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, 369 insertions, 0 deletions

diff --git a/absl/random/internal/iostream_state_saver_test.cc b/absl/random/internal/iostream_state_saver_test.cc
new file mode 100644
index 00000000..2ecbaac1
--- /dev/null
+++ b/absl/random/internal/iostream_state_saver_test.cc
@@ -0,0 +1,369 @@
+// 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.
+
+#include "absl/random/internal/iostream_state_saver.h"
+
+#include <sstream>
+#include <string>
+
+#include "gtest/gtest.h"
+
+namespace {
+
+using absl::random_internal::make_istream_state_saver;
+using absl::random_internal::make_ostream_state_saver;
+using absl::random_internal::stream_precision_helper;
+
+template <typename T>
+typename absl::enable_if_t<std::is_integral<T>::value, T>  //
+StreamRoundTrip(T t) {
+  std::stringstream ss;
+  {
+    auto saver = make_ostream_state_saver(ss);
+    ss.precision(stream_precision_helper<T>::kPrecision);
+    ss << t;
+  }
+  T result = 0;
+  {
+    auto saver = make_istream_state_saver(ss);
+    ss >> result;
+  }
+  EXPECT_FALSE(ss.fail())            //
+      << ss.str() << " "             //
+      << (ss.good() ? "good " : "")  //
+      << (ss.bad() ? "bad " : "")    //
+      << (ss.eof() ? "eof " : "")    //
+      << (ss.fail() ? "fail " : "");
+
+  return result;
+}
+
+template <typename T>
+typename absl::enable_if_t<std::is_floating_point<T>::value, T>  //
+StreamRoundTrip(T t) {
+  std::stringstream ss;
+  {
+    auto saver = make_ostream_state_saver(ss);
+    ss.precision(stream_precision_helper<T>::kPrecision);
+    ss << t;
+  }
+  T result = 0;
+  {
+    auto saver = make_istream_state_saver(ss);
+    result = absl::random_internal::read_floating_point<T>(ss);
+  }
+  EXPECT_FALSE(ss.fail())            //
+      << ss.str() << " "             //
+      << (ss.good() ? "good " : "")  //
+      << (ss.bad() ? "bad " : "")    //
+      << (ss.eof() ? "eof " : "")    //
+      << (ss.fail() ? "fail " : "");
+
+  return result;
+}
+
+TEST(IOStreamStateSaver, BasicSaverState) {
+  std::stringstream ss;
+  ss.precision(2);
+  ss.fill('x');
+  ss.flags(std::ios_base::dec | std::ios_base::right);
+
+  {
+    auto saver = make_ostream_state_saver(ss);
+    ss.precision(10);
+    EXPECT_NE('x', ss.fill());
+    EXPECT_EQ(10, ss.precision());
+    EXPECT_NE(std::ios_base::dec | std::ios_base::right, ss.flags());
+
+    ss << 1.23;
+  }
+
+  EXPECT_EQ('x', ss.fill());
+  EXPECT_EQ(2, ss.precision());
+  EXPECT_EQ(std::ios_base::dec | std::ios_base::right, ss.flags());
+}
+
+TEST(IOStreamStateSaver, RoundTripInts) {
+  const uint64_t kUintValues[] = {
+      0,
+      1,
+      static_cast<uint64_t>(-1),
+      2,
+      static_cast<uint64_t>(-2),
+
+      1 << 7,
+      1 << 8,
+      1 << 16,
+      1ull << 32,
+      1ull << 50,
+      1ull << 62,
+      1ull << 63,
+
+      (1 << 7) - 1,
+      (1 << 8) - 1,
+      (1 << 16) - 1,
+      (1ull << 32) - 1,
+      (1ull << 50) - 1,
+      (1ull << 62) - 1,
+      (1ull << 63) - 1,
+
+      static_cast<uint64_t>(-(1 << 8)),
+      static_cast<uint64_t>(-(1 << 16)),
+      static_cast<uint64_t>(-(1ll << 32)),
+      static_cast<uint64_t>(-(1ll << 50)),
+      static_cast<uint64_t>(-(1ll << 62)),
+
+      static_cast<uint64_t>(-(1 << 8) - 1),
+      static_cast<uint64_t>(-(1 << 16) - 1),
+      static_cast<uint64_t>(-(1ll << 32) - 1),
+      static_cast<uint64_t>(-(1ll << 50) - 1),
+      static_cast<uint64_t>(-(1ll << 62) - 1),
+  };
+
+  for (const uint64_t u : kUintValues) {
+    EXPECT_EQ(u, StreamRoundTrip<uint64_t>(u));
+
+    int64_t x = static_cast<int64_t>(u);
+    EXPECT_EQ(x, StreamRoundTrip<int64_t>(x));
+
+    double d = static_cast<double>(x);
+    EXPECT_EQ(d, StreamRoundTrip<double>(d));
+
+    float f = d;
+    EXPECT_EQ(f, StreamRoundTrip<float>(f));
+  }
+}
+
+TEST(IOStreamStateSaver, RoundTripFloats) {
+  static_assert(
+      stream_precision_helper<float>::kPrecision >= 9,
+      "stream_precision_helper<float>::kPrecision should be at least 9");
+
+  const float kValues[] = {
+      1,
+      std::nextafter(1.0f, 0.0f),  // 1 - epsilon
+      std::nextafter(1.0f, 2.0f),  // 1 + epsilon
+
+      1.0e+1f,
+      1.0e-1f,
+      1.0e+2f,
+      1.0e-2f,
+      1.0e+10f,
+      1.0e-10f,
+
+      0.00000051110000111311111111f,
+      -0.00000051110000111211111111f,
+
+      1.234678912345678912345e+6f,
+      1.234678912345678912345e-6f,
+      1.234678912345678912345e+30f,
+      1.234678912345678912345e-30f,
+      1.234678912345678912345e+38f,
+      1.0234678912345678912345e-38f,
+
+      // Boundary cases.
+      std::numeric_limits<float>::max(),
+      std::numeric_limits<float>::lowest(),
+      std::numeric_limits<float>::epsilon(),
+      std::nextafter(std::numeric_limits<float>::min(),
+                     1.0f),               // min + epsilon
+      std::numeric_limits<float>::min(),  // smallest normal
+      // There are some errors dealing with denorms on apple platforms.
+      std::numeric_limits<float>::denorm_min(),  // smallest denorm
+      std::numeric_limits<float>::min() / 2,
+      std::nextafter(std::numeric_limits<float>::min(),
+                     0.0f),  // denorm_max
+      std::nextafter(std::numeric_limits<float>::denorm_min(), 1.0f),
+  };
+
+  for (const float f : kValues) {
+    EXPECT_EQ(f, StreamRoundTrip<float>(f));
+    EXPECT_EQ(-f, StreamRoundTrip<float>(-f));
+
+    double d = f;
+    EXPECT_EQ(d, StreamRoundTrip<double>(d));
+    EXPECT_EQ(-d, StreamRoundTrip<double>(-d));
+
+    // Avoid undefined behavior (overflow/underflow).
+    if (d <= std::numeric_limits<int64_t>::max() &&
+        d >= std::numeric_limits<int64_t>::lowest()) {
+      int64_t x = static_cast<int64_t>(f);
+      EXPECT_EQ(x, StreamRoundTrip<int64_t>(x));
+    }
+  }
+}
+
+TEST(IOStreamStateSaver, RoundTripDoubles) {
+  static_assert(
+      stream_precision_helper<double>::kPrecision >= 17,
+      "stream_precision_helper<double>::kPrecision should be at least 17");
+
+  const double kValues[] = {
+      1,
+      std::nextafter(1.0, 0.0),  // 1 - epsilon
+      std::nextafter(1.0, 2.0),  // 1 + epsilon
+
+      1.0e+1,
+      1.0e-1,
+      1.0e+2,
+      1.0e-2,
+      1.0e+10,
+      1.0e-10,
+
+      0.00000051110000111311111111,
+      -0.00000051110000111211111111,
+
+      1.234678912345678912345e+6,
+      1.234678912345678912345e-6,
+      1.234678912345678912345e+30,
+      1.234678912345678912345e-30,
+      1.234678912345678912345e+38,
+      1.0234678912345678912345e-38,
+
+      1.0e+100,
+      1.0e-100,
+      1.234678912345678912345e+308,
+      1.0234678912345678912345e-308,
+      2.22507385850720138e-308,
+
+      // Boundary cases.
+      std::numeric_limits<double>::max(),
+      std::numeric_limits<double>::lowest(),
+      std::numeric_limits<double>::epsilon(),
+      std::nextafter(std::numeric_limits<double>::min(),
+                     1.0),                 // min + epsilon
+      std::numeric_limits<double>::min(),  // smallest normal
+      // There are some errors dealing with denorms on apple platforms.
+      std::numeric_limits<double>::denorm_min(),  // smallest denorm
+      std::numeric_limits<double>::min() / 2,
+      std::nextafter(std::numeric_limits<double>::min(),
+                     0.0),  // denorm_max
+      std::nextafter(std::numeric_limits<double>::denorm_min(), 1.0f),
+  };
+
+  for (const double d : kValues) {
+    EXPECT_EQ(d, StreamRoundTrip<double>(d));
+    EXPECT_EQ(-d, StreamRoundTrip<double>(-d));
+
+    // Avoid undefined behavior (overflow/underflow).
+    if (d <= std::numeric_limits<float>::max() &&
+        d >= std::numeric_limits<float>::lowest()) {
+      float f = static_cast<float>(d);
+      EXPECT_EQ(f, StreamRoundTrip<float>(f));
+    }
+
+    // Avoid undefined behavior (overflow/underflow).
+    if (d <= std::numeric_limits<int64_t>::max() &&
+        d >= std::numeric_limits<int64_t>::lowest()) {
+      int64_t x = static_cast<int64_t>(d);
+      EXPECT_EQ(x, StreamRoundTrip<int64_t>(x));
+    }
+  }
+}
+
+TEST(IOStreamStateSaver, RoundTripLongDoubles) {
+  // Technically, C++ only guarantees that long double is at least as large as a
+  // double.  Practically it varies from 64-bits to 128-bits.
+  //
+  // So it is best to consider long double a best-effort extended precision
+  // type.
+
+  static_assert(
+      stream_precision_helper<long double>::kPrecision >= 36,
+      "stream_precision_helper<long double>::kPrecision should be at least 36");
+
+  using real_type = long double;
+  const real_type kValues[] = {
+      1,
+      std::nextafter(1.0, 0.0),  // 1 - epsilon
+      std::nextafter(1.0, 2.0),  // 1 + epsilon
+
+      1.0e+1,
+      1.0e-1,
+      1.0e+2,
+      1.0e-2,
+      1.0e+10,
+      1.0e-10,
+
+      0.00000051110000111311111111,
+      -0.00000051110000111211111111,
+
+      1.2346789123456789123456789123456789e+6,
+      1.2346789123456789123456789123456789e-6,
+      1.2346789123456789123456789123456789e+30,
+      1.2346789123456789123456789123456789e-30,
+      1.2346789123456789123456789123456789e+38,
+      1.2346789123456789123456789123456789e-38,
+      1.2346789123456789123456789123456789e+308,
+      1.2346789123456789123456789123456789e-308,
+
+      1.0e+100,
+      1.0e-100,
+      1.234678912345678912345e+308,
+      1.0234678912345678912345e-308,
+
+      // Boundary cases.
+      std::numeric_limits<real_type>::max(),
+      std::numeric_limits<real_type>::lowest(),
+      std::numeric_limits<real_type>::epsilon(),
+      std::nextafter(std::numeric_limits<real_type>::min(),
+                     real_type(1)),           // min + epsilon
+      std::numeric_limits<real_type>::min(),  // smallest normal
+      // There are some errors dealing with denorms on apple platforms.
+      std::numeric_limits<real_type>::denorm_min(),  // smallest denorm
+      std::numeric_limits<real_type>::min() / 2,
+      std::nextafter(std::numeric_limits<real_type>::min(),
+                     0.0),  // denorm_max
+      std::nextafter(std::numeric_limits<real_type>::denorm_min(), 1.0f),
+  };
+
+  int index = -1;
+  for (const long double dd : kValues) {
+    index++;
+    EXPECT_EQ(dd, StreamRoundTrip<real_type>(dd)) << index;
+    EXPECT_EQ(-dd, StreamRoundTrip<real_type>(-dd)) << index;
+
+    // Avoid undefined behavior (overflow/underflow).
+    if (dd <= std::numeric_limits<double>::max() &&
+        dd >= std::numeric_limits<double>::lowest()) {
+      double d = static_cast<double>(dd);
+      EXPECT_EQ(d, StreamRoundTrip<double>(d));
+    }
+
+    // Avoid undefined behavior (overflow/underflow).
+    if (dd <= std::numeric_limits<int64_t>::max() &&
+        dd >= std::numeric_limits<int64_t>::lowest()) {
+      int64_t x = static_cast<int64_t>(dd);
+      EXPECT_EQ(x, StreamRoundTrip<int64_t>(x));
+    }
+  }
+}
+
+TEST(StrToDTest, DoubleMin) {
+  const char kV[] = "2.22507385850720138e-308";
+  char* end;
+  double x = std::strtod(kV, &end);
+  EXPECT_EQ(std::numeric_limits<double>::min(), x);
+  // errno may equal ERANGE.
+}
+
+TEST(StrToDTest, DoubleDenormMin) {
+  const char kV[] = "4.94065645841246544e-324";
+  char* end;
+  double x = std::strtod(kV, &end);
+  EXPECT_EQ(std::numeric_limits<double>::denorm_min(), x);
+  // errno may equal ERANGE.
+}
+
+}  // namespace