Export of internal Abseil changes

--
7a9e8d95f795be037aa2dce4e44809ad0166aaec by Samuel Benzaquen <sbenza@google.com>:

Make end() iterator be nullptr.
This makes the creation of and comparison with end() smaller and faster. `find()!=end()` becomes leaner.

PiperOrigin-RevId: 304681605

--
8f3024979446b391b79b1b60ada7d00a504d6aa6 by Derek Mauro <dmauro@google.com>:

Fix Bazel's distdir detection and prefer double brackets (bash recommendation)

PiperOrigin-RevId: 304615725

--
f1d709cb4b2b3743d548b814dd19602fb057a5e6 by Abseil Team <absl-team@google.com>:

Internal change

PiperOrigin-RevId: 304570545

--
2bbfa5bda52057e1938a96c286ad33ff64e535e0 by Gennadiy Rozental <rogeeff@google.com>:

Implement general storage case as aligned buffer.

Aside from eliminating dynamic memory allocation for flag storage, we also saving 11 bytes per int flag, 15 bytes per double and string flag.

PiperOrigin-RevId: 304511965

--
9e1aed2a95d7d060f8b906fe8c67fc3ba537b521 by Derek Mauro <dmauro@google.com>:

Use reserve to make a bad_alloc less likely in endian_test

This happened once and shouldn't have happened, so it was probably
just a flake, but might as well make this change.

PiperOrigin-RevId: 304505572

--
c2faf22ba2d4d66753390e6959494214895581f0 by Gennadiy Rozental <rogeeff@google.com>:

Use anonymous bit fields to enforce separation between const and mutable bit fields.

We also move init_control field (which is now safe) to save 8 bytes per flag (based on size_tester output)

PiperOrigin-RevId: 304505215

--
7ec51250a84bb03e826b3caad64431e91748186a by Krzysztof Kosiński <krzysio@google.com>:

Change the buffer size in AppendNumberUnit to constexpr.

PiperOrigin-RevId: 304492779

--
a6c8db1be4f421ea7b7c02f7a01b4f48bad61883 by Gennadiy Rozental <rogeeff@google.com>:

Add test cases for two word storage.

Some additional tests were added for other storage kinds as well. These came about after I started to look into a coverage output and noticed that some cases (like reading flag values via reflection) were not covered by this test at all. It does not make sense to just add tests for two word values, so I've covered other storage kinds as well.

PiperOrigin-RevId: 304432511

--
2644ecc32e1215cd6451efcb2f1054fd77e7c812 by Abseil Team <absl-team@google.com>:

Internal change

PiperOrigin-RevId: 304254681

--
4949a6b20c2bb4b9b2c811f439ccb893abc08df5 by Abseil Team <absl-team@google.com>:

Internal change

PiperOrigin-RevId: 304250274
GitOrigin-RevId: 7a9e8d95f795be037aa2dce4e44809ad0166aaec
Change-Id: I01623de87355bec5cf87cc5932a1ca44cade9aae

3 files changed, 8 insertions, 8 deletions

diff --git a/absl/random/internal/iostream_state_saver.h b/absl/random/internal/iostream_state_saver.h
index 7378829a..e6e242ee 100644
--- a/absl/random/internal/iostream_state_saver.h
+++ b/absl/random/internal/iostream_state_saver.h
@@ -192,8 +192,8 @@ struct stream_u128_helper<absl::uint128> {
 
   template <typename OStream>
   inline void write(absl::uint128 val, OStream& out) {
-    uint64_t h = Uint128High64(val);
-    uint64_t l = Uint128Low64(val);
+    uint64_t h = absl::Uint128High64(val);
+    uint64_t l = absl::Uint128Low64(val);
     out << h << out.fill() << l;
   }
 };
diff --git a/absl/random/internal/nanobenchmark_test.cc b/absl/random/internal/nanobenchmark_test.cc
index ab824ef5..f1571e26 100644
--- a/absl/random/internal/nanobenchmark_test.cc
+++ b/absl/random/internal/nanobenchmark_test.cc
@@ -53,7 +53,7 @@ void RunAll(const int argc, char* argv[]) {
   // Avoid migrating between cores - important on multi-socket systems.
   int cpu = -1;
   if (argc == 2) {
-    if (!SimpleAtoi(argv[1], &cpu)) {
+    if (!absl::SimpleAtoi(argv[1], &cpu)) {
       ABSL_RAW_LOG(FATAL, "The optional argument must be a CPU number >= 0.\n");
     }
   }
diff --git a/absl/random/internal/wide_multiply.h b/absl/random/internal/wide_multiply.h
index 6e4cf1be..0afcbe08 100644
--- a/absl/random/internal/wide_multiply.h
+++ b/absl/random/internal/wide_multiply.h
@@ -38,9 +38,9 @@ namespace random_internal {
 // MultiplyU64ToU128 multiplies two 64-bit values to a 128-bit value.
 // If an intrinsic is available, it is used, otherwise use native 32-bit
 // multiplies to construct the result.
-inline uint128 MultiplyU64ToU128(uint64_t a, uint64_t b) {
+inline absl::uint128 MultiplyU64ToU128(uint64_t a, uint64_t b) {
 #if defined(ABSL_HAVE_INTRINSIC_INT128)
-  return uint128(static_cast<__uint128_t>(a) * b);
+  return absl::uint128(static_cast<__uint128_t>(a) * b);
 #elif defined(ABSL_INTERNAL_USE_UMUL128)
   // uint64_t * uint64_t => uint128 multiply using imul intrinsic on MSVC.
   uint64_t high = 0;
@@ -93,14 +93,14 @@ struct wide_multiply {
 template <>
 struct wide_multiply<uint64_t> {
   using input_type = uint64_t;
-  using result_type = uint128;
+  using result_type = absl::uint128;
 
   static result_type multiply(uint64_t a, uint64_t b) {
     return MultiplyU64ToU128(a, b);
   }
 
-  static uint64_t hi(result_type r) { return Uint128High64(r); }
-  static uint64_t lo(result_type r) { return Uint128Low64(r); }
+  static uint64_t hi(result_type r) { return absl::Uint128High64(r); }
+  static uint64_t lo(result_type r) { return absl::Uint128Low64(r); }
 };
 #endif