Export of internal Abseil changes

--
2f49cb9009386bc67bf54a2908c8720b749c8b7f by Greg Falcon <gfalcon@google.com>:

docs: fix typo

Import of https://github.com/abseil/abseil-cpp/pull/397

PiperOrigin-RevId: 277504420

--
f2bed362c1c12d3fa9c22d11f2b918668e8c37b7 by Abseil Team <absl-team@google.com>:

Avoid our is_[copy/move]_assignable workarounds in MSVC 19.20 and on, since that release introduces a regression that breaks our workaround. We should ideally use the std forms in more cases, but branching when our workarounds fail is simpler to maintain.

PiperOrigin-RevId: 277502334

--
e33de894ffd49848b9e088f59acc9743d1661948 by Derek Mauro <dmauro@google.com>:

Update rules_cc version. The mirror.bazel.build URL does not exist
(cache expiration?)

PiperOrigin-RevId: 277498394

--
b23757b0747c64634d2d701433782c969effef19 by Abseil Team <absl-team@google.com>:

Fix https://github.com/abseil/abseil-cpp/issues/394.

PiperOrigin-RevId: 277491405

--
54c75b8b29813531c52d67cf0ba7063baae4a4f3 by Abseil Team <absl-team@google.com>:

Fix comment typos: waker => waiter.

PiperOrigin-RevId: 277376952

--
874eeaa3b3af808fc88b6355245f643674f5e36e by Abseil Team <absl-team@google.com>:

Don't use atomic ops on waiter and wakeup counts in CONDVAR waiter mode.

Just guard the waiter and wakeup counts with the mutex. This eliminates the
race.

Also fix a typo in the error message for pthread_cond_timedwait.

PiperOrigin-RevId: 277366017

--
ce8c9a63109214519b5a7eaecef2c663c4d566df by Greg Falcon <gfalcon@google.com>:

Implement the config options for our four main C++ forward compatibility types.

These options control whether the names `any`, `optional`, `string_view`, and `variant` in namespace `absl` are aliases to the corresponding C++17 types in namespace `std`.  By default, we continue to auto-detect the configuration of the compiler being run.

These options are not intended to be modified on the command line (as -D flags, say).  Instead, the options.h file can be modified by distributors of Abseil (e.g., binary packagers, maintainers of local copies of Abseil, etc.)

Changing options will change Abseil in an ODR sense.  Any program must only link in a single version of Abseil.  Linking libraries that use Abseil configured with different options is an error: there is no ABI compatibility guarantee when linking different configurations, even if the Abseil versions used are otherwise exactly identical.

PiperOrigin-RevId: 277364298

--
5ed3ad42ae43a05862070f92f9ffd07f5c1f2b81 by Chris Kennelly <ckennelly@google.com>:

Suppress -Wimplicit-int-float-conversion.

On recent builds of Clang, this is an error/warning.

PiperOrigin-RevId: 277346168

--
9b9b0055243c705189bb27d912e6d45a7789cdee by Eric Fiselier <ericwf@google.com>:

Allow building Abseil as a shared library with CMake.

By default CMake's `add_library` creates the target as a static
library. However, users can override the default using the builtin
CMake option -DBUILD_SHARED_LIBS=ON.

This changes Abseil's CMake to respect this configuration option
by removing the explicit `STATIC` in our usages of `add_library`.

PiperOrigin-RevId: 277337753

--
63a8b7b8ede3a9d851916929d6b8537f4f2508ca by Abseil Team <absl-team@google.com>:

Improve AlphaNum Hex performance by using absl::numbers_internal::FastHexToBufferZeroPad16.

PiperOrigin-RevId: 277318108

--
dd047f7e92032682d94b27732df0e4d0670e24a4 by CJ Johnson <johnsoncj@google.com>:

Internal change

PiperOrigin-RevId: 277316913

--
d19ee7074929fed08973cc5b40a844573ce1e0a6 by Abseil Team <absl-team@google.com>:

Handle invoking [[nodiscard]] functions correctly in our tests.

PiperOrigin-RevId: 277301500

--
5373f3737894ba9b8481e95e5f58c7957c00d26a by Chris Kennelly <ckennelly@google.com>:

For internal reasons, loosen visibility restrictions of `//absl/base:malloc_internal`.

As an internal-namespace interface, this module remains unsupported.  We
reserve the right to change, delete, or re-restrict visibility to this target
at any time.

PiperOrigin-RevId: 277118689

--
44e4f6655e05393a375d09b3625c192b1fde5909 by Abseil Team <absl-team@google.com>:

Fix error in example civil day comment.

PiperOrigin-RevId: 277103315

--
7657392b4ce48469106f11cdb952a0bc76a28df3 by Abseil Team <absl-team@google.com>:

Internal change

PiperOrigin-RevId: 277056076

--
c75bda76f40b01fa249b75b5a70c1f5907e56c76 by Abseil Team <absl-team@google.com>:

Suppress lifetime constant-initialization tests when building with MSVC versions > 19.0.

These are broken due to non-compliant initialization order in these versions:
https://developercommunity.visualstudio.com/content/problem/336946/class-with-constexpr-constructor-not-using-static.html
We don't know when Microsoft will fix this bug.

PiperOrigin-RevId: 277049770

--
16c3b9bf2a1796efa57f97b00bcd6227fbccca1f by Matt Calabrese <calabrese@google.com>:

Avoid our is_[copy/move]_assignable workarounds in MSVC 19.20 and on, since that release introduces a regression that breaks our workaround. We should ideally use the std forms in more cases, but branching when our workarounds fail is simpler to maintain.

PiperOrigin-RevId: 277048670

--
e91003fa3ee6026d9b80624a23fc144fa5d74810 by Chris Kennelly <ckennelly@google.com>:

Fix -Wimplicit-int-float-conversion warning in latest clang

PiperOrigin-RevId: 276771618

--
53087ca6603e86ad815f3dd9ab795cc0f79792c1 by Andy Soffer <asoffer@google.com>:

Add documentation on absl::SNPrintF.

PiperOrigin-RevId: 276694032

--
a9426af8cbd4c3a8f9053e7446c821852e41ff61 by Jorg Brown <jorg@google.com>:

Stop including kern/OSByteOrder.h in order to support __APPLE__

Apple compiles with clang now anyway, and clang has support for the
built-in compiler swap functions that are much faster than any function call to
the OS.

PiperOrigin-RevId: 276625231

--
df974be5aa5b4dc1b09c356cb8816edfc7867e63 by Jorg Brown <jorg@google.com>:

Fix the build for Android x86-64 builds, where __SSE4_2__ is defined but
_bswap64 is not.

PiperOrigin-RevId: 276542642

--
d99dc092b3a5ad17643005e55f3b3cb6b9187ccc by Jorg Brown <jorg@google.com>:

Remove a byteswap from the non-SSE path of FastHexToBufferZeroPad16

Remove the need for including absl/base/internal/endian.h from the SSE case
(since if we have the Intel SSE intrinsics, then clearly we also have the
Intel Byte-Swap intrinsics.)

PiperOrigin-RevId: 276532608

--
d67b106dc930d8558810ae3983613bb2ab1e0d36 by Abseil Team <absl-team@google.com>:

Use explicit static_cast<double> for int64_t to double conversion

This uses an explicit static_cast<double>() in the int64_t to double comparisons in duration.cc's SafeAddRepHi. This satisfies clang's -Wimplicit-int-to-float-conversion warning (with https://reviews.llvm.org/D64666). This may also make it easier for humans to realize that the comparison is happening between two floating point double precision values.  It should have no impact on the behavior or generated code.
Tested:
No behavior change
PiperOrigin-RevId: 276529211
GitOrigin-RevId: 2f49cb9009386bc67bf54a2908c8720b749c8b7f
Change-Id: I71e0781893ce219960b8290d54b20532779cb0ff

3 files changed, 35 insertions, 34 deletions

diff --git a/absl/synchronization/internal/waiter.cc b/absl/synchronization/internal/waiter.cc
index d83d8087..2f41aa5f 100644
--- a/absl/synchronization/internal/waiter.cc
+++ b/absl/synchronization/internal/waiter.cc
@@ -161,7 +161,7 @@ bool Waiter::Wait(KernelTimeout t) {
 
 void Waiter::Post() {
   if (futex_.fetch_add(1, std::memory_order_release) == 0) {
-    // We incremented from 0, need to wake a potential waker.
+    // We incremented from 0, need to wake a potential waiter.
     Poke();
   }
 }
@@ -210,8 +210,8 @@ void Waiter::Init() {
     ABSL_RAW_LOG(FATAL, "pthread_cond_init failed: %d", err2);
   }
 
-  waiter_count_.store(0, std::memory_order_relaxed);
-  wakeup_count_.store(0, std::memory_order_relaxed);
+  waiter_count_ = 0;
+  wakeup_count_ = 0;
 }
 
 bool Waiter::Wait(KernelTimeout t) {
@@ -221,24 +221,12 @@ bool Waiter::Wait(KernelTimeout t) {
   }
 
   PthreadMutexHolder h(&mu_);
-  waiter_count_.fetch_add(1, std::memory_order_relaxed);
+  ++waiter_count_;
   // Loop until we find a wakeup to consume or timeout.
   // Note that, since the thread ticker is just reset, we don't need to check
   // whether the thread is idle on the very first pass of the loop.
   bool first_pass = true;
-  while (true) {
-    int x = wakeup_count_.load(std::memory_order_relaxed);
-    if (x != 0) {
-      if (!wakeup_count_.compare_exchange_weak(x, x - 1,
-                                               std::memory_order_acquire,
-                                               std::memory_order_relaxed)) {
-        continue;  // Raced with someone, retry.
-      }
-      // Successfully consumed a wakeup, we're done.
-      waiter_count_.fetch_sub(1, std::memory_order_relaxed);
-      return true;
-    }
-
+  while (wakeup_count_ == 0) {
     if (!first_pass) MaybeBecomeIdle();
     // No wakeups available, time to wait.
     if (!t.has_timeout()) {
@@ -249,34 +237,38 @@ bool Waiter::Wait(KernelTimeout t) {
     } else {
       const int err = pthread_cond_timedwait(&cv_, &mu_, &abs_timeout);
       if (err == ETIMEDOUT) {
-        waiter_count_.fetch_sub(1, std::memory_order_relaxed);
+        --waiter_count_;
         return false;
       }
       if (err != 0) {
-        ABSL_RAW_LOG(FATAL, "pthread_cond_wait failed: %d", err);
+        ABSL_RAW_LOG(FATAL, "pthread_cond_timedwait failed: %d", err);
       }
     }
     first_pass = false;
   }
+  // Consume a wakeup and we're done.
+  --wakeup_count_;
+  --waiter_count_;
+  return true;
 }
 
 void Waiter::Post() {
-  wakeup_count_.fetch_add(1, std::memory_order_release);
-  Poke();
+  PthreadMutexHolder h(&mu_);
+  ++wakeup_count_;
+  InternalCondVarPoke();
 }
 
 void Waiter::Poke() {
-  if (waiter_count_.load(std::memory_order_relaxed) == 0) {
-    return;
-  }
-  // Potentially a waker. Take the lock and check again.
   PthreadMutexHolder h(&mu_);
-  if (waiter_count_.load(std::memory_order_relaxed) == 0) {
-    return;
-  }
-  const int err = pthread_cond_signal(&cv_);
-  if (err != 0) {
-    ABSL_RAW_LOG(FATAL, "pthread_cond_signal failed: %d", err);
+  InternalCondVarPoke();
+}
+
+void Waiter::InternalCondVarPoke() {
+  if (waiter_count_ != 0) {
+    const int err = pthread_cond_signal(&cv_);
+    if (ABSL_PREDICT_FALSE(err != 0)) {
+      ABSL_RAW_LOG(FATAL, "pthread_cond_signal failed: %d", err);
+    }
   }
 }
 
@@ -456,7 +448,7 @@ void Waiter::Poke() {
   if (waiter_count_.load(std::memory_order_relaxed) == 0) {
     return;
   }
-  // Potentially a waker. Take the lock and check again.
+  // Potentially a waiter. Take the lock and check again.
   LockHolder h(WinHelper::GetLock(this));
   if (waiter_count_.load(std::memory_order_relaxed) == 0) {
     return;
diff --git a/absl/synchronization/internal/waiter.h b/absl/synchronization/internal/waiter.h
index 09993545..c2389700 100644
--- a/absl/synchronization/internal/waiter.h
+++ b/absl/synchronization/internal/waiter.h
@@ -106,10 +106,13 @@ class Waiter {
   static_assert(sizeof(int32_t) == sizeof(futex_), "Wrong size for futex");
 
 #elif ABSL_WAITER_MODE == ABSL_WAITER_MODE_CONDVAR
+  // REQUIRES: mu_ must be held.
+  void InternalCondVarPoke();
+
   pthread_mutex_t mu_;
   pthread_cond_t cv_;
-  std::atomic<int> waiter_count_;
-  std::atomic<int> wakeup_count_;  // Unclaimed wakeups, written under lock.
+  int waiter_count_;
+  int wakeup_count_;  // Unclaimed wakeups.
 
 #elif ABSL_WAITER_MODE == ABSL_WAITER_MODE_SEM
   sem_t sem_;
diff --git a/absl/synchronization/lifetime_test.cc b/absl/synchronization/lifetime_test.cc
index 34b8875b..cc973a32 100644
--- a/absl/synchronization/lifetime_test.cc
+++ b/absl/synchronization/lifetime_test.cc
@@ -122,6 +122,11 @@ class OnDestruction {
   Function fn_;
 };
 
+// These tests require that the compiler correctly supports C++11 constant
+// initialization... but MSVC has a known regression since v19.10:
+// https://developercommunity.visualstudio.com/content/problem/336946/class-with-constexpr-constructor-not-using-static.html
+// TODO(epastor): Limit the affected range once MSVC fixes this bug.
+#if defined(__clang__) || !(defined(_MSC_VER) && _MSC_VER > 1900)
 // kConstInit
 // Test early usage.  (Declaration comes first; definitions must appear after
 // the test runner.)
@@ -151,6 +156,7 @@ OnConstruction check_still_locked([]() ABSL_NO_THREAD_SAFETY_ANALYSIS {
   const_init_sanity_mutex.AssertHeld();
   const_init_sanity_mutex.Unlock();
 });
+#endif  // defined(__clang__) || !(defined(_MSC_VER) && _MSC_VER > 1900)
 
 // Test shutdown usage.  (Declarations come first; definitions must appear after
 // the test runner.)