Export of internal Abseil changes.

--
7a6ff16a85beb730c172d5d25cf1b5e1be885c56 by Laramie Leavitt <lar@google.com>:

Internal change.

PiperOrigin-RevId: 254454546

--
ff8f9bafaefc26d451f576ea4a06d150aed63f6f by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254451562

--
deefc5b651b479ce36f0b4ef203e119c0c8936f2 by CJ Johnson <johnsoncj@google.com>:

Account for subtracting unsigned values from the size of InlinedVector

PiperOrigin-RevId: 254450625

--
3c677316a27bcadc17e41957c809ca472d5fef14 by Andy Soffer <asoffer@google.com>:

Add C++17's std::make_from_tuple to absl/utility/utility.h

PiperOrigin-RevId: 254411573

--
4ee3536a918830eeec402a28fc31a62c7c90b940 by CJ Johnson <johnsoncj@google.com>:

Adds benchmark for the rest of the InlinedVector public API

PiperOrigin-RevId: 254408378

--
e5a21a00700ee83498ff1efbf649169756463ee4 by CJ Johnson <johnsoncj@google.com>:

Updates the definition of InlinedVector::shrink_to_fit() to be exception safe and adds exception safety tests for it.

PiperOrigin-RevId: 254401387

--
2ea82e72b86d82d78b4e4712a63a55981b53c64b by Laramie Leavitt <lar@google.com>:

Use absl::InsecureBitGen in place of std::mt19937
in tests absl/random/...distribution_test.cc

PiperOrigin-RevId: 254289444

--
fa099e02c413a7ffda732415e8105cad26a90337 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254286334

--
ce34b7f36933b30cfa35b9c9a5697a792b5666e4 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254273059

--
6f9c473da7c2090c2e85a37c5f00622e8a912a89 by Jorg Brown <jorg@google.com>:

Change absl::container_internal::CompressedTuple to instantiate its
internal Storage class with the name of the type it's holding, rather
than the name of the Tuple.  This is not an externally-visible change,
other than less compiler memory is used and less debug information is
generated.

PiperOrigin-RevId: 254269285

--
8bd3c186bf2fc0c55d8a2dd6f28a5327502c9fba by Andy Soffer <asoffer@google.com>:

Adding short-hand IntervalClosed for IntervalClosedClosed and IntervalOpen for
IntervalOpenOpen.

PiperOrigin-RevId: 254252419

--
ea957f99b6a04fccd42aa05605605f3b44b1ecfd by Abseil Team <absl-team@google.com>:

Do not directly use __SIZEOF_INT128__.

In order to avoid linker errors when building with clang-cl (__fixunsdfti, __udivti3 and __fixunssfti are undefined), this CL uses ABSL_HAVE_INTRINSIC_INT128 which is not defined for clang-cl.

PiperOrigin-RevId: 254250739

--
89ab385cd26b34d64130bce856253aaba96d2345 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254242321

--
cffc793d93eca6d6bdf7de733847b6ab4a255ae9 by CJ Johnson <johnsoncj@google.com>:

Adds benchmark for InlinedVector::reserve(size_type)

PiperOrigin-RevId: 254199226

--
c90c7a9fa3c8f0c9d5114036979548b055ea2f2a by Gennadiy Rozental <rogeeff@google.com>:

Import of CCTZ from GitHub.

PiperOrigin-RevId: 254072387

--
c4c388beae016c9570ab54ffa1d52660e4a85b7b by Laramie Leavitt <lar@google.com>:

Internal cleanup.

PiperOrigin-RevId: 254062381

--
d3c992e221cc74e5372d0c8fa410170b6a43c062 by Tom Manshreck <shreck@google.com>:

Update distributions.h to Abseil standards

PiperOrigin-RevId: 254054946

--
d15ad0035c34ef11b14fadc5a4a2d3ec415f5518 by CJ Johnson <johnsoncj@google.com>:

Removes functions with only one caller from the implementation details of InlinedVector by manually inlining the definitions

PiperOrigin-RevId: 254005427

--
2f37e807efc3a8ef1f4b539bdd379917d4151520 by Andy Soffer <asoffer@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253999861

--
24ed1694b6430791d781ed533a8f8ccf6cac5856 by CJ Johnson <johnsoncj@google.com>:

Updates the definition of InlinedVector::assign(...)/InlinedVector::operator=(...) to new, exception-safe implementations with exception safety tests to boot

PiperOrigin-RevId: 253993691

--
5613d95f5a7e34a535cfaeadce801441e990843e by CJ Johnson <johnsoncj@google.com>:

Adds benchmarks for InlinedVector::shrink_to_fit()

PiperOrigin-RevId: 253989647

--
2a96ddfdac40bbb8cb6a7f1aeab90917067c6e63 by Abseil Team <absl-team@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253927497

--
bf1aff8fc9ffa921ad74643e9525ecf25b0d8dc1 by Andy Soffer <asoffer@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253920512

--
bfc03f4a3dcda3cf3a4b84bdb84cda24e3394f41 by Laramie Leavitt <lar@google.com>:

Internal change.

PiperOrigin-RevId: 253886486

--
05036cfcc078ca7c5f581a00dfb0daed568cbb69 by Eric Fiselier <ericwf@google.com>:

Don't include `winsock2.h` because it drags in `windows.h` and friends,
and they define awful macros like OPAQUE, ERROR, and more. This has the
potential to break abseil users.

Instead we only forward declare `timeval` and require Windows users
include `winsock2.h` themselves. This is both inconsistent and poor QoI, but so
including 'windows.h' is bad too.

PiperOrigin-RevId: 253852615
GitOrigin-RevId: 7a6ff16a85beb730c172d5d25cf1b5e1be885c56
Change-Id: Icd6aff87da26f29ec8915da856f051129987cef6

1 files changed, 228 insertions, 0 deletions

diff --git a/absl/random/internal/randen_engine.h b/absl/random/internal/randen_engine.h
new file mode 100644
index 000000000000..02212a137c36
--- /dev/null
+++ b/absl/random/internal/randen_engine.h
@@ -0,0 +1,228 @@
+// Copyright 2017 The Abseil Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      https://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef ABSL_RANDOM_INTERNAL_RANDEN_ENGINE_H_
+#define ABSL_RANDOM_INTERNAL_RANDEN_ENGINE_H_
+
+#include <algorithm>
+#include <cinttypes>
+#include <cstdlib>
+#include <iostream>
+#include <iterator>
+#include <limits>
+#include <type_traits>
+
+#include "absl/meta/type_traits.h"
+#include "absl/random/internal/iostream_state_saver.h"
+#include "absl/random/internal/randen.h"
+
+namespace absl {
+namespace random_internal {
+
+// Deterministic pseudorandom byte generator with backtracking resistance
+// (leaking the state does not compromise prior outputs). Based on Reverie
+// (see "A Robust and Sponge-Like PRNG with Improved Efficiency") instantiated
+// with an improved Simpira-like permutation.
+// Returns values of type "T" (must be a built-in unsigned integer type).
+//
+// RANDen = RANDom generator or beetroots in Swiss High German.
+// 'Strong' (well-distributed, unpredictable, backtracking-resistant) random
+// generator, faster in some benchmarks than std::mt19937_64 and pcg64_c32.
+template <typename T>
+class alignas(16) randen_engine {
+ public:
+  // C++11 URBG interface:
+  using result_type = T;
+  static_assert(std::is_unsigned<result_type>::value,
+                "randen_engine template argument must be a built-in unsigned "
+                "integer type");
+
+  static constexpr result_type(min)() {
+    return (std::numeric_limits<result_type>::min)();
+  }
+
+  static constexpr result_type(max)() {
+    return (std::numeric_limits<result_type>::max)();
+  }
+
+  explicit randen_engine(result_type seed_value = 0) { seed(seed_value); }
+
+  template <class SeedSequence,
+            typename = typename absl::enable_if_t<
+                !std::is_same<SeedSequence, randen_engine>::value>>
+  explicit randen_engine(SeedSequence&& seq) {
+    seed(seq);
+  }
+
+  randen_engine(const randen_engine&) = default;
+
+  // Returns random bits from the buffer in units of result_type.
+  result_type operator()() {
+    // Refill the buffer if needed (unlikely).
+    if (next_ >= kStateSizeT) {
+      next_ = kCapacityT;
+      impl_.Generate(state_);
+    }
+
+    return state_[next_++];
+  }
+
+  template <class SeedSequence>
+  typename absl::enable_if_t<
+      !std::is_convertible<SeedSequence, result_type>::value>
+  seed(SeedSequence&& seq) {
+    // Zeroes the state.
+    seed();
+    reseed(seq);
+  }
+
+  void seed(result_type seed_value = 0) {
+    next_ = kStateSizeT;
+    // Zeroes the inner state and fills the outer state with seed_value to
+    // mimics behaviour of reseed
+    std::fill(std::begin(state_), std::begin(state_) + kCapacityT, 0);
+    std::fill(std::begin(state_) + kCapacityT, std::end(state_), seed_value);
+  }
+
+  // Inserts entropy into (part of) the state. Calling this periodically with
+  // sufficient entropy ensures prediction resistance (attackers cannot predict
+  // future outputs even if state is compromised).
+  template <class SeedSequence>
+  void reseed(SeedSequence& seq) {
+    using sequence_result_type = typename SeedSequence::result_type;
+    static_assert(sizeof(sequence_result_type) == 4,
+                  "SeedSequence::result_type must be 32-bit");
+
+    constexpr size_t kBufferSize =
+        Randen::kSeedBytes / sizeof(sequence_result_type);
+    alignas(16) sequence_result_type buffer[kBufferSize];
+
+    // Randen::Absorb XORs the seed into state, which is then mixed by a call
+    // to Randen::Generate. Seeding with only the provided entropy is preferred
+    // to using an arbitrary generate() call, so use [rand.req.seed_seq]
+    // size as a proxy for the number of entropy units that can be generated
+    // without relying on seed sequence mixing...
+    const size_t entropy_size = seq.size();
+    if (entropy_size < kBufferSize) {
+      // ... and only request that many values, or 256-bits, when unspecified.
+      const size_t requested_entropy = (entropy_size == 0) ? 8u : entropy_size;
+      std::fill(std::begin(buffer) + requested_entropy, std::end(buffer), 0);
+      seq.generate(std::begin(buffer), std::begin(buffer) + requested_entropy);
+      // The Randen paper suggests preferentially initializing even-numbered
+      // 128-bit vectors of the randen state (there are 16 such vectors).
+      // The seed data is merged into the state offset by 128-bits, which
+      // implies prefering seed bytes [16..31, ..., 208..223]. Since the
+      // buffer is 32-bit values, we swap the corresponding buffer positions in
+      // 128-bit chunks.
+      size_t dst = kBufferSize;
+      while (dst > 7) {
+        // leave the odd bucket as-is.
+        dst -= 4;
+        size_t src = dst >> 1;
+        // swap 128-bits into the even bucket
+        std::swap(buffer[--dst], buffer[--src]);
+        std::swap(buffer[--dst], buffer[--src]);
+        std::swap(buffer[--dst], buffer[--src]);
+        std::swap(buffer[--dst], buffer[--src]);
+      }
+    } else {
+      seq.generate(std::begin(buffer), std::end(buffer));
+    }
+    impl_.Absorb(buffer, state_);
+
+    // Generate will be called when operator() is called
+    next_ = kStateSizeT;
+  }
+
+  void discard(uint64_t count) {
+    uint64_t step = std::min<uint64_t>(kStateSizeT - next_, count);
+    count -= step;
+
+    constexpr uint64_t kRateT = kStateSizeT - kCapacityT;
+    while (count > 0) {
+      next_ = kCapacityT;
+      impl_.Generate(state_);
+      step = std::min<uint64_t>(kRateT, count);
+      count -= step;
+    }
+    next_ += step;
+  }
+
+  bool operator==(const randen_engine& other) const {
+    return next_ == other.next_ &&
+           std::equal(std::begin(state_), std::end(state_),
+                      std::begin(other.state_));
+  }
+
+  bool operator!=(const randen_engine& other) const {
+    return !(*this == other);
+  }
+
+  template <class CharT, class Traits>
+  friend std::basic_ostream<CharT, Traits>& operator<<(
+      std::basic_ostream<CharT, Traits>& os,  // NOLINT(runtime/references)
+      const randen_engine<T>& engine) {       // NOLINT(runtime/references)
+    using numeric_type =
+        typename random_internal::stream_format_type<result_type>::type;
+    auto saver = random_internal::make_ostream_state_saver(os);
+    for (const auto& elem : engine.state_) {
+      // In the case that `elem` is `uint8_t`, it must be cast to something
+      // larger so that it prints as an integer rather than a character. For
+      // simplicity, apply the cast all circumstances.
+      os << static_cast<numeric_type>(elem) << os.fill();
+    }
+    os << engine.next_;
+    return os;
+  }
+
+  template <class CharT, class Traits>
+  friend std::basic_istream<CharT, Traits>& operator>>(
+      std::basic_istream<CharT, Traits>& is,  // NOLINT(runtime/references)
+      randen_engine<T>& engine) {             // NOLINT(runtime/references)
+    using numeric_type =
+        typename random_internal::stream_format_type<result_type>::type;
+    result_type state[kStateSizeT];
+    size_t next;
+    for (auto& elem : state) {
+      // It is not possible to read uint8_t from wide streams, so it is
+      // necessary to read a wider type and then cast it to uint8_t.
+      numeric_type value;
+      is >> value;
+      elem = static_cast<result_type>(value);
+    }
+    is >> next;
+    if (is.fail()) {
+      return is;
+    }
+    std::memcpy(engine.state_, state, sizeof(engine.state_));
+    engine.next_ = next;
+    return is;
+  }
+
+ private:
+  static constexpr size_t kStateSizeT =
+      Randen::kStateBytes / sizeof(result_type);
+  static constexpr size_t kCapacityT =
+      Randen::kCapacityBytes / sizeof(result_type);
+
+  // First kCapacityT are `inner', the others are accessible random bits.
+  alignas(16) result_type state_[kStateSizeT];
+  size_t next_;  // index within state_
+  Randen impl_;
+};
+
+}  // namespace random_internal
+}  // namespace absl
+
+#endif  // ABSL_RANDOM_INTERNAL_RANDEN_ENGINE_H_