa->syntax & RE_DOT_NEWLINE))
+	  || (ch == '\0' && (mctx->dfa->syntax & RE_DOT_NOT_NULL)))
+	return 0;
+      break;
+
+    default:
+      return 0;
+    }
+
+  if (node->constraint)
+    {
+      /* The node has constraints.  Check whether the current context
+	 satisfies the constraints.  */
+      unsigned int context = re_string_context_at (&mctx->input, idx,
+						   mctx->eflags);
+      if (NOT_SATISFY_NEXT_CONSTRAINT (node->constraint, context))
+	return 0;
+    }
+
+  return 1;
+}
+
+/* Extend the buffers, if the buffers have run out.  */
+
+static reg_errcode_t
+internal_function
+extend_buffers (re_match_context_t *mctx)
+{
+  reg_errcode_t ret;
+  re_string_t *pstr = &mctx->input;
+
+  /* Avoid overflow.  */
+  if (BE (INT_MAX / 2 / sizeof (re_dfastate_t *) <= pstr->bufs_len, 0))
+    return REG_ESPACE;
+
+  /* Double the lengths of the buffers.  */
+  ret = re_string_realloc_buffers (pstr, pstr->bufs_len * 2);
+  if (BE (ret != REG_NOERROR, 0))
+    return ret;
+
+  if (mctx->state_log != NULL)
+    {
+      /* And double the length of state_log.  */
+      /* XXX We have no indication of the size of this buffer.  If this
+	 allocation fail we have no indication that the state_log array
+	 does not have the right size.  */
+      re_dfastate_t **new_array = re_realloc (mctx->state_log, re_dfastate_t *,
+					      pstr->bufs_len + 1);
+      if (BE (new_array == NULL, 0))
+	return REG_ESPACE;
+      mctx->state_log = new_array;
+    }
+
+  /* Then reconstruct the buffers.  */
+  if (pstr->icase)
+    {
+#ifdef RE_ENABLE_I18N
+      if (pstr->mb_cur_max > 1)
+	{
+	  ret = build_wcs_upper_buffer (pstr);
+	  if (BE (ret != REG_NOERROR, 0))
+	    return ret;
+	}
+      else
+#endif /* RE_ENABLE_I18N  */
+	build_upper_buffer (pstr);
+    }
+  else
+    {
+#ifdef RE_ENABLE_I18N
+      if (pstr->mb_cur_max > 1)
+	build_wcs_buffer (pstr);
+      else
+#endif /* RE_ENABLE_I18N  */
+	{
+	  if (pstr->trans != NULL)
+	    re_string_translate_buffer (pstr);
+	}
+    }
+  return REG_NOERROR;
+}
+
+
+/* Functions for matching context.  */
+
+/* Initialize MCTX.  */
+
+static reg_errcode_t
+internal_function
+match_ctx_init (re_match_context_t *mctx, int eflags, int n)
+{
+  mctx->eflags = eflags;
+  mctx->match_last = -1;
+  if (n > 0)
+    {
+      mctx->bkref_ents = re_malloc (struct re_backref_cache_entry, n);
+      mctx->sub_tops = re_malloc (re_sub_match_top_t *, n);
+      if (BE (mctx->bkref_ents == NULL || mctx->sub_tops == NULL, 0))
+	return REG_ESPACE;
+    }
+  /* Already zero-ed by the caller.
+     else
+       mctx->bkref_ents = NULL;
+     mctx->nbkref_ents = 0;
+     mctx->nsub_tops = 0;  */
+  mctx->abkref_ents = n;
+  mctx->max_mb_elem_len = 1;
+  mctx->asub_tops = n;
+  return REG_NOERROR;
+}
+
+/* Clean the entries which depend on the current input in MCTX.
+   This function must be invoked when the matcher changes the start index
+   of the input, or changes the input string.  */
+
+static void
+internal_function
+match_ctx_clean (re_match_context_t *mctx)
+{
+  int st_idx;
+  for (st_idx = 0; st_idx < mctx->nsub_tops; ++st_idx)
+    {
+      int sl_idx;
+      re_sub_match_top_t *top = mctx->sub_tops[st_idx];
+      for (sl_idx = 0; sl_idx < top->nlasts; ++sl_idx)
+	{
+	  re_sub_match_last_t *last = top->lasts[sl_idx];
+	  re_free (last->path.array);
+	  re_free (last);
+	}
+      re_free (top->lasts);
+      if (top->path)
+	{
+	  re_free (top->path->array);
+	  re_free (top->path);
+	}
+      free (top);
+    }
+
+  mctx->nsub_tops = 0;
+  mctx->nbkref_ents = 0;
+}
+
+/* Free all the memory associated with MCTX.  */
+
+static void
+internal_function
+match_ctx_free (re_match_context_t *mctx)
+{
+  /* First, free all the memory associated with MCTX->SUB_TOPS.  */
+  match_ctx_clean (mctx);
+  re_free (mctx->sub_tops);
+  re_free (mctx->bkref_ents);
+}
+
+/* Add a new backreference entry to MCTX.
+   Note that we assume that caller never call this function with duplicate
+   entry, and call with STR_IDX which isn't smaller than any existing entry.
+*/
+
+static reg_errcode_t
+internal_function
+match_ctx_add_entry (re_match_context_t *mctx, int node, int str_idx, int from,
+		     int to)
+{
+  if (mctx->nbkref_ents >= mctx->abkref_ents)
+    {
+      struct re_backref_cache_entry* new_entry;
+      new_entry = re_realloc (mctx->bkref_ents, struct re_backref_cache_entry,
+			      mctx->abkref_ents * 2);
+      if (BE (new_entry == NULL, 0))
+	{
+	  re_free (mctx->bkref_ents);
+	  return REG_ESPACE;
+	}
+      mctx->bkref_ents = new_entry;
+      memset (mctx->bkref_ents + mctx->nbkref_ents, '\0',
+	      sizeof (struct re_backref_cache_entry) * mctx->abkref_ents);
+      mctx->abkref_ents *= 2;
+    }
+  if (mctx->nbkref_ents > 0
+      && mctx->bkref_ents[mctx->nbkref_ents - 1].str_idx == str_idx)
+    mctx->bkref_ents[mctx->nbkref_ents - 1].more = 1;
+
+  mctx->bkref_ents[mctx->nbkref_ents].node = node;
+  mctx->bkref_ents[mctx->nbkref_ents].str_idx = str_idx;
+  mctx->bkref_ents[mctx->nbkref_ents].subexp_from = from;
+  mctx->bkref_ents[mctx->nbkref_ents].subexp_to = to;
+
+  /* This is a cache that saves negative results of check_dst_limits_calc_pos.
+     If bit N is clear, means that this entry won't epsilon-transition to
+     an OP_OPEN_SUBEXP or OP_CLOSE_SUBEXP for the N+1-th subexpression.  If
+     it is set, check_dst_limits_calc_pos_1 will recurse and try to find one
+     such node.
+
+     A backreference does not epsilon-transition unless it is empty, so set
+     to all zeros if FROM != TO.  */
+  mctx->bkref_ents[mctx->nbkref_ents].eps_reachable_subexps_map
+    = (from == to ? ~0 : 0);
+
+  mctx->bkref_ents[mctx->nbkref_ents++].more = 0;
+  if (mctx->max_mb_elem_len < to - from)
+    mctx->max_mb_elem_len = to - from;
+  return REG_NOERROR;
+}
+
+/* Search for the first entry which has the same str_idx, or -1 if none is
+   found.  Note that MCTX->BKREF_ENTS is already sorted by MCTX->STR_IDX.  */
+
+static int
+internal_function
+search_cur_bkref_entry (const re_match_context_t *mctx, int str_idx)
+{
+  int left, right, mid, last;
+  last = right = mctx->nbkref_ents;
+  for (left = 0; left < right;)
+    {
+      mid = left + (right - left) / 2;
+      if (mctx->bkref_ents[mid].str_idx < str_idx)
+	left = mid + 1;
+      else
+	right = mid;
+    }
+  if (left < last && mctx->bkref_ents[left].str_idx == str_idx)
+    return left;
+  else
+    return -1;
+}
+
+/* Register the node NODE, whose type is OP_OPEN_SUBEXP, and which matches
+   at STR_IDX.  */
+
+static reg_errcode_t
+internal_function
+match_ctx_add_subtop (re_match_context_t *mctx, int node, int str_idx)
+{
+#ifdef DEBUG
+  assert (mctx->sub_tops != NULL);
+  assert (mctx->asub_tops > 0);
+#endif
+  if (BE (mctx->nsub_tops == mctx->asub_tops, 0))
+    {
+      int new_asub_tops = mctx->asub_tops * 2;
+      re_sub_match_top_t **new_array = re_realloc (mctx->sub_tops,
+						   re_sub_match_top_t *,
+						   new_asub_tops);
+      if (BE (new_array == NULL, 0))
+	return REG_ESPACE;
+      mctx->sub_tops = new_array;
+      mctx->asub_tops = new_asub_tops;
+    }
+  mctx->sub_tops[mctx->nsub_tops] = calloc (1, sizeof (re_sub_match_top_t));
+  if (BE (mctx->sub_tops[mctx->nsub_tops] == NULL, 0))
+    return REG_ESPACE;
+  mctx->sub_tops[mctx->nsub_tops]->node = node;
+  mctx->sub_tops[mctx->nsub_tops++]->str_idx = str_idx;
+  return REG_NOERROR;
+}
+
+/* Register the node NODE, whose type is OP_CLOSE_SUBEXP, and which matches
+   at STR_IDX, whose corresponding OP_OPEN_SUBEXP is SUB_TOP.  */
+
+static re_sub_match_last_t *
+internal_function
+match_ctx_add_sublast (re_sub_match_top_t *subtop, int node, int str_idx)
+{
+  re_sub_match_last_t *new_entry;
+  if (BE (subtop->nlasts == subtop->alasts, 0))
+    {
+      int new_alasts = 2 * subtop->alasts + 1;
+      re_sub_match_last_t **new_array = re_realloc (subtop->lasts,
+						    re_sub_match_last_t *,
+						    new_alasts);
+      if (BE (new_array == NULL, 0))
+	return NULL;
+      subtop->lasts = new_array;
+      subtop->alasts = new_alasts;
+    }
+  new_entry = calloc (1, sizeof (re_sub_match_last_t));
+  if (BE (new_entry != NULL, 1))
+    {
+      subtop->lasts[subtop->nlasts] = new_entry;
+      new_entry->node = node;
+      new_entry->str_idx = str_idx;
+      ++subtop->nlasts;
+    }
+  return new_entry;
+}
+
+static void
+internal_function
+sift_ctx_init (re_sift_context_t *sctx, re_dfastate_t **sifted_sts,
+	       re_dfastate_t **limited_sts, int last_node, int last_str_idx)
+{
+  sctx->sifted_states = sifted_sts;
+  sctx->limited_states = limited_sts;
+  sctx->last_node = last_node;
+  sctx->last_str_idx = last_str_idx;
+  re_node_set_init_empty (&sctx->limits);
+}
diff --git a/compat/setenv.c b/compat/setenv.c
new file mode 100644
index 000000000000..7849f258d201
--- /dev/null
+++ b/compat/setenv.c
@@ -0,0 +1,40 @@
+#include "../git-compat-util.h"
+
+int gitsetenv(const char *name, const char *value, int replace)
+{
+	int out;
+	size_t namelen, valuelen;
+	char *envstr;
+
+	if (!name || strchr(name, '=') || !value) {
+		errno = EINVAL;
+		return -1;
+	}
+	if (!replace) {
+		char *oldval = NULL;
+		oldval = getenv(name);
+		if (oldval) return 0;
+	}
+
+	namelen = strlen(name);
+	valuelen = strlen(value);
+	envstr = malloc(st_add3(namelen, valuelen, 2));
+	if (!envstr) {
+		errno = ENOMEM;
+		return -1;
+	}
+
+	memcpy(envstr, name, namelen);
+	envstr[namelen] = '=';
+	memcpy(envstr + namelen + 1, value, valuelen);
+	envstr[namelen + valuelen + 1] = 0;
+
+	out = putenv(envstr);
+	/* putenv(3) makes the argument string part of the environment,
+	 * and changing that string modifies the environment --- which
+	 * means we do not own that storage anymore.  Do not free
+	 * envstr.
+	 */
+
+	return out;
+}
diff --git a/compat/sha1-chunked.c b/compat/sha1-chunked.c
new file mode 100644
index 000000000000..6adfcfd54051
--- /dev/null
+++ b/compat/sha1-chunked.c
@@ -0,0 +1,19 @@
+#include "cache.h"
+
+int git_SHA1_Update_Chunked(platform_SHA_CTX *c, const void *data, size_t len)
+{
+	size_t nr;
+	size_t total = 0;
+	const char *cdata = (const char*)data;
+
+	while (len) {
+		nr = len;
+		if (nr > SHA1_MAX_BLOCK_SIZE)
+			nr = SHA1_MAX_BLOCK_SIZE;
+		platform_SHA1_Update(c, cdata, nr);
+		total += nr;
+		cdata += nr;
+		len -= nr;
+	}
+	return total;
+}
diff --git a/compat/sha1-chunked.h b/compat/sha1-chunked.h
new file mode 100644
index 000000000000..7b2df28eec45
--- /dev/null
+++ b/compat/sha1-chunked.h
@@ -0,0 +1,2 @@
+
+int git_SHA1_Update_Chunked(platform_SHA_CTX *c, const void *data, size_t len);
diff --git a/compat/snprintf.c b/compat/snprintf.c
new file mode 100644
index 000000000000..0b1168853778
--- /dev/null
+++ b/compat/snprintf.c
@@ -0,0 +1,69 @@
+#include "../git-compat-util.h"
+
+/*
+ * The size parameter specifies the available space, i.e. includes
+ * the trailing NUL byte; but Windows's vsnprintf uses the entire
+ * buffer and avoids the trailing NUL, should the buffer be exactly
+ * big enough for the result. Defining SNPRINTF_SIZE_CORR to 1 will
+ * therefore remove 1 byte from the reported buffer size, so we
+ * always have room for a trailing NUL byte.
+ */
+#ifndef SNPRINTF_SIZE_CORR
+#if defined(WIN32) && (!defined(__GNUC__) || __GNUC__ < 4) && (!defined(_MSC_VER) || _MSC_VER < 1900)
+#define SNPRINTF_SIZE_CORR 1
+#else
+#define SNPRINTF_SIZE_CORR 0
+#endif
+#endif
+
+#undef vsnprintf
+int git_vsnprintf(char *str, size_t maxsize, const char *format, va_list ap)
+{
+	va_list cp;
+	char *s;
+	int ret = -1;
+
+	if (maxsize > 0) {
+		va_copy(cp, ap);
+		ret = vsnprintf(str, maxsize-SNPRINTF_SIZE_CORR, format, cp);
+		va_end(cp);
+		if (ret == maxsize-1)
+			ret = -1;
+		/* Windows does not NUL-terminate if result fills buffer */
+		str[maxsize-1] = 0;
+	}
+	if (ret != -1)
+		return ret;
+
+	s = NULL;
+	if (maxsize < 128)
+		maxsize = 128;
+
+	while (ret == -1) {
+		maxsize *= 4;
+		str = realloc(s, maxsize);
+		if (! str)
+			break;
+		s = str;
+		va_copy(cp, ap);
+		ret = vsnprintf(str, maxsize-SNPRINTF_SIZE_CORR, format, cp);
+		va_end(cp);
+		if (ret == maxsize-1)
+			ret = -1;
+	}
+	free(s);
+	return ret;
+}
+
+int git_snprintf(char *str, size_t maxsize, const char *format, ...)
+{
+	va_list ap;
+	int ret;
+
+	va_start(ap, format);
+	ret = git_vsnprintf(str, maxsize, format, ap);
+	va_end(ap);
+
+	return ret;
+}
+
diff --git a/compat/stat.c b/compat/stat.c
new file mode 100644
index 000000000000..a2d3931cb73c
--- /dev/null
+++ b/compat/stat.c
@@ -0,0 +1,48 @@
+#define _POSIX_C_SOURCE 200112L
+#include <sys/stat.h>  /* *stat, S_IS* */
+#include <sys/types.h> /* mode_t       */
+
+static inline mode_t mode_native_to_git(mode_t native_mode)
+{
+	mode_t perm_bits = native_mode & 07777;
+	if (S_ISREG(native_mode))
+		return 0100000 | perm_bits;
+	if (S_ISDIR(native_mode))
+		return 0040000 | perm_bits;
+	if (S_ISLNK(native_mode))
+		return 0120000 | perm_bits;
+	if (S_ISBLK(native_mode))
+		return 0060000 | perm_bits;
+	if (S_ISCHR(native_mode))
+		return 0020000 | perm_bits;
+	if (S_ISFIFO(native_mode))
+		return 0010000 | perm_bits;
+	if (S_ISSOCK(native_mode))
+		return 0140000 | perm_bits;
+	/* Non-standard type bits were given. */
+	return perm_bits;
+}
+
+int git_stat(const char *path, struct stat *buf)
+{
+	int rc = stat(path, buf);
+	if (rc == 0)
+		buf->st_mode = mode_native_to_git(buf->st_mode);
+	return rc;
+}
+
+int git_fstat(int fd, struct stat *buf)
+{
+	int rc = fstat(fd, buf);
+	if (rc == 0)
+		buf->st_mode = mode_native_to_git(buf->st_mode);
+	return rc;
+}
+
+int git_lstat(const char *path, struct stat *buf)
+{
+	int rc = lstat(path, buf);
+	if (rc == 0)
+		buf->st_mode = mode_native_to_git(buf->st_mode);
+	return rc;
+}
diff --git a/compat/strcasestr.c b/compat/strcasestr.c
new file mode 100644
index 000000000000..26896deca64c
--- /dev/null
+++ b/compat/strcasestr.c
@@ -0,0 +1,22 @@
+#include "../git-compat-util.h"
+
+char *gitstrcasestr(const char *haystack, const char *needle)
+{
+	int nlen = strlen(needle);
+	int hlen = strlen(haystack) - nlen + 1;
+	int i;
+
+	for (i = 0; i < hlen; i++) {
+		int j;
+		for (j = 0; j < nlen; j++) {
+			unsigned char c1 = haystack[i+j];
+			unsigned char c2 = needle[j];
+			if (toupper(c1) != toupper(c2))
+				goto next;
+		}
+		return (char *) haystack + i;
+	next:
+		;
+	}
+	return NULL;
+}
diff --git a/compat/strdup.c b/compat/strdup.c
new file mode 100644
index 000000000000..f3fb978eb3cc
--- /dev/null
+++ b/compat/strdup.c
@@ -0,0 +1,11 @@
+#include "../git-compat-util.h"
+
+char *gitstrdup(const char *s1)
+{
+	size_t len = strlen(s1) + 1;
+	char *s2 = malloc(len);
+
+	if (s2)
+		memcpy(s2, s1, len);
+	return s2;
+}
diff --git a/compat/strlcpy.c b/compat/strlcpy.c
new file mode 100644
index 000000000000..4024c360301e
--- /dev/null
+++ b/compat/strlcpy.c
@@ -0,0 +1,13 @@
+#include "../git-compat-util.h"
+
+size_t gitstrlcpy(char *dest, const char *src, size_t size)
+{
+	size_t ret = strlen(src);
+
+	if (size) {
+		size_t len = (ret >= size) ? size - 1 : ret;
+		memcpy(dest, src, len);
+		dest[len] = '\0';
+	}
+	return ret;
+}
diff --git a/compat/strtoimax.c b/compat/strtoimax.c
new file mode 100644
index 000000000000..ac09ed89e710
--- /dev/null
+++ b/compat/strtoimax.c
@@ -0,0 +1,10 @@
+#include "../git-compat-util.h"
+
+intmax_t gitstrtoimax (const char *nptr, char **endptr, int base)
+{
+#if defined(NO_STRTOULL)
+	return strtol(nptr, endptr, base);
+#else
+	return strtoll(nptr, endptr, base);
+#endif
+}
diff --git a/compat/strtoumax.c b/compat/strtoumax.c
new file mode 100644
index 000000000000..5541353a77a2
--- /dev/null
+++ b/compat/strtoumax.c
@@ -0,0 +1,10 @@
+#include "../git-compat-util.h"
+
+uintmax_t gitstrtoumax (const char *nptr, char **endptr, int base)
+{
+#if defined(NO_STRTOULL)
+	return strtoul(nptr, endptr, base);
+#else
+	return strtoull(nptr, endptr, base);
+#endif
+}
diff --git a/compat/terminal.c b/compat/terminal.c
new file mode 100644
index 000000000000..fa13ee672db3
--- /dev/null
+++ b/compat/terminal.c
@@ -0,0 +1,147 @@
+#include "git-compat-util.h"
+#include "compat/terminal.h"
+#include "sigchain.h"
+#include "strbuf.h"
+
+#if defined(HAVE_DEV_TTY) || defined(GIT_WINDOWS_NATIVE)
+
+static void restore_term(void);
+
+static void restore_term_on_signal(int sig)
+{
+	restore_term();
+	sigchain_pop(sig);
+	raise(sig);
+}
+
+#ifdef HAVE_DEV_TTY
+
+#define INPUT_PATH "/dev/tty"
+#define OUTPUT_PATH "/dev/tty"
+
+static int term_fd = -1;
+static struct termios old_term;
+
+static void restore_term(void)
+{
+	if (term_fd < 0)
+		return;
+
+	tcsetattr(term_fd, TCSAFLUSH, &old_term);
+	close(term_fd);
+	term_fd = -1;
+}
+
+static int disable_echo(void)
+{
+	struct termios t;
+
+	term_fd = open("/dev/tty", O_RDWR);
+	if (tcgetattr(term_fd, &t) < 0)
+		goto error;
+
+	old_term = t;
+	sigchain_push_common(restore_term_on_signal);
+
+	t.c_lflag &= ~ECHO;
+	if (!tcsetattr(term_fd, TCSAFLUSH, &t))
+		return 0;
+
+error:
+	close(term_fd);
+	term_fd = -1;
+	return -1;
+}
+
+#elif defined(GIT_WINDOWS_NATIVE)
+
+#define INPUT_PATH "CONIN$"
+#define OUTPUT_PATH "CONOUT$"
+#define FORCE_TEXT "t"
+
+static HANDLE hconin = INVALID_HANDLE_VALUE;
+static DWORD cmode;
+
+static void restore_term(void)
+{
+	if (hconin == INVALID_HANDLE_VALUE)
+		return;
+
+	SetConsoleMode(hconin, cmode);
+	CloseHandle(hconin);
+	hconin = INVALID_HANDLE_VALUE;
+}
+
+static int disable_echo(void)
+{
+	hconin = CreateFile("CONIN$", GENERIC_READ | GENERIC_WRITE,
+	    FILE_SHARE_READ, NULL, OPEN_EXISTING,
+	    FILE_ATTRIBUTE_NORMAL, NULL);
+	if (hconin == INVALID_HANDLE_VALUE)
+		return -1;
+
+	GetConsoleMode(hconin, &cmode);
+	sigchain_push_common(restore_term_on_signal);
+	if (!SetConsoleMode(hconin, cmode & (~ENABLE_ECHO_INPUT))) {
+		CloseHandle(hconin);
+		hconin = INVALID_HANDLE_VALUE;
+		return -1;
+	}
+
+	return 0;
+}
+
+#endif
+
+#ifndef FORCE_TEXT
+#define FORCE_TEXT
+#endif
+
+char *git_terminal_prompt(const char *prompt, int echo)
+{
+	static struct strbuf buf = STRBUF_INIT;
+	int r;
+	FILE *input_fh, *output_fh;
+
+	input_fh = fopen(INPUT_PATH, "r" FORCE_TEXT);
+	if (!input_fh)
+		return NULL;
+
+	output_fh = fopen(OUTPUT_PATH, "w" FORCE_TEXT);
+	if (!output_fh) {
+		fclose(input_fh);
+		return NULL;
+	}
+
+	if (!echo && disable_echo()) {
+		fclose(input_fh);
+		fclose(output_fh);
+		return NULL;
+	}
+
+	fputs(prompt, output_fh);
+	fflush(output_fh);
+
+	r = strbuf_getline_lf(&buf, input_fh);
+	if (!echo) {
+		putc('\n', output_fh);
+		fflush(output_fh);
+	}
+
+	restore_term();
+	fclose(input_fh);
+	fclose(output_fh);
+
+	if (r == EOF)
+		return NULL;
+	return buf.buf;
+}
+
+#else
+
+char *git_terminal_prompt(const char *prompt, int echo)
+{
+	return getpass(prompt);
+}
+
+#endif
diff --git a/compat/terminal.h b/compat/terminal.h
new file mode 100644
index 000000000000..97db7cd69d65
--- /dev/null
+++ b/compat/terminal.h
@@ -0,0 +1,6 @@
+#ifndef COMPAT_TERMINAL_H
+#define COMPAT_TERMINAL_H
+
+char *git_terminal_prompt(const char *prompt, int echo);
+
+#endif /* COMPAT_TERMINAL_H */
diff --git a/compat/unsetenv.c b/compat/unsetenv.c
new file mode 100644
index 000000000000..bf5fd7063bc9
--- /dev/null
+++ b/compat/unsetenv.c
@@ -0,0 +1,27 @@
+#include "../git-compat-util.h"
+
+void gitunsetenv (const char *name)
+{
+#if !defined(__MINGW32__)
+     extern char **environ;
+#endif
+     int src, dst;
+     size_t nmln;
+
+     nmln = strlen(name);
+
+     for (src = dst = 0; environ[src]; ++src) {
+	  size_t enln;
+	  enln = strlen(environ[src]);
+	  if (enln > nmln) {
+               /* might match, and can test for '=' safely */
+	       if (0 == strncmp (environ[src], name, nmln)
+		   && '=' == environ[src][nmln])
+		    /* matches, so skip */
+		    continue;
+	  }
+	  environ[dst] = environ[src];
+	  ++dst;
+     }
+     environ[dst] = NULL;
+}
diff --git a/compat/vcbuild/.gitignore b/compat/vcbuild/.gitignore
new file mode 100644
index 000000000000..8f8b794ef39f
--- /dev/null
+++ b/compat/vcbuild/.gitignore
@@ -0,0 +1,3 @@
+/vcpkg/
+/MSVC-DEFS-GEN
+/VCPKG-DEFS
diff --git a/compat/vcbuild/README b/compat/vcbuild/README
new file mode 100644
index 000000000000..1b6dabf5a231
--- /dev/null
+++ b/compat/vcbuild/README
@@ -0,0 +1,112 @@
+The Steps to Build Git with VS2015 or VS2017 from the command line.
+
+1. Install the "vcpkg" open source package manager and build essential
+   third-party libraries.  The steps for this have been captured in a
+   set of convenience scripts.  These can be run from a stock Command
+   Prompt or from an SDK bash window:
+
+   $ cd <repo_root>
+   $ ./compat/vcbuild/vcpkg_install.bat
+
+   The vcpkg tools and all of the third-party sources will be installed
+   in this folder:
+      <repo_root>/compat/vcbuild/vcpkg/
+
+   A file will be created with a set of Makefile macros pointing to a
+   unified "include", "lib", and "bin" directory (release and debug) for
+   all of the required packages.  This file will be included by the main
+   Makefile:
+      <repo_root>/compat/vcbuild/MSVC-DEFS-GEN
+
+2. OPTIONALLY copy the third-party *.dll and *.pdb files into the repo
+   root to make it easier to run and debug git.exe without having to
+   manipulate your PATH.  This is especially true for debug sessions in
+   Visual Studio.
+
+   Use ONE of the following forms which should match how you want to
+   compile git.exe.
+
+   $ ./compat/vcbuild/vcpkg_copy_packages.bat debug
+   $ ./compat/vcbuild/vcpkg_copy_packages.bat release
+
+3. Build git using MSVC from an SDK bash window using one of the
+   following commands:
+
+   $ make MSVC=1
+   $ make MSVC=1 DEBUG=1
+
+================================================================
+
+Alternatively, run `make vcxproj` and then load the generated `git.sln` in
+Visual Studio. The initial build will install the vcpkg system and build the
+dependencies automatically. This will take a while.
+
+Instead of generating the `git.sln` file yourself (which requires a full Git
+for Windows SDK), you may want to consider fetching the `vs/master` branch of
+https://github.com/git-for-windows/git instead (which is updated automatically
+via CI running `make vcxproj`). The `vs/master` branch does not require a Git
+for Windows to build, but you can run the test scripts in a regular Git Bash.
+
+Note that `make vcxproj` will automatically add and commit the generated `.sln`
+and `.vcxproj` files to the repo. This is necessary to allow building a
+fully-testable Git in Visual Studio, where a regular Git Bash can be used to
+run the test scripts (as opposed to a full Git for Windows SDK): a number of
+build targets, such as Git commands implemented as Unix shell scripts (where
+`@@SHELL_PATH@@` and other placeholders are interpolated) require a full-blown
+Git for Windows SDK (which is about 10x the size of a regular Git for Windows
+installation).
+
+If your plan is to open a Pull Request with Git for Windows, it is a good idea
+to drop this commit before submitting.
+
+================================================================
+The Steps of Build Git with VS2008
+
+1. You need the build environment, which contains the Git dependencies
+   to be able to compile, link and run Git with MSVC.
+
+   You can either use the binary repository:
+
+       WWW: http://repo.or.cz/w/msvcgit.git
+       Git: git clone git://repo.or.cz/msvcgit.git
+       Zip: http://repo.or.cz/w/msvcgit.git?a=snapshot;h=master;sf=zip
+
+   and call the setup_32bit_env.cmd batch script before compiling Git,
+   (see repo/package README for details), or the source repository:
+
+       WWW: http://repo.or.cz/w/gitbuild.git
+       Git: git clone git://repo.or.cz/gitbuild.git
+       Zip: (None, as it's a project with submodules)
+
+   and build the support libs as instructed in that repo/package.
+
+2. Ensure you have the msysgit environment in your path, so you have
+   GNU Make, bash and perl available.
+
+       WWW: http://repo.or.cz/w/msysgit.git
+       Git: git clone git://repo.or.cz/msysgit.git
+       Zip: http://repo.or.cz/w/msysgit.git?a=snapshot;h=master;sf=zip
+
+   This environment is also needed when you use the resulting
+   executables, since Git might need to run scripts which are part of
+   the git operations.
+
+3. Inside Git's directory run the command:
+       make command-list.h
+   to generate the command-list.h file needed to compile git.
+
+4. Then either build Git with the GNU Make Makefile in the Git projects
+   root
+       make MSVC=1
+   or generate Visual Studio solution/projects (.sln/.vcproj) with the
+   command
+       perl contrib/buildsystems/generate -g Vcproj
+   and open and build the solution with the IDE
+       devenv git.sln /useenv
+   or build with the IDE build engine directly from the command line
+       devenv git.sln /useenv /build "Release|Win32"
+   The /useenv option is required, so Visual Studio picks up the
+   environment variables for the support libraries required to build
+   Git, which you set up in step 1.
+
+Done!
diff --git a/compat/vcbuild/find_vs_env.bat b/compat/vcbuild/find_vs_env.bat
new file mode 100644
index 000000000000..40194dd2309b
--- /dev/null
+++ b/compat/vcbuild/find_vs_env.bat
@@ -0,0 +1,168 @@
+@ECHO OFF
+REM ================================================================
+REM You can use either GCC (the default) or MSVC to build git
+REM using the GIT-SDK command line tools.
+REM        $ make
+REM        $ make MSVC=1
+REM
+REM GIT-SDK BASH windows inherit environment variables with all of
+REM the bin/lib/include paths for GCC.  It DOES NOT inherit values
+REM for the corresponding MSVC tools.
+REM
+REM During normal (non-git) Windows development, you launch one
+REM of the provided "developer command prompts" to set environment
+REM variables for the MSVC tools.
+REM
+REM Therefore, to allow MSVC command line builds of git from BASH
+REM and MAKE, we must blend these two different worlds.  This script
+REM attempts to do that.
+REM ================================================================
+REM This BAT file starts in a plain (non-developer) command prompt,
+REM searches for the "best" commmand prompt setup script, installs
+REM it into the current CMD process, and exports the various MSVC
+REM environment variables for use by MAKE.
+REM
+REM The output of this script should be written to a make "include
+REM file" and referenced by the top-level Makefile.
+REM
+REM See "config.mak.uname" (look for compat/vcbuild/MSVC-DEFS-GEN).
+REM ================================================================
+REM The provided command prompts are custom to each VS release and
+REM filled with lots of internal knowledge (such as Registry settings);
+REM even their names vary by release, so it is not appropriate for us
+REM to look inside them.  Rather, just run them in a subordinate
+REM process and extract the settings we need.
+REM ================================================================
+REM
+REM Current (VS2017 and beyond)
+REM -------------------
+REM Visual Studio 2017 introduced a new installation layout and
+REM support for side-by-side installation of multiple versions of
+REM VS2017.  Furthermore, these can all coexist with installations
+REM of previous versions of VS (which have a completely different
+REM layout on disk).
+REM
+REM VS2017 Update 2 introduced a "vswhere.exe" command:
+REM https://github.com/Microsoft/vswhere
+REM https://blogs.msdn.microsoft.com/heaths/2017/02/25/vswhere-available/
+REM https://blogs.msdn.microsoft.com/vcblog/2017/03/06/finding-the-visual-c-compiler-tools-in-visual-studio-2017/
+REM
+REM VS2015
+REM ------
+REM Visual Studio 2015 uses the traditional VcVarsAll.
+REM
+REM Earlier Versions
+REM ----------------
+REM Currently unsupported.
+REM
+REM ================================================================
+REM Note: Throughout this script we use "dir <path> && <cmd>" rather
+REM than "if exist <path>" because of script problems with pathnames
+REM containing spaces.
+REM ================================================================
+
+REM Sanitize PATH to prevent git-sdk paths from confusing "wmic.exe"
+REM (called internally in some of the system BAT files).
+SET PATH=%SystemRoot%\system32;%SystemRoot%;%SystemRoot%\System32\Wbem;
+
+REM ================================================================
+
+:current
+   SET vs_where=C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe
+   dir "%vs_where%" >nul 2>nul && GOTO have_vs_where
+   GOTO not_2017
+
+:have_vs_where
+   REM Try to use VsWhere to get the location of VsDevCmd.
+
+   REM Keep VsDevCmd from cd'ing away.
+   SET VSCMD_START_DIR=.
+
+   REM Get the root of the VS product installation.
+   FOR /F "usebackq tokens=*" %%i IN (`"%vs_where%" -latest -requires Microsoft.VisualStudio.Workload.NativeDesktop -property installationPath`) DO @SET vs_ip=%%i
+
+   SET vs_devcmd=%vs_ip%\Common7\Tools\VsDevCmd.bat
+   dir "%vs_devcmd%" >nul 2>nul && GOTO have_vs_devcmd
+   GOTO not_2017
+
+:have_vs_devcmd
+   REM Use VsDevCmd to setup the environment of this process.
+   REM Setup CL for building 64-bit apps using 64-bit tools.
+   @call "%vs_devcmd%" -no_logo -arch=x64 -host_arch=x64
+
+   SET tgt=%VSCMD_ARG_TGT_ARCH%
+
+   SET mn=%VCToolsInstallDir%
+   SET msvc_includes=-I"%mn%INCLUDE"
+   SET msvc_libs=-L"%mn%lib\%tgt%"
+   SET msvc_bin_dir=%mn%bin\Host%VSCMD_ARG_HOST_ARCH%\%tgt%
+
+   SET sdk_dir=%WindowsSdkDir%
+   SET sdk_ver=%WindowsSDKVersion%
+   SET si=%sdk_dir%Include\%sdk_ver%
+   SET sdk_includes=-I"%si%ucrt" -I"%si%um" -I"%si%shared"
+   SET sl=%sdk_dir%lib\%sdk_ver%
+   SET sdk_libs=-L"%sl%ucrt\%tgt%" -L"%sl%um\%tgt%"
+
+   SET vs_ver=%VisualStudioVersion%
+
+   GOTO print_vars
+
+REM ================================================================
+
+:not_2017
+   REM See if VS2015 is installed.
+
+   SET vs_2015_bat=C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\vcvarsall.bat
+   dir "%vs_2015_bat%" >nul 2>nul && GOTO have_vs_2015
+   GOTO not_2015
+
+:have_vs_2015
+   REM Use VcVarsAll like the "x64 Native" command prompt.
+   REM Setup CL for building 64-bit apps using 64-bit tools.
+   @call "%vs_2015_bat%" amd64
+
+   REM Note that in VS2015 they use "x64" in some contexts and "amd64" in others.
+   SET mn=C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\
+   SET msvc_includes=-I"%mn%INCLUDE"
+   SET msvc_libs=-L"%mn%lib\amd64"
+   SET msvc_bin_dir=%mn%bin\amd64
+
+   SET sdk_dir=%WindowsSdkDir%
+   SET sdk_ver=%WindowsSDKVersion%
+   SET si=%sdk_dir%Include\%sdk_ver%
+   SET sdk_includes=-I"%si%ucrt" -I"%si%um" -I"%si%shared" -I"%si%winrt"
+   SET sl=%sdk_dir%lib\%sdk_ver%
+   SET sdk_libs=-L"%sl%ucrt\x64" -L"%sl%um\x64"
+
+   SET vs_ver=%VisualStudioVersion%
+
+   GOTO print_vars
+
+REM ================================================================
+
+:not_2015
+   echo "ERROR: unsupported VS version (older than VS2015)" >&2
+   EXIT /B 1
+
+REM ================================================================
+
+:print_vars
+   REM Dump the essential vars to stdout to allow the main
+   REM Makefile to include it.  See config.mak.uname.
+   REM Include DOS-style and BASH-style path for bin dir.
+
+   echo msvc_bin_dir=%msvc_bin_dir%
+   SET X1=%msvc_bin_dir:C:=/C%
+   SET X2=%X1:\=/%
+   echo msvc_bin_dir_msys=%X2%
+
+   echo msvc_includes=%msvc_includes%
+   echo msvc_libs=%msvc_libs%
+
+   echo sdk_includes=%sdk_includes%
+   echo sdk_libs=%sdk_libs%
+
+   echo vs_ver=%vs_ver%
+
+   EXIT /B 0
diff --git a/compat/vcbuild/include/sys/param.h b/compat/vcbuild/include/sys/param.h
new file mode 100644
index 000000000000..0d8552a2c6dc
--- /dev/null
+++ b/compat/vcbuild/include/sys/param.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/time.h b/compat/vcbuild/include/sys/time.h
new file mode 100644
index 000000000000..0d8552a2c6dc
--- /dev/null
+++ b/compat/vcbuild/include/sys/time.h
@@ -0,0 +1 @@
+/* Intentionally empty file to support building git with MSVC */
diff --git a/compat/vcbuild/include/sys/utime.h b/compat/vcbuild/include/sys/utime.h
new file mode 100644
index 000000000000..582589c70a1b
--- /dev/null
+++ b/compat/vcbuild/include/sys/utime.h
@@ -0,0 +1,34 @@
+#ifndef	_UTIME_H_
+#define	_UTIME_H_
+/*
+ * UTIME.H
+ * This file has no copyright assigned and is placed in the Public Domain.
+ * This file is a part of the mingw-runtime package.
+ *
+ * The mingw-runtime package and its code is distributed in the hope that it
+ * will be useful but WITHOUT ANY WARRANTY.  ALL WARRANTIES, EXPRESSED OR
+ * IMPLIED ARE HEREBY DISCLAIMED.  This includes but is not limited to
+ * warranties of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You are free to use this package and its code without limitation.
+ */
+
+/*
+ * Structure used by _utime function.
+ */
+struct _utimbuf
+{
+	time_t	actime;		/* Access time */
+	time_t	modtime;	/* Modification time */
+};
+
+#ifndef	_NO_OLDNAMES
+/* NOTE: Must be the same as _utimbuf above. */
+struct utimbuf
+{
+	time_t	actime;
+	time_t	modtime;
+};
+#endif	/* Not _NO_OLDNAMES */
+
+#endif
diff --git a/compat/vcbuild/include/unistd.h b/compat/vcbuild/include/unistd.h
new file mode 100644
index 000000000000..3a959d124ca7
--- /dev/null
+++ b/compat/vcbuild/include/unistd.h
@@ -0,0 +1,103 @@
+#ifndef _UNISTD_
+#define _UNISTD_
+
+/* Win32 define for porting git*/
+
+#ifndef _MODE_T_
+#define	_MODE_T_
+typedef unsigned short _mode_t;
+
+#ifndef	_NO_OLDNAMES
+typedef _mode_t	mode_t;
+#endif
+#endif	/* Not _MODE_T_ */
+
+#ifndef _SSIZE_T_
+#define _SSIZE_T_
+typedef long _ssize_t;
+
+#ifndef	_OFF_T_
+#define	_OFF_T_
+typedef long _off_t;
+
+#ifndef	_NO_OLDNAMES
+typedef _off_t	off_t;
+#endif
+#endif	/* Not _OFF_T_ */
+
+
+#ifndef	_NO_OLDNAMES
+typedef _ssize_t ssize_t;
+#endif
+#endif /* Not _SSIZE_T_ */
+
+typedef signed char int8_t;
+typedef unsigned char   uint8_t;
+typedef short  int16_t;
+typedef unsigned short  uint16_t;
+typedef int  int32_t;
+typedef unsigned   uint32_t;
+typedef long long  int64_t;
+typedef unsigned long long   uint64_t;
+
+typedef long long  intmax_t;
+typedef unsigned long long uintmax_t;
+
+typedef int64_t off64_t;
+
+#if !defined(_MSC_VER) || _MSC_VER < 1600
+#define INTMAX_MIN  _I64_MIN
+#define INTMAX_MAX  _I64_MAX
+#define UINTMAX_MAX _UI64_MAX
+
+#define UINT32_MAX 0xffffffff  /* 4294967295U */
+#else
+#include <stdint.h>
+#endif
+
+#define STDIN_FILENO  0
+#define STDOUT_FILENO 1
+#define STDERR_FILENO 2
+
+/* Some defines for _access nAccessMode (MS doesn't define them, but
+ * it doesn't seem to hurt to add them). */
+#define	F_OK	0	/* Check for file existence */
+/* Well maybe it does hurt.  On newer versions of MSVCRT, an access mode
+   of 1 causes invalid parameter error. */
+#define	X_OK	0	/* MS access() doesn't check for execute permission. */
+#define	W_OK	2	/* Check for write permission */
+#define	R_OK	4	/* Check for read permission */
+
+#define	_S_IFIFO	0x1000	/* FIFO */
+#define	_S_IFCHR	0x2000	/* Character */
+#define	_S_IFBLK	0x3000	/* Block: Is this ever set under w32? */
+#define	_S_IFDIR	0x4000	/* Directory */
+#define	_S_IFREG	0x8000	/* Regular */
+
+#define	_S_IFMT		0xF000	/* File type mask */
+
+#define	_S_IXUSR	_S_IEXEC
+#define	_S_IWUSR	_S_IWRITE
+#define	_S_IRUSR	_S_IREAD
+#define	_S_ISDIR(m)	(((m) & _S_IFMT) == _S_IFDIR)
+
+#define	S_IFIFO		_S_IFIFO
+#define	S_IFCHR		_S_IFCHR
+#define	S_IFBLK		_S_IFBLK
+#define	S_IFDIR		_S_IFDIR
+#define	S_IFREG		_S_IFREG
+#define	S_IFMT		_S_IFMT
+#define	S_IEXEC		_S_IEXEC
+#define	S_IWRITE	_S_IWRITE
+#define	S_IREAD		_S_IREAD
+#define	S_IRWXU		_S_IRWXU
+#define	S_IXUSR		_S_IXUSR
+#define	S_IWUSR		_S_IWUSR
+#define	S_IRUSR		_S_IRUSR
+
+
+#define	S_ISDIR(m)	(((m) & S_IFMT) == S_IFDIR)
+#define	S_ISREG(m)	(((m) & S_IFMT) == S_IFREG)
+#define	S_ISFIFO(m)	(((m) & S_IFMT) == S_IFIFO)
+
+#endif
diff --git a/compat/vcbuild/include/utime.h b/compat/vcbuild/include/utime.h
new file mode 100644
index 000000000000..8285f38fdefe
--- /dev/null
+++ b/compat/vcbuild/include/utime.h
@@ -0,0 +1 @@
+#include <sys/utime.h>
diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl
new file mode 100755
index 000000000000..c7b021bfac7a
--- /dev/null
+++ b/compat/vcbuild/scripts/clink.pl
@@ -0,0 +1,85 @@
+#!/usr/bin/perl -w
+######################################################################
+# Compiles or links files
+#
+# This is a wrapper to facilitate the compilation of Git with MSVC
+# using GNU Make as the build system. So, instead of manipulating the
+# Makefile into something nasty, just to support non-space arguments
+# etc, we use this wrapper to fix the command line options
+#
+# Copyright (C) 2009 Marius Storm-Olsen <mstormo@gmail.com>
+######################################################################
+use strict;
+my @args = ();
+my @cflags = ();
+my @lflags = ();
+my $is_linking = 0;
+my $is_debug = 0;
+while (@ARGV) {
+	my $arg = shift @ARGV;
+	if ("$arg" eq "-DDEBUG") {
+	    # Some vcpkg-based libraries have different names for release
+	    # and debug versions.  This hack assumes that -DDEBUG comes
+	    # before any "-l*" flags.
+	    $is_debug = 1;
+	}
+	if ("$arg" =~ /^-[DIMGOZ]/) {
+		push(@cflags, $arg);
+	} elsif ("$arg" eq "-o") {
+		my $file_out = shift @ARGV;
+		if ("$file_out" =~ /exe$/) {
+			$is_linking = 1;
+			# Create foo.exe and foo.pdb
+			push(@args, "-OUT:$file_out");
+		} else {
+			# Create foo.o and foo.o.pdb
+			push(@args, "-Fo$file_out");
+			push(@args, "-Fd$file_out.pdb");
+		}
+	} elsif ("$arg" eq "-lz") {
+	    if ($is_debug) {
+		push(@args, "zlibd.lib");
+	    } else{
+		push(@args, "zlib.lib");
+	    }
+	} elsif ("$arg" eq "-liconv") {
+		push(@args, "libiconv.lib");
+	} elsif ("$arg" eq "-lcrypto") {
+		push(@args, "libeay32.lib");
+	} elsif ("$arg" eq "-lssl") {
+		push(@args, "ssleay32.lib");
+	} elsif ("$arg" eq "-lcurl") {
+		my $lib = "";
+		# Newer vcpkg definitions call this libcurl_imp.lib; Do we
+		# need to use that instead?
+		foreach my $flag (@lflags) {
+			if ($flag =~ /^-LIBPATH:(.*)/) {
+				foreach my $l ("libcurl_imp.lib", "libcurl.lib") {
+					if (-f "$1/$l") {
+						$lib = $l;
+						last;
+					}
+				}
+			}
+		}
+		push(@args, $lib);
+	} elsif ("$arg" eq "-lexpat") {
+		push(@args, "expat.lib");
+	} elsif ("$arg" =~ /^-L/ && "$arg" ne "-LTCG") {
+		$arg =~ s/^-L/-LIBPATH:/;
+		push(@lflags, $arg);
+	} elsif ("$arg" =~ /^-R/) {
+		# eat
+	} else {
+		push(@args, $arg);
+	}
+}
+if ($is_linking) {
+	push(@args, @lflags);
+	unshift(@args, "link.exe");
+} else {
+	unshift(@args, "cl.exe");
+	push(@args, @cflags);
+}
+printf(STDERR "**** @args\n\n\n") if (!defined($ENV{'QUIET_GEN'}));
+exit (system(@args) != 0);
diff --git a/compat/vcbuild/scripts/lib.pl b/compat/vcbuild/scripts/lib.pl
new file mode 100755
index 000000000000..d8054e469fe8
--- /dev/null
+++ b/compat/vcbuild/scripts/lib.pl
@@ -0,0 +1,26 @@
+#!/usr/bin/perl -w
+######################################################################
+# Libifies files on Windows
+#
+# This is a wrapper to facilitate the compilation of Git with MSVC
+# using GNU Make as the build system. So, instead of manipulating the
+# Makefile into something nasty, just to support non-space arguments
+# etc, we use this wrapper to fix the command line options
+#
+# Copyright (C) 2009 Marius Storm-Olsen <mstormo@gmail.com>
+######################################################################
+use strict;
+my @args = ();
+while (@ARGV) {
+	my $arg = shift @ARGV;
+	if ("$arg" eq "rcs") {
+		# Consume the rcs option
+	} elsif ("$arg" =~ /\.a$/) {
+		push(@args, "-OUT:$arg");
+	} else {
+		push(@args, $arg);
+	}
+}
+unshift(@args, "lib.exe");
+# printf("**** @args\n");
+exit (system(@args) != 0);
diff --git a/compat/vcbuild/vcpkg_copy_dlls.bat b/compat/vcbuild/vcpkg_copy_dlls.bat
new file mode 100644
index 000000000000..13661c14f870
--- /dev/null
+++ b/compat/vcbuild/vcpkg_copy_dlls.bat
@@ -0,0 +1,39 @@
+@ECHO OFF
+REM ================================================================
+REM This script is an optional step. It copies the *.dll and *.pdb
+REM files (created by vcpkg_install.bat) into the top-level directory
+REM of the repo so that you can type "./git.exe" and find them without
+REM having to fixup your PATH.
+REM
+REM NOTE: Because the names of some DLL files change between DEBUG and
+REM NOTE: RELEASE builds when built using "vcpkg.exe", you will need
+REM NOTE: to copy up the corresponding version.
+REM ================================================================
+
+	SETLOCAL EnableDelayedExpansion
+
+	@FOR /F "delims=" %%D IN ("%~dp0") DO @SET cwd=%%~fD
+	cd %cwd%
+
+	SET arch=x64-windows
+	SET inst=%cwd%vcpkg\installed\%arch%
+
+	IF [%1]==[release] (
+		echo Copying RELEASE mode DLLs to repo root...
+	) ELSE IF [%1]==[debug] (
+		SET inst=%inst%\debug
+		echo Copying DEBUG mode DLLs to repo root...
+	) ELSE (
+		echo ERROR: Invalid argument.
+		echo Usage: %~0 release
+		echo Usage: %~0 debug
+		EXIT /B 1
+	)
+
+	xcopy /e/s/v/y %inst%\bin\*.dll ..\..\
+	xcopy /e/s/v/y %inst%\bin\*.pdb ..\..\
+
+	xcopy /e/s/v/y %inst%\bin\*.dll ..\..\t\helper\
+	xcopy /e/s/v/y %inst%\bin\*.pdb ..\..\t\helper\
+
+	EXIT /B 0
diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat
new file mode 100644
index 000000000000..ebd0bad242a8
--- /dev/null
+++ b/compat/vcbuild/vcpkg_install.bat
@@ -0,0 +1,80 @@
+@ECHO OFF
+REM ================================================================
+REM This script installs the "vcpkg" source package manager and uses
+REM it to build the third-party libraries that git requires when it
+REM is built using MSVC.
+REM
+REM [1] Install VCPKG.
+REM     [a] Create <root>/compat/vcbuild/vcpkg/
+REM     [b] Download "vcpkg".
+REM     [c] Compile using the currently installed version of VS.
+REM     [d] Create <root>/compat/vcbuild/vcpkg/vcpkg.exe
+REM
+REM [2] Install third-party libraries.
+REM     [a] Download each (which may also install CMAKE).
+REM     [b] Compile in RELEASE mode and install in:
+REM         vcpkg/installed/<arch>/{bin,lib}
+REM     [c] Compile in DEBUG mode and install in:
+REM         vcpkg/installed/<arch>/debug/{bin,lib}
+REM     [d] Install headers in:
+REM         vcpkg/installed/<arch>/include
+REM
+REM [3] Create a set of MAKE definitions for the top-level
+REM     Makefile to allow "make MSVC=1" to find the above
+REM     third-party libraries.
+REM     [a] Write vcpkg/VCPGK-DEFS
+REM
+REM https://blogs.msdn.microsoft.com/vcblog/2016/09/19/vcpkg-a-tool-to-acquire-and-build-c-open-source-libraries-on-windows/
+REM https://github.com/Microsoft/vcpkg
+REM https://vcpkg.readthedocs.io/en/latest/
+REM ================================================================
+
+	SETLOCAL EnableDelayedExpansion
+
+	@FOR /F "delims=" %%D IN ("%~dp0") DO @SET cwd=%%~fD
+	cd %cwd%
+
+	dir vcpkg\vcpkg.exe >nul 2>nul && GOTO :install_libraries
+
+	echo Fetching vcpkg in %cwd%vcpkg
+	git.exe clone https://github.com/Microsoft/vcpkg vcpkg
+	IF ERRORLEVEL 1 ( EXIT /B 1 )
+
+	cd vcpkg
+	echo Building vcpkg
+	powershell -exec bypass scripts\bootstrap.ps1
+	IF ERRORLEVEL 1 ( EXIT /B 1 )
+
+	echo Successfully installed %cwd%vcpkg\vcpkg.exe
+
+:install_libraries
+	SET arch=x64-windows
+
+	echo Installing third-party libraries...
+	FOR %%i IN (zlib expat libiconv openssl libssh2 curl) DO (
+	    cd %cwd%vcpkg
+	    IF NOT EXIST "packages\%%i_%arch%" CALL :sub__install_one %%i
+	    IF ERRORLEVEL 1 ( EXIT /B 1 )
+	)
+
+:install_defines
+	cd %cwd%
+	SET inst=%cwd%vcpkg\installed\%arch%
+
+	echo vcpkg_inc=-I"%inst%\include">VCPKG-DEFS
+	echo vcpkg_rel_lib=-L"%inst%\lib">>VCPKG-DEFS
+	echo vcpkg_rel_bin="%inst%\bin">>VCPKG-DEFS
+	echo vcpkg_dbg_lib=-L"%inst%\debug\lib">>VCPKG-DEFS
+	echo vcpkg_dbg_bin="%inst%\debug\bin">>VCPKG-DEFS
+
+	EXIT /B 0
+
+
+:sub__install_one
+	echo     Installing package %1...
+
+	.\vcpkg.exe install %1:%arch%
+	IF ERRORLEVEL 1 ( EXIT /B 1 )
+
+	echo     Finished %1
+	goto :EOF
diff --git a/compat/win32.h b/compat/win32.h
new file mode 100644
index 000000000000..a97e880757b6
--- /dev/null
+++ b/compat/win32.h
@@ -0,0 +1,41 @@
+#ifndef WIN32_H
+#define WIN32_H
+
+/* common Win32 functions for MinGW and Cygwin */
+#ifndef GIT_WINDOWS_NATIVE	/* Not defined for Cygwin */
+#include <windows.h>
+#endif
+
+static inline int file_attr_to_st_mode (DWORD attr)
+{
+	int fMode = S_IREAD;
+	if (attr & FILE_ATTRIBUTE_DIRECTORY)
+		fMode |= S_IFDIR;
+	else
+		fMode |= S_IFREG;
+	if (!(attr & FILE_ATTRIBUTE_READONLY))
+		fMode |= S_IWRITE;
+	return fMode;
+}
+
+static inline int get_file_attr(const char *fname, WIN32_FILE_ATTRIBUTE_DATA *fdata)
+{
+	if (GetFileAttributesExA(fname, GetFileExInfoStandard, fdata))
+		return 0;
+
+	switch (GetLastError()) {
+	case ERROR_ACCESS_DENIED:
+	case ERROR_SHARING_VIOLATION:
+	case ERROR_LOCK_VIOLATION:
+	case ERROR_SHARING_BUFFER_EXCEEDED:
+		return EACCES;
+	case ERROR_BUFFER_OVERFLOW:
+		return ENAMETOOLONG;
+	case ERROR_NOT_ENOUGH_MEMORY:
+		return ENOMEM;
+	default:
+		return ENOENT;
+	}
+}
+
+#endif
diff --git a/compat/win32/alloca.h b/compat/win32/alloca.h
new file mode 100644
index 000000000000..c0d7985b7edd
--- /dev/null
+++ b/compat/win32/alloca.h
@@ -0,0 +1 @@
+#include <malloc.h>
diff --git a/compat/win32/dirent.c b/compat/win32/dirent.c
new file mode 100644
index 000000000000..52420ec7d4da
--- /dev/null
+++ b/compat/win32/dirent.c
@@ -0,0 +1,92 @@
+#include "../../git-compat-util.h"
+
+struct DIR {
+	struct dirent dd_dir; /* includes d_type */
+	HANDLE dd_handle;     /* FindFirstFile handle */
+	int dd_stat;          /* 0-based index */
+};
+
+static inline void finddata2dirent(struct dirent *ent, WIN32_FIND_DATAW *fdata)
+{
+	/* convert UTF-16 name to UTF-8 */
+	xwcstoutf(ent->d_name, fdata->cFileName, sizeof(ent->d_name));
+
+	/* Set file type, based on WIN32_FIND_DATA */
+	if (fdata->dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)
+		ent->d_type = DT_DIR;
+	else
+		ent->d_type = DT_REG;
+}
+
+DIR *opendir(const char *name)
+{
+	wchar_t pattern[MAX_PATH + 2]; /* + 2 for '/' '*' */
+	WIN32_FIND_DATAW fdata;
+	HANDLE h;
+	int len;
+	DIR *dir;
+
+	/* convert name to UTF-16 and check length < MAX_PATH */
+	if ((len = xutftowcs_path(pattern, name)) < 0)
+		return NULL;
+
+	/* append optional '/' and wildcard '*' */
+	if (len && !is_dir_sep(pattern[len - 1]))
+		pattern[len++] = '/';
+	pattern[len++] = '*';
+	pattern[len] = 0;
+
+	/* open find handle */
+	h = FindFirstFileW(pattern, &fdata);
+	if (h == INVALID_HANDLE_VALUE) {
+		DWORD err = GetLastError();
+		errno = (err == ERROR_DIRECTORY) ? ENOTDIR : err_win_to_posix(err);
+		return NULL;
+	}
+
+	/* initialize DIR structure and copy first dir entry */
+	dir = xmalloc(sizeof(DIR));
+	dir->dd_handle = h;
+	dir->dd_stat = 0;
+	finddata2dirent(&dir->dd_dir, &fdata);
+	return dir;
+}
+
+struct dirent *readdir(DIR *dir)
+{
+	if (!dir) {
+		errno = EBADF; /* No set_errno for mingw */
+		return NULL;
+	}
+
+	/* if first entry, dirent has already been set up by opendir */
+	if (dir->dd_stat) {
+		/* get next entry and convert from WIN32_FIND_DATA to dirent */
+		WIN32_FIND_DATAW fdata;
+		if (FindNextFileW(dir->dd_handle, &fdata)) {
+			finddata2dirent(&dir->dd_dir, &fdata);
+		} else {
+			DWORD lasterr = GetLastError();
+			/* POSIX says you shouldn't set errno when readdir can't
+			   find any more files; so, if another error we leave it set. */
+			if (lasterr != ERROR_NO_MORE_FILES)
+				errno = err_win_to_posix(lasterr);
+			return NULL;
+		}
+	}
+
+	++dir->dd_stat;
+	return &dir->dd_dir;
+}
+
+int closedir(DIR *dir)
+{
+	if (!dir) {
+		errno = EBADF;
+		return -1;
+	}
+
+	FindClose(dir->dd_handle);
+	free(dir);
+	return 0;
+}
diff --git a/compat/win32/dirent.h b/compat/win32/dirent.h
new file mode 100644
index 000000000000..058207e4bfed
--- /dev/null
+++ b/compat/win32/dirent.h
@@ -0,0 +1,20 @@
+#ifndef DIRENT_H
+#define DIRENT_H
+
+typedef struct DIR DIR;
+
+#define DT_UNKNOWN 0
+#define DT_DIR     1
+#define DT_REG     2
+#define DT_LNK     3
+
+struct dirent {
+	unsigned char d_type;      /* file type to prevent lstat after readdir */
+	char d_name[MAX_PATH * 3]; /* file name (* 3 for UTF-8 conversion) */
+};
+
+DIR *opendir(const char *dirname);
+struct dirent *readdir(DIR *dir);
+int closedir(DIR *dir);
+
+#endif /* DIRENT_H */
diff --git a/compat/win32/git.manifest b/compat/win32/git.manifest
new file mode 100644
index 000000000000..771e3cce4315
--- /dev/null
+++ b/compat/win32/git.manifest
@@ -0,0 +1,25 @@
+<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
+<assembly xmlns="urn:schemas-microsoft-com:asm.v1" manifestVersion="1.0">
+	<assemblyIdentity type="win32" name="Git" version="0.0.0.1" />
+	<trustInfo xmlns="urn:schemas-microsoft-com:asm.v2">
+		<security>
+			<requestedPrivileges>
+				<requestedExecutionLevel level="asInvoker" uiAccess="false" />
+			</requestedPrivileges>
+		</security>
+	</trustInfo>
+	<compatibility xmlns="urn:schemas-microsoft-com:compatibility.v1">
+		<application>
+			<!-- Windows Vista -->
+			<supportedOS Id="{e2011457-1546-43c5-a5fe-008deee3d3f0}"/>
+			<!-- Windows 7 -->
+			<supportedOS Id="{35138b9a-5d96-4fbd-8e2d-a2440225f93a}"/>
+			<!-- Windows 8 -->
+			<supportedOS Id="{4a2f28e3-53b9-4441-ba9c-d69d4a4a6e38}"/>
+			<!-- Windows 8.1 -->
+			<supportedOS Id="{1f676c76-80e1-4239-95bb-83d0f6d0da78}"/>
+			<!-- Windows 10 -->
+			<supportedOS Id="{8e0f7a12-bfb3-4fe8-b9a5-48fd50a15a9a}"/>
+		</application>
+	</compatibility>
+</assembly>
diff --git a/compat/win32/lazyload.h b/compat/win32/lazyload.h
new file mode 100644
index 000000000000..9e631c8593ff
--- /dev/null
+++ b/compat/win32/lazyload.h
@@ -0,0 +1,57 @@
+#ifndef LAZYLOAD_H
+#define LAZYLOAD_H
+
+/*
+ * A pair of macros to simplify loading of DLL functions. Example:
+ *
+ *   DECLARE_PROC_ADDR(kernel32.dll, BOOL, CreateHardLinkW,
+ *                     LPCWSTR, LPCWSTR, LPSECURITY_ATTRIBUTES);
+ *
+ *   if (!INIT_PROC_ADDR(CreateHardLinkW))
+ *           return error("Could not find CreateHardLinkW() function";
+ *
+ *   if (!CreateHardLinkW(source, target, NULL))
+ *           return error("could not create hardlink from %S to %S",
+ *                        source, target);
+ */
+
+struct proc_addr {
+	const char *const dll;
+	const char *const function;
+	FARPROC pfunction;
+	unsigned initialized : 1;
+};
+
+/* Declares a function to be loaded dynamically from a DLL. */
+#define DECLARE_PROC_ADDR(dll, rettype, function, ...) \
+	static struct proc_addr proc_addr_##function = \
+	{ #dll, #function, NULL, 0 }; \
+	static rettype (WINAPI *function)(__VA_ARGS__)
+
+/*
+ * Loads a function from a DLL (once-only).
+ * Returns non-NULL function pointer on success.
+ * Returns NULL + errno == ENOSYS on failure.
+ * This function is not thread-safe.
+ */
+#define INIT_PROC_ADDR(function) \
+	(function = get_proc_addr(&proc_addr_##function))
+
+static inline void *get_proc_addr(struct proc_addr *proc)
+{
+	/* only do this once */
+	if (!proc->initialized) {
+		HANDLE hnd;
+		proc->initialized = 1;
+		hnd = LoadLibraryExA(proc->dll, NULL,
+				     LOAD_LIBRARY_SEARCH_SYSTEM32);
+		if (hnd)
+			proc->pfunction = GetProcAddress(hnd, proc->function);
+	}
+	/* set ENOSYS if DLL or function was not found */
+	if (!proc->pfunction)
+		errno = ENOSYS;
+	return proc->pfunction;
+}
+
+#endif
diff --git a/compat/win32/path-utils.c b/compat/win32/path-utils.c
new file mode 100644
index 000000000000..d9d3641de857
--- /dev/null
+++ b/compat/win32/path-utils.c
@@ -0,0 +1,28 @@
+#include "../../git-compat-util.h"
+
+int win32_skip_dos_drive_prefix(char **path)
+{
+	int ret = has_dos_drive_prefix(*path);
+	*path += ret;
+	return ret;
+}
+
+int win32_offset_1st_component(const char *path)
+{
+	char *pos = (char *)path;
+
+	/* unc paths */
+	if (!skip_dos_drive_prefix(&pos) &&
+			is_dir_sep(pos[0]) && is_dir_sep(pos[1])) {
+		/* skip server name */
+		pos = strpbrk(pos + 2, "\\/");
+		if (!pos)
+			return 0; /* Error: malformed unc path */
+
+		do {
+			pos++;
+		} while (*pos && !is_dir_sep(*pos));
+	}
+
+	return pos + is_dir_sep(*pos) - path;
+}
diff --git a/compat/win32/path-utils.h b/compat/win32/path-utils.h
new file mode 100644
index 000000000000..0f70d439204f
--- /dev/null
+++ b/compat/win32/path-utils.h
@@ -0,0 +1,20 @@
+#define has_dos_drive_prefix(path) \
+	(isalpha(*(path)) && (path)[1] == ':' ? 2 : 0)
+int win32_skip_dos_drive_prefix(char **path);
+#define skip_dos_drive_prefix win32_skip_dos_drive_prefix
+static inline int win32_is_dir_sep(int c)
+{
+	return c == '/' || c == '\\';
+}
+#define is_dir_sep win32_is_dir_sep
+static inline char *win32_find_last_dir_sep(const char *path)
+{
+	char *ret = NULL;
+	for (; *path; ++path)
+		if (is_dir_sep(*path))
+			ret = (char *)path;
+	return ret;
+}
+#define find_last_dir_sep win32_find_last_dir_sep
+int win32_offset_1st_component(const char *path);
+#define offset_1st_component win32_offset_1st_component
diff --git a/compat/win32/pthread.c b/compat/win32/pthread.c
new file mode 100644
index 000000000000..2e7eead42cb0
--- /dev/null
+++ b/compat/win32/pthread.c
@@ -0,0 +1,58 @@
+/*
+ * Copyright (C) 2009 Andrzej K. Haczewski <ahaczewski@gmail.com>
+ *
+ * DISCLAIMER: The implementation is Git-specific, it is subset of original
+ * Pthreads API, without lots of other features that Git doesn't use.
+ * Git also makes sure that the passed arguments are valid, so there's
+ * no need for double-checking.
+ */
+
+#include "../../git-compat-util.h"
+#include "pthread.h"
+
+#include <errno.h>
+#include <limits.h>
+
+static unsigned __stdcall win32_start_routine(void *arg)
+{
+	pthread_t *thread = arg;
+	thread->tid = GetCurrentThreadId();
+	thread->arg = thread->start_routine(thread->arg);
+	return 0;
+}
+
+int pthread_create(pthread_t *thread, const void *unused,
+		   void *(*start_routine)(void*), void *arg)
+{
+	thread->arg = arg;
+	thread->start_routine = start_routine;
+	thread->handle = (HANDLE)
+		_beginthreadex(NULL, 0, win32_start_routine, thread, 0, NULL);
+
+	if (!thread->handle)
+		return errno;
+	else
+		return 0;
+}
+
+int win32_pthread_join(pthread_t *thread, void **value_ptr)
+{
+	DWORD result = WaitForSingleObject(thread->handle, INFINITE);
+	switch (result) {
+		case WAIT_OBJECT_0:
+			if (value_ptr)
+				*value_ptr = thread->arg;
+			return 0;
+		case WAIT_ABANDONED:
+			return EINVAL;
+		default:
+			return err_win_to_posix(GetLastError());
+	}
+}
+
+pthread_t pthread_self(void)
+{
+	pthread_t t = { NULL };
+	t.tid = GetCurrentThreadId();
+	return t;
+}
diff --git a/compat/win32/pthread.h b/compat/win32/pthread.h
new file mode 100644
index 000000000000..c6cb8dd21900
--- /dev/null
+++ b/compat/win32/pthread.h
@@ -0,0 +1,100 @@
+/*
+ * Header used to adapt pthread-based POSIX code to Windows API threads.
+ *
+ * Copyright (C) 2009 Andrzej K. Haczewski <ahaczewski@gmail.com>
+ */
+
+#ifndef PTHREAD_H
+#define PTHREAD_H
+
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif
+
+#include <windows.h>
+
+/*
+ * Defines that adapt Windows API threads to pthreads API
+ */
+#define pthread_mutex_t CRITICAL_SECTION
+
+static inline int return_0(int i) {
+	return 0;
+}
+#define pthread_mutex_init(a,b) return_0((InitializeCriticalSection((a)), 0))
+#define pthread_mutex_destroy(a) DeleteCriticalSection((a))
+#define pthread_mutex_lock EnterCriticalSection
+#define pthread_mutex_unlock LeaveCriticalSection
+
+typedef int pthread_mutexattr_t;
+#define pthread_mutexattr_init(a) (*(a) = 0)
+#define pthread_mutexattr_destroy(a) do {} while (0)
+#define pthread_mutexattr_settype(a, t) 0
+#define PTHREAD_MUTEX_RECURSIVE 0
+
+#define pthread_cond_t CONDITION_VARIABLE
+
+#define pthread_cond_init(a,b) InitializeConditionVariable((a))
+#define pthread_cond_destroy(a) do {} while (0)
+#define pthread_cond_wait(a,b) return_0(SleepConditionVariableCS((a), (b), INFINITE))
+#define pthread_cond_signal WakeConditionVariable
+#define pthread_cond_broadcast WakeAllConditionVariable
+
+/*
+ * Simple thread creation implementation using pthread API
+ */
+typedef struct {
+	HANDLE handle;
+	void *(*start_routine)(void*);
+	void *arg;
+	DWORD tid;
+} pthread_t;
+
+extern int pthread_create(pthread_t *thread, const void *unused,
+			  void *(*start_routine)(void*), void *arg);
+
+/*
+ * To avoid the need of copying a struct, we use small macro wrapper to pass
+ * pointer to win32_pthread_join instead.
+ */
+#define pthread_join(a, b) win32_pthread_join(&(a), (b))
+
+extern int win32_pthread_join(pthread_t *thread, void **value_ptr);
+
+#define pthread_equal(t1, t2) ((t1).tid == (t2).tid)
+extern pthread_t pthread_self(void);
+
+static inline void NORETURN pthread_exit(void *ret)
+{
+	ExitThread((DWORD)(intptr_t)ret);
+}
+
+typedef DWORD pthread_key_t;
+static inline int pthread_key_create(pthread_key_t *keyp, void (*destructor)(void *value))
+{
+	return (*keyp = TlsAlloc()) == TLS_OUT_OF_INDEXES ? EAGAIN : 0;
+}
+
+static inline int pthread_key_delete(pthread_key_t key)
+{
+	return TlsFree(key) ? 0 : EINVAL;
+}
+
+static inline int pthread_setspecific(pthread_key_t key, const void *value)
+{
+	return TlsSetValue(key, (void *)value) ? 0 : EINVAL;
+}
+
+static inline void *pthread_getspecific(pthread_key_t key)
+{
+	return TlsGetValue(key);
+}
+
+#ifndef __MINGW64_VERSION_MAJOR
+static inline int pthread_sigmask(int how, const sigset_t *set, sigset_t *oset)
+{
+	return 0;
+}
+#endif
+
+#endif /* PTHREAD_H */
diff --git a/compat/win32/syslog.c b/compat/win32/syslog.c
new file mode 100644
index 000000000000..161978d720ae
--- /dev/null
+++ b/compat/win32/syslog.c
@@ -0,0 +1,80 @@
+#include "../../git-compat-util.h"
+
+static HANDLE ms_eventlog;
+
+void openlog(const char *ident, int logopt, int facility)
+{
+	if (ms_eventlog)
+		return;
+
+	ms_eventlog = RegisterEventSourceA(NULL, ident);
+
+	if (!ms_eventlog)
+		warning("RegisterEventSource() failed: %lu", GetLastError());
+}
+
+void syslog(int priority, const char *fmt, ...)
+{
+	WORD logtype;
+	char *str, *pos;
+	int str_len;
+	va_list ap;
+
+	if (!ms_eventlog)
+		return;
+
+	va_start(ap, fmt);
+	str_len = vsnprintf(NULL, 0, fmt, ap);
+	va_end(ap);
+
+	if (str_len < 0) {
+		warning_errno("vsnprintf failed");
+		return;
+	}
+
+	str = malloc(st_add(str_len, 1));
+	if (!str) {
+		warning_errno("malloc failed");
+		return;
+	}
+
+	va_start(ap, fmt);
+	vsnprintf(str, str_len + 1, fmt, ap);
+	va_end(ap);
+
+	while ((pos = strstr(str, "%1")) != NULL) {
+		char *oldstr = str;
+		str = realloc(str, st_add(++str_len, 1));
+		if (!str) {
+			free(oldstr);
+			warning_errno("realloc failed");
+			return;
+		}
+		memmove(pos + 2, pos + 1, strlen(pos));
+		pos[1] = ' ';
+	}
+
+	switch (priority) {
+	case LOG_EMERG:
+	case LOG_ALERT:
+	case LOG_CRIT:
+	case LOG_ERR:
+		logtype = EVENTLOG_ERROR_TYPE;
+		break;
+
+	case LOG_WARNING:
+		logtype = EVENTLOG_WARNING_TYPE;
+		break;
+
+	case LOG_NOTICE:
+	case LOG_INFO:
+	case LOG_DEBUG:
+	default:
+		logtype = EVENTLOG_INFORMATION_TYPE;
+		break;
+	}
+
+	ReportEventA(ms_eventlog, logtype, 0, 0, NULL, 1, 0,
+	    (const char **)&str, NULL);
+	free(str);
+}
diff --git a/compat/win32/syslog.h b/compat/win32/syslog.h
new file mode 100644
index 000000000000..70daa7c08b8d
--- /dev/null
+++ b/compat/win32/syslog.h
@@ -0,0 +1,20 @@
+#ifndef SYSLOG_H
+#define SYSLOG_H
+
+#define LOG_PID     0x01
+
+#define LOG_EMERG   0
+#define LOG_ALERT   1
+#define LOG_CRIT    2
+#define LOG_ERR     3
+#define LOG_WARNING 4
+#define LOG_NOTICE  5
+#define LOG_INFO    6
+#define LOG_DEBUG   7
+
+#define LOG_DAEMON  (3<<3)
+
+void openlog(const char *ident, int logopt, int facility);
+void syslog(int priority, const char *fmt, ...);
+
+#endif /* SYSLOG_H */
diff --git a/compat/win32/trace2_win32_process_info.c b/compat/win32/trace2_win32_process_info.c
new file mode 100644
index 000000000000..8ccbd1c2c6f8
--- /dev/null
+++ b/compat/win32/trace2_win32_process_info.c
@@ -0,0 +1,191 @@
+#include "../../cache.h"
+#include "../../json-writer.h"
+#include "lazyload.h"
+#include <Psapi.h>
+#include <tlHelp32.h>
+
+/*
+ * An arbitrarily chosen value to limit the size of the ancestor
+ * array built in git_processes().
+ */
+#define NR_PIDS_LIMIT 10
+
+/*
+ * Find the process data for the given PID in the given snapshot
+ * and update the PROCESSENTRY32 data.
+ */
+static int find_pid(DWORD pid, HANDLE hSnapshot, PROCESSENTRY32 *pe32)
+{
+	pe32->dwSize = sizeof(PROCESSENTRY32);
+
+	if (Process32First(hSnapshot, pe32)) {
+		do {
+			if (pe32->th32ProcessID == pid)
+				return 1;
+		} while (Process32Next(hSnapshot, pe32));
+	}
+	return 0;
+}
+
+/*
+ * Accumulate JSON array of our parent processes:
+ *     [
+ *         exe-name-parent,
+ *         exe-name-grand-parent,
+ *         ...
+ *     ]
+ *
+ * Note: we only report the filename of the process executable; the
+ *       only way to get its full pathname is to use OpenProcess()
+ *       and GetModuleFileNameEx() or QueryfullProcessImageName()
+ *       and that seems rather expensive (on top of the cost of
+ *       getting the snapshot).
+ *
+ * Note: we compute the set of parent processes by walking the PPID
+ *       link in each visited PROCESSENTRY32 record.  This search
+ *       stops when an ancestor process is not found in the snapshot
+ *       (because it exited before the current or intermediate parent
+ *       process exited).
+ *
+ *       This search may compute an incorrect result if the PPID link
+ *       refers to the PID of an exited parent and that PID has been
+ *       recycled and given to a new unrelated process.
+ *
+ *       Worse, it is possible for a child or descendant of the
+ *       current process to be given the recycled PID and cause a
+ *       PPID-cycle.  This would cause an infinite loop building our
+ *       parent process array.
+ *
+ * Note: for completeness, the "System Idle" process has PID=0 and
+ *       PPID=0 and could cause another PPID-cycle.  We don't expect
+ *       Git to be a descendant of the idle process, but because of
+ *       PID recycling, it might be possible to get a PPID link value
+ *       of 0.  This too would cause an infinite loop.
+ *
+ * Therefore, we keep an array of the visited PPIDs to guard against
+ * cycles.
+ *
+ * We use a fixed-size array rather than ALLOC_GROW to keep things
+ * simple and avoid the alloc/realloc overhead.  It is OK if we
+ * truncate the search and return a partial answer.
+ */
+static void get_processes(struct json_writer *jw, HANDLE hSnapshot)
+{
+	PROCESSENTRY32 pe32;
+	DWORD pid;
+	DWORD pid_list[NR_PIDS_LIMIT];
+	int k, nr_pids = 0;
+
+	pid = GetCurrentProcessId();
+	while (find_pid(pid, hSnapshot, &pe32)) {
+		/* Only report parents. Omit self from the JSON output. */
+		if (nr_pids)
+			jw_array_string(jw, pe32.szExeFile);
+
+		/* Check for cycle in snapshot. (Yes, it happened.) */
+		for (k = 0; k < nr_pids; k++)
+			if (pid == pid_list[k]) {
+				jw_array_string(jw, "(cycle)");
+				return;
+			}
+
+		if (nr_pids == NR_PIDS_LIMIT) {
+			jw_array_string(jw, "(truncated)");
+			return;
+		}
+
+		pid_list[nr_pids++] = pid;
+
+		pid = pe32.th32ParentProcessID;
+	}
+}
+
+/*
+ * Emit JSON data for the current and parent processes.  Individual
+ * trace2 targets can decide how to actually print it.
+ */
+static void get_ancestry(void)
+{
+	HANDLE hSnapshot = CreateToolhelp32Snapshot(TH32CS_SNAPPROCESS, 0);
+
+	if (hSnapshot != INVALID_HANDLE_VALUE) {
+		struct json_writer jw = JSON_WRITER_INIT;
+
+		jw_array_begin(&jw, 0);
+		get_processes(&jw, hSnapshot);
+		jw_end(&jw);
+
+		trace2_data_json("process", the_repository, "windows/ancestry",
+				 &jw);
+
+		jw_release(&jw);
+		CloseHandle(hSnapshot);
+	}
+}
+
+/*
+ * Is a debugger attached to the current process?
+ *
+ * This will catch debug runs (where the debugger started the process).
+ * This is the normal case.  Since this code is called during our startup,
+ * it will not report instances where a debugger is attached dynamically
+ * to a running git process, but that is relatively rare.
+ */
+static void get_is_being_debugged(void)
+{
+	if (IsDebuggerPresent())
+		trace2_data_intmax("process", the_repository,
+				   "windows/debugger_present", 1);
+}
+
+/*
+ * Emit JSON data with the peak memory usage of the current process.
+ */
+static void get_peak_memory_info(void)
+{
+	DECLARE_PROC_ADDR(psapi.dll, BOOL, GetProcessMemoryInfo, HANDLE,
+			  PPROCESS_MEMORY_COUNTERS, DWORD);
+
+	if (INIT_PROC_ADDR(GetProcessMemoryInfo)) {
+		PROCESS_MEMORY_COUNTERS pmc;
+
+		if (GetProcessMemoryInfo(GetCurrentProcess(), &pmc,
+					 sizeof(pmc))) {
+			struct json_writer jw = JSON_WRITER_INIT;
+
+			jw_object_begin(&jw, 0);
+
+#define KV(kv) #kv, (intmax_t)pmc.kv
+
+			jw_object_intmax(&jw, KV(PageFaultCount));
+			jw_object_intmax(&jw, KV(PeakWorkingSetSize));
+			jw_object_intmax(&jw, KV(PeakPagefileUsage));
+
+			jw_end(&jw);
+
+			trace2_data_json("process", the_repository,
+					 "windows/memory", &jw);
+			jw_release(&jw);
+		}
+	}
+}
+
+void trace2_collect_process_info(enum trace2_process_info_reason reason)
+{
+	if (!trace2_is_enabled())
+		return;
+
+	switch (reason) {
+	case TRACE2_PROCESS_INFO_STARTUP:
+		get_is_being_debugged();
+		get_ancestry();
+		return;
+
+	case TRACE2_PROCESS_INFO_EXIT:
+		get_peak_memory_info();
+		return;
+
+	default:
+		BUG("trace2_collect_process_info: unknown reason '%d'", reason);
+	}
+}
diff --git a/compat/win32mmap.c b/compat/win32mmap.c
new file mode 100644
index 000000000000..519d51f2b60a
--- /dev/null
+++ b/compat/win32mmap.c
@@ -0,0 +1,46 @@
+#include "../git-compat-util.h"
+
+void *git_mmap(void *start, size_t length, int prot, int flags, int fd, off_t offset)
+{
+	HANDLE osfhandle, hmap;
+	void *temp;
+	LARGE_INTEGER len;
+	uint64_t o = offset;
+	uint32_t l = o & 0xFFFFFFFF;
+	uint32_t h = (o >> 32) & 0xFFFFFFFF;
+
+	osfhandle = (HANDLE)_get_osfhandle(fd);
+	if (!GetFileSizeEx(osfhandle, &len))
+		die("mmap: could not determine filesize");
+
+	if ((length + offset) > len.QuadPart)
+		length = xsize_t(len.QuadPart - offset);
+
+	if (!(flags & MAP_PRIVATE))
+		die("Invalid usage of mmap when built with USE_WIN32_MMAP");
+
+	hmap = CreateFileMapping(osfhandle, NULL,
+		prot == PROT_READ ? PAGE_READONLY : PAGE_WRITECOPY, 0, 0, NULL);
+
+	if (!hmap) {
+		errno = EINVAL;
+		return MAP_FAILED;
+	}
+
+	temp = MapViewOfFileEx(hmap, prot == PROT_READ ?
+			FILE_MAP_READ : FILE_MAP_COPY, h, l, length, start);
+
+	if (!CloseHandle(hmap))
+		warning("unable to close file mapping handle");
+
+	if (temp)
+		return temp;
+
+	errno = GetLastError() == ERROR_COMMITMENT_LIMIT ? EFBIG : EINVAL;
+	return MAP_FAILED;
+}
+
+int git_munmap(void *start, size_t length)
+{
+	return !UnmapViewOfFile(start);
+}
diff --git a/compat/winansi.c b/compat/winansi.c
new file mode 100644
index 000000000000..cacd82c833a6
--- /dev/null
+++ b/compat/winansi.c
@@ -0,0 +1,671 @@
+/*
+ * Copyright 2008 Peter Harris <git@peter.is-a-geek.org>
+ */
+
+#undef NOGDI
+#include "../git-compat-util.h"
+#include <wingdi.h>
+#include <winreg.h>
+#include "win32.h"
+#include "win32/lazyload.h"
+
+static int fd_is_interactive[3] = { 0, 0, 0 };
+#define FD_CONSOLE 0x1
+#define FD_SWAPPED 0x2
+#define FD_MSYS    0x4
+
+/*
+ ANSI codes used by git: m, K
+
+ This file is git-specific. Therefore, this file does not attempt
+ to implement any codes that are not used by git.
+*/
+
+static HANDLE console;
+static WORD plain_attr;
+static WORD attr;
+static int negative;
+static int non_ascii_used = 0;
+static HANDLE hthread, hread, hwrite;
+static HANDLE hconsole1, hconsole2;
+
+#ifdef __MINGW32__
+#if !defined(__MINGW64_VERSION_MAJOR) || __MINGW64_VERSION_MAJOR < 5
+typedef struct _CONSOLE_FONT_INFOEX {
+	ULONG cbSize;
+	DWORD nFont;
+	COORD dwFontSize;
+	UINT FontFamily;
+	UINT FontWeight;
+	WCHAR FaceName[LF_FACESIZE];
+} CONSOLE_FONT_INFOEX, *PCONSOLE_FONT_INFOEX;
+#endif
+#endif
+
+static void warn_if_raster_font(void)
+{
+	DWORD fontFamily = 0;
+	DECLARE_PROC_ADDR(kernel32.dll, BOOL, GetCurrentConsoleFontEx,
+			HANDLE, BOOL, PCONSOLE_FONT_INFOEX);
+
+	/* don't bother if output was ascii only */
+	if (!non_ascii_used)
+		return;
+
+	/* GetCurrentConsoleFontEx is available since Vista */
+	if (INIT_PROC_ADDR(GetCurrentConsoleFontEx)) {
+		CONSOLE_FONT_INFOEX cfi;
+		cfi.cbSize = sizeof(cfi);
+		if (GetCurrentConsoleFontEx(console, 0, &cfi))
+			fontFamily = cfi.FontFamily;
+	} else {
+		/* pre-Vista: check default console font in registry */
+		HKEY hkey;
+		if (ERROR_SUCCESS == RegOpenKeyExA(HKEY_CURRENT_USER, "Console",
+				0, KEY_READ, &hkey)) {
+			DWORD size = sizeof(fontFamily);
+			RegQueryValueExA(hkey, "FontFamily", NULL, NULL,
+					(LPVOID) &fontFamily, &size);
+			RegCloseKey(hkey);
+		}
+	}
+
+	if (!(fontFamily & TMPF_TRUETYPE)) {
+		const wchar_t *msg = L"\nWarning: Your console font probably "
+			L"doesn\'t support Unicode. If you experience strange "
+			L"characters in the output, consider switching to a "
+			L"TrueType font such as Consolas!\n";
+		DWORD dummy;
+		WriteConsoleW(console, msg, wcslen(msg), &dummy, NULL);
+	}
+}
+
+static int is_console(int fd)
+{
+	CONSOLE_SCREEN_BUFFER_INFO sbi;
+	DWORD mode;
+	HANDLE hcon;
+
+	static int initialized = 0;
+
+	/* get OS handle of the file descriptor */
+	hcon = (HANDLE) _get_osfhandle(fd);
+	if (hcon == INVALID_HANDLE_VALUE)
+		return 0;
+
+	/* check if its a device (i.e. console, printer, serial port) */
+	if (GetFileType(hcon) != FILE_TYPE_CHAR)
+		return 0;
+
+	/* check if its a handle to a console output screen buffer */
+	if (!fd) {
+		if (!GetConsoleMode(hcon, &mode))
+			return 0;
+		/*
+		 * This code path is only reached if there is no console
+		 * attached to stdout/stderr, i.e. we will not need to output
+		 * any text to any console, therefore we might just as well
+		 * use black as foreground color.
+		 */
+		sbi.wAttributes = 0;
+	} else if (!GetConsoleScreenBufferInfo(hcon, &sbi))
+		return 0;
+
+	if (fd >= 0 && fd <= 2)
+		fd_is_interactive[fd] |= FD_CONSOLE;
+
+	/* initialize attributes */
+	if (!initialized) {
+		console = hcon;
+		attr = plain_attr = sbi.wAttributes;
+		negative = 0;
+		initialized = 1;
+	}
+
+	return 1;
+}
+
+#define BUFFER_SIZE 4096
+#define MAX_PARAMS 16
+
+static void write_console(unsigned char *str, size_t len)
+{
+	/* only called from console_thread, so a static buffer will do */
+	static wchar_t wbuf[2 * BUFFER_SIZE + 1];
+	DWORD dummy;
+
+	/* convert utf-8 to utf-16 */
+	int wlen = xutftowcsn(wbuf, (char*) str, ARRAY_SIZE(wbuf), len);
+	if (wlen < 0) {
+		wchar_t *err = L"[invalid]";
+		WriteConsoleW(console, err, wcslen(err), &dummy, NULL);
+		return;
+	}
+
+	/* write directly to console */
+	WriteConsoleW(console, wbuf, wlen, &dummy, NULL);
+
+	/* remember if non-ascii characters are printed */
+	if (wlen != len)
+		non_ascii_used = 1;
+}
+
+#define FOREGROUND_ALL (FOREGROUND_RED | FOREGROUND_GREEN | FOREGROUND_BLUE)
+#define BACKGROUND_ALL (BACKGROUND_RED | BACKGROUND_GREEN | BACKGROUND_BLUE)
+
+static void set_console_attr(void)
+{
+	WORD attributes = attr;
+	if (negative) {
+		attributes &= ~FOREGROUND_ALL;
+		attributes &= ~BACKGROUND_ALL;
+
+		/* This could probably use a bitmask
+		   instead of a series of ifs */
+		if (attr & FOREGROUND_RED)
+			attributes |= BACKGROUND_RED;
+		if (attr & FOREGROUND_GREEN)
+			attributes |= BACKGROUND_GREEN;
+		if (attr & FOREGROUND_BLUE)
+			attributes |= BACKGROUND_BLUE;
+
+		if (attr & BACKGROUND_RED)
+			attributes |= FOREGROUND_RED;
+		if (attr & BACKGROUND_GREEN)
+			attributes |= FOREGROUND_GREEN;
+		if (attr & BACKGROUND_BLUE)
+			attributes |= FOREGROUND_BLUE;
+	}
+	SetConsoleTextAttribute(console, attributes);
+}
+
+static void erase_in_line(void)
+{
+	CONSOLE_SCREEN_BUFFER_INFO sbi;
+	DWORD dummy; /* Needed for Windows 7 (or Vista) regression */
+
+	if (!console)
+		return;
+
+	GetConsoleScreenBufferInfo(console, &sbi);
+	FillConsoleOutputCharacterA(console, ' ',
+		sbi.dwSize.X - sbi.dwCursorPosition.X, sbi.dwCursorPosition,
+		&dummy);
+}
+
+static void set_attr(char func, const int *params, int paramlen)
+{
+	int i;
+	switch (func) {
+	case 'm':
+		for (i = 0; i < paramlen; i++) {
+			switch (params[i]) {
+			case 0: /* reset */
+				attr = plain_attr;
+				negative = 0;
+				break;
+			case 1: /* bold */
+				attr |= FOREGROUND_INTENSITY;
+				break;
+			case 2:  /* faint */
+			case 22: /* normal */
+				attr &= ~FOREGROUND_INTENSITY;
+				break;
+			case 3:  /* italic */
+				/* Unsupported */
+				break;
+			case 4:  /* underline */
+			case 21: /* double underline */
+				/* Wikipedia says this flag does nothing */
+				/* Furthermore, mingw doesn't define this flag
+				attr |= COMMON_LVB_UNDERSCORE; */
+				break;
+			case 24: /* no underline */
+				/* attr &= ~COMMON_LVB_UNDERSCORE; */
+				break;
+			case 5:  /* slow blink */
+			case 6:  /* fast blink */
+				/* We don't have blink, but we do have
+				   background intensity */
+				attr |= BACKGROUND_INTENSITY;
+				break;
+			case 25: /* no blink */
+				attr &= ~BACKGROUND_INTENSITY;
+				break;
+			case 7:  /* negative */
+				negative = 1;
+				break;
+			case 27: /* positive */
+				negative = 0;
+				break;
+			case 8:  /* conceal */
+			case 28: /* reveal */
+				/* Unsupported */
+				break;
+			case 30: /* Black */
+				attr &= ~FOREGROUND_ALL;
+				break;
+			case 31: /* Red */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_RED;
+				break;
+			case 32: /* Green */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_GREEN;
+				break;
+			case 33: /* Yellow */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_RED | FOREGROUND_GREEN;
+				break;
+			case 34: /* Blue */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_BLUE;
+				break;
+			case 35: /* Magenta */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_RED | FOREGROUND_BLUE;
+				break;
+			case 36: /* Cyan */
+				attr &= ~FOREGROUND_ALL;
+				attr |= FOREGROUND_GREEN | FOREGROUND_BLUE;
+				break;
+			case 37: /* White */
+				attr |= FOREGROUND_RED |
+					FOREGROUND_GREEN |
+					FOREGROUND_BLUE;
+				break;
+			case 38: /* Unknown */
+				break;
+			case 39: /* reset */
+				attr &= ~FOREGROUND_ALL;
+				attr |= (plain_attr & FOREGROUND_ALL);
+				break;
+			case 40: /* Black */
+				attr &= ~BACKGROUND_ALL;
+				break;
+			case 41: /* Red */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_RED;
+				break;
+			case 42: /* Green */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_GREEN;
+				break;
+			case 43: /* Yellow */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_RED | BACKGROUND_GREEN;
+				break;
+			case 44: /* Blue */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_BLUE;
+				break;
+			case 45: /* Magenta */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_RED | BACKGROUND_BLUE;
+				break;
+			case 46: /* Cyan */
+				attr &= ~BACKGROUND_ALL;
+				attr |= BACKGROUND_GREEN | BACKGROUND_BLUE;
+				break;
+			case 47: /* White */
+				attr |= BACKGROUND_RED |
+					BACKGROUND_GREEN |
+					BACKGROUND_BLUE;
+				break;
+			case 48: /* Unknown */
+				break;
+			case 49: /* reset */
+				attr &= ~BACKGROUND_ALL;
+				attr |= (plain_attr & BACKGROUND_ALL);
+				break;
+			default:
+				/* Unsupported code */
+				break;
+			}
+		}
+		set_console_attr();
+		break;
+	case 'K':
+		erase_in_line();
+		break;
+	default:
+		/* Unsupported code */
+		break;
+	}
+}
+
+enum {
+	TEXT = 0, ESCAPE = 033, BRACKET = '['
+};
+
+static DWORD WINAPI console_thread(LPVOID unused)
+{
+	unsigned char buffer[BUFFER_SIZE];
+	DWORD bytes;
+	int start, end = 0, c, parampos = 0, state = TEXT;
+	int params[MAX_PARAMS];
+
+	while (1) {
+		/* read next chunk of bytes from the pipe */
+		if (!ReadFile(hread, buffer + end, BUFFER_SIZE - end, &bytes,
+				NULL)) {
+			/* exit if pipe has been closed or disconnected */
+			if (GetLastError() == ERROR_PIPE_NOT_CONNECTED ||
+					GetLastError() == ERROR_BROKEN_PIPE)
+				break;
+			/* ignore other errors */
+			continue;
+		}
+
+		/* scan the bytes and handle ANSI control codes */
+		bytes += end;
+		start = end = 0;
+		while (end < bytes) {
+			c = buffer[end++];
+			switch (state) {
+			case TEXT:
+				if (c == ESCAPE) {
+					/* print text seen so far */
+					if (end - 1 > start)
+						write_console(buffer + start,
+							end - 1 - start);
+
+					/* then start parsing escape sequence */
+					start = end - 1;
+					memset(params, 0, sizeof(params));
+					parampos = 0;
+					state = ESCAPE;
+				}
+				break;
+
+			case ESCAPE:
+				/* continue if "\033[", otherwise bail out */
+				state = (c == BRACKET) ? BRACKET : TEXT;
+				break;
+
+			case BRACKET:
+				/* parse [0-9;]* into array of parameters */
+				if (c >= '0' && c <= '9') {
+					params[parampos] *= 10;
+					params[parampos] += c - '0';
+				} else if (c == ';') {
+					/*
+					 * next parameter, bail out if out of
+					 * bounds
+					 */
+					parampos++;
+					if (parampos >= MAX_PARAMS)
+						state = TEXT;
+				} else {
+					/*
+					 * end of escape sequence, change
+					 * console attributes
+					 */
+					set_attr(c, params, parampos + 1);
+					start = end;
+					state = TEXT;
+				}
+				break;
+			}
+		}
+
+		/* print remaining text unless parsing an escape sequence */
+		if (state == TEXT && end > start) {
+			/* check for incomplete UTF-8 sequences and fix end */
+			if (buffer[end - 1] >= 0x80) {
+				if (buffer[end -1] >= 0xc0)
+					end--;
+				else if (end - 1 > start &&
+						buffer[end - 2] >= 0xe0)
+					end -= 2;
+				else if (end - 2 > start &&
+						buffer[end - 3] >= 0xf0)
+					end -= 3;
+			}
+
+			/* print remaining complete UTF-8 sequences */
+			if (end > start)
+				write_console(buffer + start, end - start);
+
+			/* move remaining bytes to the front */
+			if (end < bytes)
+				memmove(buffer, buffer + end, bytes - end);
+			end = bytes - end;
+		} else {
+			/* all data has been consumed, mark buffer empty */
+			end = 0;
+		}
+	}
+
+	/* check if the console font supports unicode */
+	warn_if_raster_font();
+
+	CloseHandle(hread);
+	return 0;
+}
+
+static void winansi_exit(void)
+{
+	/* flush all streams */
+	_flushall();
+
+	/* signal console thread to exit */
+	FlushFileBuffers(hwrite);
+	DisconnectNamedPipe(hwrite);
+
+	/* wait for console thread to copy remaining data */
+	WaitForSingleObject(hthread, INFINITE);
+
+	/* cleanup handles... */
+	CloseHandle(hwrite);
+	CloseHandle(hthread);
+}
+
+static void die_lasterr(const char *fmt, ...)
+{
+	va_list params;
+	va_start(params, fmt);
+	errno = err_win_to_posix(GetLastError());
+	die_errno(fmt, params);
+	va_end(params);
+}
+
+#undef dup2
+int winansi_dup2(int oldfd, int newfd)
+{
+	int ret = dup2(oldfd, newfd);
+
+	if (!ret && newfd >= 0 && newfd <= 2)
+		fd_is_interactive[newfd] = oldfd < 0 || oldfd > 2 ?
+			0 : fd_is_interactive[oldfd];
+
+	return ret;
+}
+
+static HANDLE duplicate_handle(HANDLE hnd)
+{
+	HANDLE hresult, hproc = GetCurrentProcess();
+	if (!DuplicateHandle(hproc, hnd, hproc, &hresult, 0, TRUE,
+			DUPLICATE_SAME_ACCESS))
+		die_lasterr("DuplicateHandle(%li) failed",
+			(long) (intptr_t) hnd);
+	return hresult;
+}
+
+static HANDLE swap_osfhnd(int fd, HANDLE new_handle)
+{
+	/*
+	 * Create a copy of the original handle associated with fd
+	 * because the original will get closed when we dup2().
+	 */
+	HANDLE handle = (HANDLE)_get_osfhandle(fd);
+	HANDLE duplicate = duplicate_handle(handle);
+
+	/* Create a temp fd associated with the already open "new_handle". */
+	int new_fd = _open_osfhandle((intptr_t)new_handle, O_BINARY);
+
+	assert((fd == 1) || (fd == 2));
+
+	/*
+	 * Use stock dup2() to re-bind fd to the new handle.  Note that
+	 * this will implicitly close(1) and close both fd=1 and the
+	 * originally associated handle.  It will open a new fd=1 and
+	 * call DuplicateHandle() on the handle associated with new_fd.
+	 * It is because of this implicit close() that we created the
+	 * copy of the original.
+	 *
+	 * Note that we need to update the cached console handle to the
+	 * duplicated one because the dup2() call will implicitly close
+	 * the original one.
+	 *
+	 * Note that dup2() when given target := {0,1,2} will also
+	 * call SetStdHandle(), so we don't need to worry about that.
+	 */
+	if (console == handle)
+		console = duplicate;
+	dup2(new_fd, fd);
+
+	/* Close the temp fd.  This explicitly closes "new_handle"
+	 * (because it has been associated with it).
+	 */
+	close(new_fd);
+
+	if (fd == 2)
+		setvbuf(stderr, NULL, _IONBF, BUFSIZ);
+	fd_is_interactive[fd] |= FD_SWAPPED;
+
+	return duplicate;
+}
+
+#ifdef DETECT_MSYS_TTY
+
+#include <winternl.h>
+
+#if defined(_MSC_VER)
+
+typedef struct _OBJECT_NAME_INFORMATION
+{
+	UNICODE_STRING Name;
+	WCHAR NameBuffer[0];
+} OBJECT_NAME_INFORMATION, *POBJECT_NAME_INFORMATION;
+
+#define ObjectNameInformation 1
+
+#else
+#include <ntstatus.h>
+#endif
+
+static void detect_msys_tty(int fd)
+{
+	ULONG result;
+	BYTE buffer[1024];
+	POBJECT_NAME_INFORMATION nameinfo = (POBJECT_NAME_INFORMATION) buffer;
+	PWSTR name;
+
+	/* check if fd is a pipe */
+	HANDLE h = (HANDLE) _get_osfhandle(fd);
+	if (GetFileType(h) != FILE_TYPE_PIPE)
+		return;
+
+	/* get pipe name */
+	if (!NT_SUCCESS(NtQueryObject(h, ObjectNameInformation,
+			buffer, sizeof(buffer) - 2, &result)))
+		return;
+	name = nameinfo->Name.Buffer;
+	name[nameinfo->Name.Length / sizeof(*name)] = 0;
+
+	/*
+	 * Check if this could be a MSYS2 pty pipe ('msys-XXXX-ptyN-XX')
+	 * or a cygwin pty pipe ('cygwin-XXXX-ptyN-XX')
+	 */
+	if ((!wcsstr(name, L"msys-") && !wcsstr(name, L"cygwin-")) ||
+			!wcsstr(name, L"-pty"))
+		return;
+
+	if (fd == 2)
+		setvbuf(stderr, NULL, _IONBF, BUFSIZ);
+	fd_is_interactive[fd] |= FD_MSYS;
+}
+
+#endif
+
+/*
+ * Wrapper for isatty().  Most calls in the main git code
+ * call isatty(1 or 2) to see if the instance is interactive
+ * and should: be colored, show progress, paginate output.
+ * We lie and give results for what the descriptor WAS at
+ * startup (and ignore any pipe redirection we internally
+ * do).
+ */
+#undef isatty
+int winansi_isatty(int fd)
+{
+	if (fd >= 0 && fd <= 2)
+		return fd_is_interactive[fd] != 0;
+	return isatty(fd);
+}
+
+void winansi_init(void)
+{
+	int con1, con2;
+	wchar_t name[32];
+
+	/* check if either stdout or stderr is a console output screen buffer */
+	con1 = is_console(1);
+	con2 = is_console(2);
+
+	/* Also compute console bit for fd 0 even though we don't need the result here. */
+	is_console(0);
+
+	if (!con1 && !con2) {
+#ifdef DETECT_MSYS_TTY
+		/* check if stdin / stdout / stderr are MSYS2 pty pipes */
+		detect_msys_tty(0);
+		detect_msys_tty(1);
+		detect_msys_tty(2);
+#endif
+		return;
+	}
+
+	/* create a named pipe to communicate with the console thread */
+	if (swprintf(name, ARRAY_SIZE(name) - 1, L"\\\\.\\pipe\\winansi%lu",
+		     GetCurrentProcessId()) < 0)
+		die("Could not initialize winansi pipe name");
+	hwrite = CreateNamedPipeW(name, PIPE_ACCESS_OUTBOUND,
+		PIPE_TYPE_BYTE | PIPE_WAIT, 1, BUFFER_SIZE, 0, 0, NULL);
+	if (hwrite == INVALID_HANDLE_VALUE)
+		die_lasterr("CreateNamedPipe failed");
+
+	hread = CreateFileW(name, GENERIC_READ, 0, NULL, OPEN_EXISTING, 0, NULL);
+	if (hread == INVALID_HANDLE_VALUE)
+		die_lasterr("CreateFile for named pipe failed");
+
+	/* start console spool thread on the pipe's read end */
+	hthread = CreateThread(NULL, 0, console_thread, NULL, 0, NULL);
+	if (hthread == INVALID_HANDLE_VALUE)
+		die_lasterr("CreateThread(console_thread) failed");
+
+	/* schedule cleanup routine */
+	if (atexit(winansi_exit))
+		die_errno("atexit(winansi_exit) failed");
+
+	/* redirect stdout / stderr to the pipe */
+	if (con1)
+		hconsole1 = swap_osfhnd(1, duplicate_handle(hwrite));
+	if (con2)
+		hconsole2 = swap_osfhnd(2, duplicate_handle(hwrite));
+}
+
+/*
+ * Returns the real console handle if stdout / stderr is a pipe redirecting
+ * to the console. Allows spawn / exec to pass the console to the next process.
+ */
+HANDLE winansi_get_osfhandle(int fd)
+{
+	if (fd == 1 && (fd_is_interactive[1] & FD_SWAPPED))
+		return hconsole1;
+	if (fd == 2 && (fd_is_interactive[2] & FD_SWAPPED))
+		return hconsole2;
+
+	return (HANDLE)_get_osfhandle(fd);
+}