Export of internal Abseil changes

--
f13697e3d33803f9667d124072da4f6dd8bfbf85 by Andy Soffer <asoffer@google.com>:

Addressing https://github.com/abseil/abseil-cpp/issues/314, fixing
CMakeLists.txt to reference ABSL_TEST_COPTS rather than ABSL_DEFAULT_COPTS.

ABSL_TEST_COPTS should be preferred for all tests so that they are configured consistently (moreover, CMake should agree with Bazel).

PiperOrigin-RevId: 274932312

--
c31c24a1fa6bb98136adf51ef37c0818ac366690 by Derek Mauro <dmauro@google.com>:

Silence MSAN in the stack consumption test utility

PiperOrigin-RevId: 274912950

--
2412913c05a246cd527cd4c31452f126e9129f3a by CJ Johnson <johnsoncj@google.com>:

Internal change

PiperOrigin-RevId: 274847103

--
75e984a93b5760873501b96ac3229ccfd955daf8 by Abseil Team <absl-team@google.com>:

Reformat BUILD file to current standards.

PiperOrigin-RevId: 274815392

--
a2780e085f1df1e4ca2c814a58c893d1b78a1d9c by Samuel Benzaquen <sbenza@google.com>:

Fix invalid result regarding leading zeros in the exponent.

PiperOrigin-RevId: 274808017

--
dd402e1cb5c4ebacb576372ae24bf289d729d323 by Samuel Benzaquen <sbenza@google.com>:

Make string_view's relational operators constexpr when possible.

PiperOrigin-RevId: 274807873

--
b4ef32565653a5da1cb8bb8d0351586d23519658 by Abseil Team <absl-team@google.com>:

Internal rework.

PiperOrigin-RevId: 274787159

--
70d81971c5914e6785b8e8a9d4f6eb2655dd62c0 by Gennadiy Rozental <rogeeff@google.com>:

Internal rework.

PiperOrigin-RevId: 274715557

--
14f5b0440e353b899cafaaa15b53e77f98f401af by Gennadiy Rozental <rogeeff@google.com>:

Make deprecated statements about ParseFLag/UnparseFlag consistent in a file.

PiperOrigin-RevId: 274668123

--
2e85adbdbb92612e4d750bc34fbca3333128b42d by Abseil Team <absl-team@google.com>:

Allow absl::c_equal to be used with arrays.

This is achieved by allowing container size computation for arrays.

PiperOrigin-RevId: 274426830

--
219719f107226d328773e6cec99fb473f5d3119c by Gennadiy Rozental <rogeeff@google.com>:

Release correct extension interfaces to support usage of absl::Time and absl::Duration as ABSL_FLAG

PiperOrigin-RevId: 274273788

--
47a77f93fda23b69b4a6bdbd506fe643c69a5579 by Gennadiy Rozental <rogeeff@google.com>:

Rework of flags persistence/FlagSaver internals.

PiperOrigin-RevId: 274225213

--
7807be3fe757c19e3b0c487298387683d4c9f5b3 by Abseil Team <absl-team@google.com>:

Switch reference to sdkddkver.h to lowercase, matching conventions used in the Windows SDK and other uses. This helps to avoid confusion on case-sensitive filesystems.

PiperOrigin-RevId: 274061877

--
561304090087a19f1d10f0475f564fe132ebf06e by Andy Getzendanner <durandal@google.com>:

Fix ABSL_WAITER_MODE detection for mingw

Import of https://github.com/abseil/abseil-cpp/pull/342

PiperOrigin-RevId: 274030071

--
9b3caac2cf202b9d440dfa1b4ffd538ac4bf715b by Derek Mauro <dmauro@google.com>:

Support using Abseil with the musl libc implementation.

Only test changes were required:
  * Workaround for a bug in sigaltstack() on musl
  * printf-style pointer formatting (%p) is implementation defined,
    so verify StrFromat produces something compatible
  * Fix detection of feenableexcept()

PiperOrigin-RevId: 274011666

--
73e8a938fc139e1cc8670d4513a445bacc855539 by Abseil Team <absl-team@google.com>:

nvcc workaround: explicitly specify the definition of node_handle::Base

PiperOrigin-RevId: 274011392

--
ab9cc6d042aca7d48e16c504ab10eab39433f4b2 by Andy Soffer <asoffer@google.com>:

Internal change

PiperOrigin-RevId: 273996318

--
e567c4979ca99c7e71821ec1523b8f5edd2c76ac by Abseil Team <absl-team@google.com>:

Introduce a type alias to work around an nvcc bug.

On the previous code, nvcc gets confused thinking that T has to be a parameter
pack, as IsDecomposable accepts one.

PiperOrigin-RevId: 273980472

--
105b6e6339b77a32f4432de05f44cd3f9c436751 by Eric Fiselier <ericwf@google.com>:

Import of CCTZ from GitHub.

PiperOrigin-RevId: 273955589

--
8feb87ff1d7e721fe094855e67c19539d5e582b7 by Abseil Team <absl-team@google.com>:

Avoid dual-exporting scheduling_mode.h

PiperOrigin-RevId: 273825112

--
fbc37854776d295dae98fb9d06a541f296daab95 by Andy Getzendanner <durandal@google.com>:

Fix ABSL_HAVE_ALARM check on mingw

Import of https://github.com/abseil/abseil-cpp/pull/341

PiperOrigin-RevId: 273817839

--
6aedcd63a735b9133e143b043744ba0a25407f6f by Andy Soffer <asoffer@google.com>:

Remove bit_gen_view.h now that all callers have been migrated to bit_gen_ref.h
Tested:
TGP - https://test.corp.google.com/ui#id=OCL:273762409:BASE:273743370:1570639020744:3001bcb5

PiperOrigin-RevId: 273810331

--
6573de24a66ba715c579f7f32b5c48a1d743c7f8 by Abseil Team <absl-team@google.com>:

Internal change.

PiperOrigin-RevId: 273589963

--
91c8c28b6dca26d98b39e8e06a8ed17c701ff793 by Abseil Team <absl-team@google.com>:

Update macro name for `ABSL_GUARDED_BY()` in the example section.

PiperOrigin-RevId: 273286983

--
0ff7d1a93d70f8ecd693f8dbb98b7a4a016ca2a4 by Abseil Team <absl-team@google.com>:

Fix potential integer overflow in the absl time library.

In absl::FromTM, the tm.tm_year is added by 1900 regarding that tm.tm_year represents the years since 1900. This change checks integer overflow before doing the arithmetic operation.

PiperOrigin-RevId: 273092952

--
b41c2a1310086807be09a833099ae6d4009f037c by Gennadiy Rozental <rogeeff@google.com>:

Correctly Unlock the global mutex in case of concurrent flag initialization.

Fixes #386

PiperOrigin-RevId: 272979749

--
c53103e71b2a6063af3c6d4ff68aa2d8f9ae9e06 by Abseil Team <absl-team@google.com>:

Try to become idle only when there is no wakeup.

Immediately after waking up (when futex wait returns), the current thread tries
to become idle doing bunch of memory loads and a branch.  Problem is that there
is a good chance that we woke up due to a wakeup, especially for actively used
threads.  For such wakeups, calling MaybeBecomeIdle() would be a waste of
cycles.

Instead, call MaybeBecomeIdle() only when we are sure there is no wakeup.  For
idle threads the net effect should be the same.  For active, threads this will
be more efficient.

Moreover, since MaybeBecomeIdle() is called before waiting on the futex, the
current thread will try to become idle before sleeping.  This should result
in more accurate idleness and more efficient release of thread resources.

PiperOrigin-RevId: 272940381
GitOrigin-RevId: f13697e3d33803f9667d124072da4f6dd8bfbf85
Change-Id: I36de05aec12595183725652dd362dfa58fb095d0

1 files changed, 9 insertions, 2 deletions

diff --git a/absl/time/internal/cctz/src/time_zone_impl.cc b/absl/time/internal/cctz/src/time_zone_impl.cc
index a26151d3e917..a241e951d1fb 100644
--- a/absl/time/internal/cctz/src/time_zone_impl.cc
+++ b/absl/time/internal/cctz/src/time_zone_impl.cc
@@ -14,6 +14,7 @@
 
 #include "time_zone_impl.h"
 
+#include <deque>
 #include <mutex>
 #include <string>
 #include <unordered_map>
@@ -91,8 +92,14 @@ bool time_zone::Impl::LoadTimeZone(const std::string& name, time_zone* tz) {
 void time_zone::Impl::ClearTimeZoneMapTestOnly() {
   std::lock_guard<std::mutex> lock(TimeZoneMutex());
   if (time_zone_map != nullptr) {
-    // Existing time_zone::Impl* entries are in the wild, so we simply
-    // leak them.  Future requests will result in reloading the data.
+    // Existing time_zone::Impl* entries are in the wild, so we can't delete
+    // them. Instead, we move them to a private container, where they are
+    // logically unreachable but not "leaked".  Future requests will result
+    // in reloading the data.
+    static auto* cleared = new std::deque<const time_zone::Impl*>;
+    for (const auto& element : *time_zone_map) {
+      cleared->push_back(element.second);
+    }
     time_zone_map->clear();
   }
 }