Abseil Common Libraries (C++) (grcp 依赖) https://abseil.io/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1055 lines
42 KiB

7 years ago
// 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
7 years ago
//
// 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.
//
// -----------------------------------------------------------------------------
// mutex.h
// -----------------------------------------------------------------------------
//
// This header file defines a `Mutex` -- a mutually exclusive lock -- and the
// most common type of synchronization primitive for facilitating locks on
// shared resources. A mutex is used to prevent multiple threads from accessing
// and/or writing to a shared resource concurrently.
//
// Unlike a `std::mutex`, the Abseil `Mutex` provides the following additional
// features:
// * Conditional predicates intrinsic to the `Mutex` object
Export of internal Abseil changes. -- 898e99cae89ca4cc27f86f719148f020d521dd58 by Abseil Team <absl-team@google.com>: Update comment. PiperOrigin-RevId: 204323401 -- b9d14db8b8a9dfb0e1cfb5967aaa0de1c4e94c42 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 204178059 -- f3b5a667611a588aa06fea9168e997ef5cffa7ac by Abseil Team <absl-team@google.com>: Fix a potential reinterpret_cast compile error in absl::InlinedVector The current code will trigger a reinterpret_cast error enhanced by llvm r336738. PiperOrigin-RevId: 204131536 -- cc87d7a8302ad4471c1a25781d6ec19c2ce1524e by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 203979040 -- bc5cae3cfc72af1d3e786d5a3b59a47e205afec9 by Gennadiy Rozental <rogeeff@google.com>: Internal: add internal logging hooks PiperOrigin-RevId: 203850605 -- 503655c248f557f677c920078613522b010e73c8 by Derek Mauro <dmauro@google.com>: Cleanup stacktrace_config.h Instead of listing the platforms that aren't supported, list the ones we do support, and fallback to stacktrace_unimplemented-inl.inc at the end. Previously any platform that wasn't listed gets "#error Not supported yet". GitHub issue #135 PiperOrigin-RevId: 203823079 -- cb69925c4cacc14558cf103a09f218c37f466a16 by Abseil Team <absl-team@google.com>: Fix a minor typo in absl::variant documentation. PiperOrigin-RevId: 203679877 -- 23a0e4db10039011fa5fd879fb73d2f2bbd17301 by Abseil Team <absl-team@google.com>: Format .bzl files with buildifier PiperOrigin-RevId: 203461813 -- 1ad02616bdb715dfdc7f1a73313e383f4ededa03 by Abseil Team <absl-team@google.com>: Make the absl::SleepFor() tests treat any successful retry within a 48x deadline as a total success, thereby reducing flakiness. PiperOrigin-RevId: 203401603 -- b3dccbbc6563ea0173527076a3fff21433d48f47 by Abseil Team <absl-team@google.com>: Replace config_setting.values{"compiler"} with config_setting.flag_values{"@bazel_tools//tools/cpp:compiler"} Due to changes in Bazel we need to change the way "compiler" is specified in config_setting. This will not change the behavior of the config_setting itself. PiperOrigin-RevId: 203345693 -- 170f1692537460a4ba1756d34852275899c2339b by Matt Armstrong <marmstrong@google.com>: Address test flakiness in the TimeoutTest. The basic idea is to loop forever until the expected scheduling delays are observed (within reasonable bounds), and only then assert the other invariants. PiperOrigin-RevId: 203188162 GitOrigin-RevId: 898e99cae89ca4cc27f86f719148f020d521dd58 Change-Id: Ie853ec050afa3a04c519393afe666bc03e11dc87
6 years ago
// * Shared/reader locks, in addition to standard exclusive/writer locks
7 years ago
// * Deadlock detection and debug support.
//
// The following helper classes are also defined within this file:
//
// MutexLock - An RAII wrapper to acquire and release a `Mutex` for exclusive/
// write access within the current scope.
// ReaderMutexLock
// - An RAII wrapper to acquire and release a `Mutex` for shared/read
// access within the current scope.
//
// WriterMutexLock
// - Alias for `MutexLock` above, designed for use in distinguishing
// reader and writer locks within code.
//
// In addition to simple mutex locks, this file also defines ways to perform
// locking under certain conditions.
//
// Condition - (Preferred) Used to wait for a particular predicate that
// depends on state protected by the `Mutex` to become true.
// CondVar - A lower-level variant of `Condition` that relies on
// application code to explicitly signal the `CondVar` when
// a condition has been met.
//
// See below for more information on using `Condition` or `CondVar`.
//
// Mutexes and mutex behavior can be quite complicated. The information within
// this header file is limited, as a result. Please consult the Mutex guide for
// more complete information and examples.
#ifndef ABSL_SYNCHRONIZATION_MUTEX_H_
#define ABSL_SYNCHRONIZATION_MUTEX_H_
#include <atomic>
#include <cstdint>
#include <string>
Export of internal Abseil changes. -- 636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>: Add move semantics to absl::container_internal::CompressedTuple PiperOrigin-RevId: 225394165 -- 43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>: Add a constructor to allow for global absl::Mutex instances. This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration. What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally. Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern. PiperOrigin-RevId: 225389447 -- 7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 225373389 -- fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>: Update absl/time/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225246853 -- 9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>: Update absl/synchronisation/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225237980 -- a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>: Internal cleanup PiperOrigin-RevId: 225226813 -- 48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>: Use a shim #define for wchar_t in msvc in int128. On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type. This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header. https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem. PiperOrigin-RevId: 225223756 GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8 Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074
6 years ago
#include "absl/base/const_init.h"
7 years ago
#include "absl/base/internal/identity.h"
#include "absl/base/internal/low_level_alloc.h"
#include "absl/base/internal/thread_identity.h"
#include "absl/base/internal/tsan_mutex_interface.h"
7 years ago
#include "absl/base/port.h"
#include "absl/base/thread_annotations.h"
#include "absl/synchronization/internal/kernel_timeout.h"
#include "absl/synchronization/internal/per_thread_sem.h"
#include "absl/time/time.h"
// Decide if we should use the non-production implementation because
// the production implementation hasn't been fully ported yet.
#ifdef ABSL_INTERNAL_USE_NONPROD_MUTEX
#error ABSL_INTERNAL_USE_NONPROD_MUTEX cannot be directly set
#elif defined(ABSL_LOW_LEVEL_ALLOC_MISSING)
#define ABSL_INTERNAL_USE_NONPROD_MUTEX 1
#include "absl/synchronization/internal/mutex_nonprod.inc"
#endif
namespace absl {
class Condition;
Changes imported from Abseil "staging" branch: - b527a3e4b36b644ac424e3c525b1cd393f6f6c40 Fix some typos in the usage examples by Jorg Brown <jorg@google.com> - 82be4a9adf3bb0ddafc0d46274969c99afffe870 Fix typo in optional.h comment. by Abseil Team <absl-team@google.com> - d6ee63bf8fc51fba074c23b33cebc28c808d7f07 Remove internal-only identifiers from code. by Daniel Katz <katzdm@google.com> - f9c3ad2f0d73f53b21603638af8b4bed636e79f4 Use easier understandable names for absl::StartsWith and ... by Abseil Team <absl-team@google.com> - 7c16c14fefee89c927b8789d6043c4691bcffc9b Add -Wno-missing-prototypes back to the LLVM copts. by Derek Mauro <dmauro@google.com> - 2f4b7d2e50c7023240242f1e15db60ccd7e8768d IWYU | absl/strings by Juemin Yang <jueminyang@google.com> - a99cbcc1daa34a2d6a2bb26de275e05173cc77e9 IWYU | absl/type by Juemin Yang <jueminyang@google.com> - 12e1146d0fc76c071d7e0ebaabb62f0a984fae66 Use LLVM_FLAGS and LLVM_TEST_FLAGS when --compiler=llvm. by Derek Mauro <dmauro@google.com> - cd6bea616abda558d0bace5bd77455662a233688 IWYU | absl/debugging by Juemin Yang <jueminyang@google.com> - d9a7382e59d46a8581b6b7a31cd5a48bb89326e9 IWYU | absl/synchronization by Juemin Yang <jueminyang@google.com> - 07ec7d6d5a4a666f4183c5d0ed9c342baa7b24bc IWYU | absl/numeric by Juemin Yang <jueminyang@google.com> - 12bfe40051f4270f8707e191af5652f83f2f750c Remove the RoundTrip{Float,Double}ToBuffer routines from ... by Jorg Brown <jorg@google.com> - eeb4fd67c9d97f66cb9475c3c5e51ab132f1c810 Adds conversion functions for converting between absl/tim... by Greg Miller <jgm@google.com> - 59a2108d05d4ea85dc5cc11e49b2cd2335d4295a Change Substitute to use %.6g formatting rather than 15/1... by Jorg Brown <jorg@google.com> - 394becb48e0fcd161642cdaac5120d32567e0ef8 IWYU | absl/meta by Juemin Yang <jueminyang@google.com> - 1e5da6e8da336699b2469dcf6dda025b9b0ec4c9 Rewrite atomic_hook.h to not use std::atomic<T*> under Wi... by Greg Falcon <gfalcon@google.com> GitOrigin-RevId: b527a3e4b36b644ac424e3c525b1cd393f6f6c40 Change-Id: I14e331d91c956ef045ac7927091a9f179716de0c
7 years ago
struct SynchWaitParams;
7 years ago
// -----------------------------------------------------------------------------
// Mutex
// -----------------------------------------------------------------------------
//
// A `Mutex` is a non-reentrant (aka non-recursive) Mutually Exclusive lock
// on some resource, typically a variable or data structure with associated
// invariants. Proper usage of mutexes prevents concurrent access by different
// threads to the same resource.
//
// A `Mutex` has two basic operations: `Mutex::Lock()` and `Mutex::Unlock()`.
// The `Lock()` operation *acquires* a `Mutex` (in a state known as an
// *exclusive* -- or write -- lock), while the `Unlock()` operation *releases* a
// Mutex. During the span of time between the Lock() and Unlock() operations,
// a mutex is said to be *held*. By design all mutexes support exclusive/write
// locks, as this is the most common way to use a mutex.
//
// The `Mutex` state machine for basic lock/unlock operations is quite simple:
//
// | | Lock() | Unlock() |
// |----------------+------------+----------|
// | Free | Exclusive | invalid |
// | Exclusive | blocks | Free |
//
// Attempts to `Unlock()` must originate from the thread that performed the
// corresponding `Lock()` operation.
//
// An "invalid" operation is disallowed by the API. The `Mutex` implementation
// is allowed to do anything on an invalid call, including but not limited to
// crashing with a useful error message, silently succeeding, or corrupting
// data structures. In debug mode, the implementation attempts to crash with a
// useful error message.
//
// `Mutex` is not guaranteed to be "fair" in prioritizing waiting threads; it
// is, however, approximately fair over long periods, and starvation-free for
// threads at the same priority.
//
// The lock/unlock primitives are now annotated with lock annotations
// defined in (base/thread_annotations.h). When writing multi-threaded code,
// you should use lock annotations whenever possible to document your lock
// synchronization policy. Besides acting as documentation, these annotations
// also help compilers or static analysis tools to identify and warn about
// issues that could potentially result in race conditions and deadlocks.
//
// For more information about the lock annotations, please see
// [Thread Safety Analysis](http://clang.llvm.org/docs/ThreadSafetyAnalysis.html)
// in the Clang documentation.
//
// See also `MutexLock`, below, for scoped `Mutex` acquisition.
class ABSL_LOCKABLE Mutex {
7 years ago
public:
Export of internal Abseil changes. -- 636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>: Add move semantics to absl::container_internal::CompressedTuple PiperOrigin-RevId: 225394165 -- 43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>: Add a constructor to allow for global absl::Mutex instances. This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration. What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally. Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern. PiperOrigin-RevId: 225389447 -- 7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 225373389 -- fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>: Update absl/time/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225246853 -- 9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>: Update absl/synchronisation/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225237980 -- a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>: Internal cleanup PiperOrigin-RevId: 225226813 -- 48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>: Use a shim #define for wchar_t in msvc in int128. On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type. This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header. https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem. PiperOrigin-RevId: 225223756 GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8 Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074
6 years ago
// Creates a `Mutex` that is not held by anyone. This constructor is
// typically used for Mutexes allocated on the heap or the stack.
//
// To create `Mutex` instances with static storage duration
// (e.g. a namespace-scoped or global variable), see
// `Mutex::Mutex(absl::kConstInit)` below instead.
7 years ago
Mutex();
Export of internal Abseil changes. -- 636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>: Add move semantics to absl::container_internal::CompressedTuple PiperOrigin-RevId: 225394165 -- 43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>: Add a constructor to allow for global absl::Mutex instances. This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration. What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally. Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern. PiperOrigin-RevId: 225389447 -- 7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 225373389 -- fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>: Update absl/time/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225246853 -- 9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>: Update absl/synchronisation/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225237980 -- a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>: Internal cleanup PiperOrigin-RevId: 225226813 -- 48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>: Use a shim #define for wchar_t in msvc in int128. On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type. This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header. https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem. PiperOrigin-RevId: 225223756 GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8 Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074
6 years ago
// Creates a mutex with static storage duration. A global variable
// constructed this way avoids the lifetime issues that can occur on program
// startup and shutdown. (See absl/base/const_init.h.)
//
// For Mutexes allocated on the heap and stack, instead use the default
// constructor, which can interact more fully with the thread sanitizer.
//
// Example usage:
// namespace foo {
// ABSL_CONST_INIT Mutex mu(absl::kConstInit);
// }
explicit constexpr Mutex(absl::ConstInitType);
Export of internal Abseil changes. -- bdce7e57e9e886eff1114d0266781b443f7ec639 by Derek Mauro <dmauro@google.com>: Change {Get|Set}EnvironmentVariable to {Get|Set}EnvironmentVariableA for compatibility with /DUNICODE. PiperOrigin-RevId: 239229514 -- 2276ed502326a044a84060d34eb19d499e3a3be2 by Derek Mauro <dmauro@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 239228622 -- a462efb970ff43b08a362ef2343fb75ac1295a50 by Derek Mauro <dmauro@google.com>: Adding linking of CoreFoundation to CMakeLists in absl/time. Import https://github.com/abseil/abseil-cpp/pull/280. Fix #283 PiperOrigin-RevId: 239220785 -- fc23327b97f940c682aae1956cf7a1bf87f88c06 by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of gcc (8.3.0 today) with libstdc++ and bazel. PiperOrigin-RevId: 239220448 -- 418c08a8f6a53e63b84e39473035774417ca3aa7 by Derek Mauro <dmauro@google.com>: Disable part of the variant exeception safety test on move assignment when using versions of libstd++ that contain a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87431#c7 PiperOrigin-RevId: 239062455 -- 799722217aeda79679577843c91d5be62cbcbb42 by Matt Calabrese <calabrese@google.com>: Add internal-only IsSwappable traits corresponding to std::is_swappable and std::is_nothrow_swappable, which are used with the swap implementations of optional and variant. PiperOrigin-RevId: 239049448 -- aa46a036038a3de5c68ac5e5d3b4bf76f818d2ea by CJ Johnson <johnsoncj@google.com>: Make InlinedVectorStorage constructor explicit PiperOrigin-RevId: 239044361 -- 17949715b3aa21c794701f69f2154e91b6acabc3 by CJ Johnson <johnsoncj@google.com>: Add absl namesapce to internal/inlined_vector.h PiperOrigin-RevId: 239030789 -- 834628325953078cc08ed10d23bb8890e5bec897 by Derek Mauro <dmauro@google.com>: Add test script that uses Docker to build Abseil with gcc-4.8, libstdc++, and cmake. PiperOrigin-RevId: 239028433 -- 80fe24149ed73ed2ced995ad1e372fb060c60427 by CJ Johnson <johnsoncj@google.com>: Factors data members of InlinedVector into an impl type called InlinedVectorStorage so that (in future changes) the contents of a vector can be grouped together with a single pointer. PiperOrigin-RevId: 239021086 -- 585331436d5d4d79f845e45dcf79d918a0dc6169 by Derek Mauro <dmauro@google.com>: Add -Wno-missing-field-initializers to gcc compiler flags. gcc-4.x has spurious missing field initializer warnings. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36750 PiperOrigin-RevId: 239017217 -- 94602fe4e33ee3a552a7f2939c0f57a992f55075 by Abseil Team <absl-team@google.com>: Formatting fixes. PiperOrigin-RevId: 238983038 -- a1c1b63c08505574e0a8c491561840cecb2bb93e by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libc++ and bazel. PiperOrigin-RevId: 238669118 -- e525f8d20bc2f79a0d69336b902f63858f3bff9d by Derek Mauro <dmauro@google.com>: Disable the test optionalTest.InPlaceTSFINAEBug until libc++ is updated. PiperOrigin-RevId: 238661703 -- f99a2a0b5ec424a059678f7f226600f137b4c74e by Derek Mauro <dmauro@google.com>: Correct the check for the FlatHashMap-Any test bug (list conditions instead of platforms when possible) PiperOrigin-RevId: 238653344 -- 777928035dbcbf39f361eb7d10dc3696822f692f by Jon Cohen <cohenjon@google.com>: Add install rules for Abseil CMake. These are attempted to be limited to in-project installation. This serves two purposes -- first it's morally the same as using Abseil in-source, except you don't have to rebuild us every time. Second, the presence of an install rule makes life massively simpler for package manager maintainers. Currently this doesn't install absl tests or testonly libraries. This can be added in a follow-up patch. Fixes #38, Fixes #80, Closes #182 PiperOrigin-RevId: 238645836 -- ded1c6ce697c191b7a6ff14572b3e6d183117b2c by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libstdc++ and bazel. PiperOrigin-RevId: 238517815 GitOrigin-RevId: bdce7e57e9e886eff1114d0266781b443f7ec639 Change-Id: I6f745869cb8ef63851891ccac05ae9a7dd241c4f
6 years ago
7 years ago
~Mutex();
// Mutex::Lock()
//
// Blocks the calling thread, if necessary, until this `Mutex` is free, and
// then acquires it exclusively. (This lock is also known as a "write lock.")
void Lock() ABSL_EXCLUSIVE_LOCK_FUNCTION();
7 years ago
// Mutex::Unlock()
//
// Releases this `Mutex` and returns it from the exclusive/write state to the
// free state. Caller must hold the `Mutex` exclusively.
void Unlock() ABSL_UNLOCK_FUNCTION();
7 years ago
// Mutex::TryLock()
//
// If the mutex can be acquired without blocking, does so exclusively and
// returns `true`. Otherwise, returns `false`. Returns `true` with high
// probability if the `Mutex` was free.
bool TryLock() ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(true);
7 years ago
// Mutex::AssertHeld()
//
// Return immediately if this thread holds the `Mutex` exclusively (in write
// mode). Otherwise, may report an error (typically by crashing with a
// diagnostic), or may return immediately.
void AssertHeld() const ABSL_ASSERT_EXCLUSIVE_LOCK();
7 years ago
// ---------------------------------------------------------------------------
// Reader-Writer Locking
// ---------------------------------------------------------------------------
// A Mutex can also be used as a starvation-free reader-writer lock.
// Neither read-locks nor write-locks are reentrant/recursive to avoid
// potential client programming errors.
//
// The Mutex API provides `Writer*()` aliases for the existing `Lock()`,
// `Unlock()` and `TryLock()` methods for use within applications mixing
// reader/writer locks. Using `Reader*()` and `Writer*()` operations in this
// manner can make locking behavior clearer when mixing read and write modes.
//
// Introducing reader locks necessarily complicates the `Mutex` state
// machine somewhat. The table below illustrates the allowed state transitions
// of a mutex in such cases. Note that ReaderLock() may block even if the lock
// is held in shared mode; this occurs when another thread is blocked on a
// call to WriterLock().
//
// ---------------------------------------------------------------------------
// Operation: WriterLock() Unlock() ReaderLock() ReaderUnlock()
// ---------------------------------------------------------------------------
// State
// ---------------------------------------------------------------------------
// Free Exclusive invalid Shared(1) invalid
// Shared(1) blocks invalid Shared(2) or blocks Free
// Shared(n) n>1 blocks invalid Shared(n+1) or blocks Shared(n-1)
// Exclusive blocks Free blocks invalid
// ---------------------------------------------------------------------------
//
// In comments below, "shared" refers to a state of Shared(n) for any n > 0.
// Mutex::ReaderLock()
//
// Blocks the calling thread, if necessary, until this `Mutex` is either free,
// or in shared mode, and then acquires a share of it. Note that
// `ReaderLock()` will block if some other thread has an exclusive/writer lock
// on the mutex.
void ReaderLock() ABSL_SHARED_LOCK_FUNCTION();
7 years ago
// Mutex::ReaderUnlock()
//
// Releases a read share of this `Mutex`. `ReaderUnlock` may return a mutex to
// the free state if this thread holds the last reader lock on the mutex. Note
// that you cannot call `ReaderUnlock()` on a mutex held in write mode.
void ReaderUnlock() ABSL_UNLOCK_FUNCTION();
7 years ago
// Mutex::ReaderTryLock()
//
// If the mutex can be acquired without blocking, acquires this mutex for
// shared access and returns `true`. Otherwise, returns `false`. Returns
// `true` with high probability if the `Mutex` was free or shared.
bool ReaderTryLock() ABSL_SHARED_TRYLOCK_FUNCTION(true);
7 years ago
// Mutex::AssertReaderHeld()
//
// Returns immediately if this thread holds the `Mutex` in at least shared
// mode (read mode). Otherwise, may report an error (typically by
// crashing with a diagnostic), or may return immediately.
void AssertReaderHeld() const ABSL_ASSERT_SHARED_LOCK();
7 years ago
// Mutex::WriterLock()
// Mutex::WriterUnlock()
// Mutex::WriterTryLock()
//
// Aliases for `Mutex::Lock()`, `Mutex::Unlock()`, and `Mutex::TryLock()`.
//
// These methods may be used (along with the complementary `Reader*()`
// methods) to distingish simple exclusive `Mutex` usage (`Lock()`,
7 years ago
// etc.) from reader/writer lock usage.
void WriterLock() ABSL_EXCLUSIVE_LOCK_FUNCTION() { this->Lock(); }
7 years ago
void WriterUnlock() ABSL_UNLOCK_FUNCTION() { this->Unlock(); }
7 years ago
bool WriterTryLock() ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(true) {
7 years ago
return this->TryLock();
}
// ---------------------------------------------------------------------------
// Conditional Critical Regions
// ---------------------------------------------------------------------------
// Conditional usage of a `Mutex` can occur using two distinct paradigms:
//
// * Use of `Mutex` member functions with `Condition` objects.
// * Use of the separate `CondVar` abstraction.
//
// In general, prefer use of `Condition` and the `Mutex` member functions
// listed below over `CondVar`. When there are multiple threads waiting on
// distinctly different conditions, however, a battery of `CondVar`s may be
// more efficient. This section discusses use of `Condition` objects.
//
// `Mutex` contains member functions for performing lock operations only under
// certain conditions, of class `Condition`. For correctness, the `Condition`
// must return a boolean that is a pure function, only of state protected by
// the `Mutex`. The condition must be invariant w.r.t. environmental state
// such as thread, cpu id, or time, and must be `noexcept`. The condition will
// always be invoked with the mutex held in at least read mode, so you should
// not block it for long periods or sleep it on a timer.
//
// Since a condition must not depend directly on the current time, use
// `*WithTimeout()` member function variants to make your condition
// effectively true after a given duration, or `*WithDeadline()` variants to
// make your condition effectively true after a given time.
//
// The condition function should have no side-effects aside from debug
// logging; as a special exception, the function may acquire other mutexes
// provided it releases all those that it acquires. (This exception was
// required to allow logging.)
// Mutex::Await()
//
// Unlocks this `Mutex` and blocks until simultaneously both `cond` is `true`
// and this `Mutex` can be reacquired, then reacquires this `Mutex` in the
// same mode in which it was previously held. If the condition is initially
// `true`, `Await()` *may* skip the release/re-acquire step.
//
// `Await()` requires that this thread holds this `Mutex` in some mode.
void Await(const Condition &cond);
// Mutex::LockWhen()
// Mutex::ReaderLockWhen()
// Mutex::WriterLockWhen()
//
Export of internal Abseil changes. -- 898e99cae89ca4cc27f86f719148f020d521dd58 by Abseil Team <absl-team@google.com>: Update comment. PiperOrigin-RevId: 204323401 -- b9d14db8b8a9dfb0e1cfb5967aaa0de1c4e94c42 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 204178059 -- f3b5a667611a588aa06fea9168e997ef5cffa7ac by Abseil Team <absl-team@google.com>: Fix a potential reinterpret_cast compile error in absl::InlinedVector The current code will trigger a reinterpret_cast error enhanced by llvm r336738. PiperOrigin-RevId: 204131536 -- cc87d7a8302ad4471c1a25781d6ec19c2ce1524e by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 203979040 -- bc5cae3cfc72af1d3e786d5a3b59a47e205afec9 by Gennadiy Rozental <rogeeff@google.com>: Internal: add internal logging hooks PiperOrigin-RevId: 203850605 -- 503655c248f557f677c920078613522b010e73c8 by Derek Mauro <dmauro@google.com>: Cleanup stacktrace_config.h Instead of listing the platforms that aren't supported, list the ones we do support, and fallback to stacktrace_unimplemented-inl.inc at the end. Previously any platform that wasn't listed gets "#error Not supported yet". GitHub issue #135 PiperOrigin-RevId: 203823079 -- cb69925c4cacc14558cf103a09f218c37f466a16 by Abseil Team <absl-team@google.com>: Fix a minor typo in absl::variant documentation. PiperOrigin-RevId: 203679877 -- 23a0e4db10039011fa5fd879fb73d2f2bbd17301 by Abseil Team <absl-team@google.com>: Format .bzl files with buildifier PiperOrigin-RevId: 203461813 -- 1ad02616bdb715dfdc7f1a73313e383f4ededa03 by Abseil Team <absl-team@google.com>: Make the absl::SleepFor() tests treat any successful retry within a 48x deadline as a total success, thereby reducing flakiness. PiperOrigin-RevId: 203401603 -- b3dccbbc6563ea0173527076a3fff21433d48f47 by Abseil Team <absl-team@google.com>: Replace config_setting.values{"compiler"} with config_setting.flag_values{"@bazel_tools//tools/cpp:compiler"} Due to changes in Bazel we need to change the way "compiler" is specified in config_setting. This will not change the behavior of the config_setting itself. PiperOrigin-RevId: 203345693 -- 170f1692537460a4ba1756d34852275899c2339b by Matt Armstrong <marmstrong@google.com>: Address test flakiness in the TimeoutTest. The basic idea is to loop forever until the expected scheduling delays are observed (within reasonable bounds), and only then assert the other invariants. PiperOrigin-RevId: 203188162 GitOrigin-RevId: 898e99cae89ca4cc27f86f719148f020d521dd58 Change-Id: Ie853ec050afa3a04c519393afe666bc03e11dc87
6 years ago
// Blocks until simultaneously both `cond` is `true` and this `Mutex` can
7 years ago
// be acquired, then atomically acquires this `Mutex`. `LockWhen()` is
// logically equivalent to `*Lock(); Await();` though they may have different
// performance characteristics.
void LockWhen(const Condition &cond) ABSL_EXCLUSIVE_LOCK_FUNCTION();
7 years ago
void ReaderLockWhen(const Condition &cond) ABSL_SHARED_LOCK_FUNCTION();
7 years ago
void WriterLockWhen(const Condition &cond) ABSL_EXCLUSIVE_LOCK_FUNCTION() {
7 years ago
this->LockWhen(cond);
}
// ---------------------------------------------------------------------------
// Mutex Variants with Timeouts/Deadlines
// ---------------------------------------------------------------------------
// Mutex::AwaitWithTimeout()
// Mutex::AwaitWithDeadline()
//
// If `cond` is initially true, do nothing, or act as though `cond` is
// initially false.
//
// If `cond` is initially false, unlock this `Mutex` and block until
// simultaneously:
// - either `cond` is true or the {timeout has expired, deadline has passed}
// and
// - this `Mutex` can be reacquired,
// then reacquire this `Mutex` in the same mode in which it was previously
// held, returning `true` iff `cond` is `true` on return.
//
// Deadlines in the past are equivalent to an immediate deadline.
// Negative timeouts are equivalent to a zero timeout.
//
// This method requires that this thread holds this `Mutex` in some mode.
bool AwaitWithTimeout(const Condition &cond, absl::Duration timeout);
bool AwaitWithDeadline(const Condition &cond, absl::Time deadline);
// Mutex::LockWhenWithTimeout()
// Mutex::ReaderLockWhenWithTimeout()
// Mutex::WriterLockWhenWithTimeout()
//
// Blocks until simultaneously both:
// - either `cond` is `true` or the timeout has expired, and
// - this `Mutex` can be acquired,
// then atomically acquires this `Mutex`, returning `true` iff `cond` is
// `true` on return.
//
// Negative timeouts are equivalent to a zero timeout.
bool LockWhenWithTimeout(const Condition &cond, absl::Duration timeout)
ABSL_EXCLUSIVE_LOCK_FUNCTION();
7 years ago
bool ReaderLockWhenWithTimeout(const Condition &cond, absl::Duration timeout)
ABSL_SHARED_LOCK_FUNCTION();
7 years ago
bool WriterLockWhenWithTimeout(const Condition &cond, absl::Duration timeout)
ABSL_EXCLUSIVE_LOCK_FUNCTION() {
7 years ago
return this->LockWhenWithTimeout(cond, timeout);
}
// Mutex::LockWhenWithDeadline()
// Mutex::ReaderLockWhenWithDeadline()
// Mutex::WriterLockWhenWithDeadline()
//
// Blocks until simultaneously both:
// - either `cond` is `true` or the deadline has been passed, and
// - this `Mutex` can be acquired,
// then atomically acquires this Mutex, returning `true` iff `cond` is `true`
// on return.
//
// Deadlines in the past are equivalent to an immediate deadline.
bool LockWhenWithDeadline(const Condition &cond, absl::Time deadline)
ABSL_EXCLUSIVE_LOCK_FUNCTION();
7 years ago
bool ReaderLockWhenWithDeadline(const Condition &cond, absl::Time deadline)
ABSL_SHARED_LOCK_FUNCTION();
7 years ago
bool WriterLockWhenWithDeadline(const Condition &cond, absl::Time deadline)
ABSL_EXCLUSIVE_LOCK_FUNCTION() {
7 years ago
return this->LockWhenWithDeadline(cond, deadline);
}
// ---------------------------------------------------------------------------
// Debug Support: Invariant Checking, Deadlock Detection, Logging.
// ---------------------------------------------------------------------------
// Mutex::EnableInvariantDebugging()
//
// If `invariant`!=null and if invariant debugging has been enabled globally,
// cause `(*invariant)(arg)` to be called at moments when the invariant for
// this `Mutex` should hold (for example: just after acquire, just before
// release).
//
// The routine `invariant` should have no side-effects since it is not
// guaranteed how many times it will be called; it should check the invariant
// and crash if it does not hold. Enabling global invariant debugging may
// substantially reduce `Mutex` performance; it should be set only for
// non-production runs. Optimization options may also disable invariant
// checks.
void EnableInvariantDebugging(void (*invariant)(void *), void *arg);
// Mutex::EnableDebugLog()
//
// Cause all subsequent uses of this `Mutex` to be logged via
// `ABSL_RAW_LOG(INFO)`. Log entries are tagged with `name` if no previous
// call to `EnableInvariantDebugging()` or `EnableDebugLog()` has been made.
//
// Note: This method substantially reduces `Mutex` performance.
void EnableDebugLog(const char *name);
// Deadlock detection
// Mutex::ForgetDeadlockInfo()
//
// Forget any deadlock-detection information previously gathered
// about this `Mutex`. Call this method in debug mode when the lock ordering
// of a `Mutex` changes.
void ForgetDeadlockInfo();
// Mutex::AssertNotHeld()
//
// Return immediately if this thread does not hold this `Mutex` in any
// mode; otherwise, may report an error (typically by crashing with a
// diagnostic), or may return immediately.
//
// Currently this check is performed only if all of:
// - in debug mode
// - SetMutexDeadlockDetectionMode() has been set to kReport or kAbort
// - number of locks concurrently held by this thread is not large.
// are true.
void AssertNotHeld() const;
// Special cases.
// A `MuHow` is a constant that indicates how a lock should be acquired.
// Internal implementation detail. Clients should ignore.
typedef const struct MuHowS *MuHow;
// Mutex::InternalAttemptToUseMutexInFatalSignalHandler()
//
// Causes the `Mutex` implementation to prepare itself for re-entry caused by
// future use of `Mutex` within a fatal signal handler. This method is
// intended for use only for last-ditch attempts to log crash information.
// It does not guarantee that attempts to use Mutexes within the handler will
// not deadlock; it merely makes other faults less likely.
//
// WARNING: This routine must be invoked from a signal handler, and the
// signal handler must either loop forever or terminate the process.
// Attempts to return from (or `longjmp` out of) the signal handler once this
// call has been made may cause arbitrary program behaviour including
// crashes and deadlocks.
static void InternalAttemptToUseMutexInFatalSignalHandler();
private:
#ifdef ABSL_INTERNAL_USE_NONPROD_MUTEX
friend class CondVar;
synchronization_internal::MutexImpl *impl() { return impl_.get(); }
synchronization_internal::SynchronizationStorage<
synchronization_internal::MutexImpl>
impl_;
#else
std::atomic<intptr_t> mu_; // The Mutex state.
// Post()/Wait() versus associated PerThreadSem; in class for required
// friendship with PerThreadSem.
static inline void IncrementSynchSem(Mutex *mu,
base_internal::PerThreadSynch *w);
static inline bool DecrementSynchSem(
Mutex *mu, base_internal::PerThreadSynch *w,
synchronization_internal::KernelTimeout t);
// slow path acquire
void LockSlowLoop(SynchWaitParams *waitp, int flags);
// wrappers around LockSlowLoop()
bool LockSlowWithDeadline(MuHow how, const Condition *cond,
synchronization_internal::KernelTimeout t,
int flags);
void LockSlow(MuHow how, const Condition *cond,
int flags) ABSL_ATTRIBUTE_COLD;
// slow path release
void UnlockSlow(SynchWaitParams *waitp) ABSL_ATTRIBUTE_COLD;
// Common code between Await() and AwaitWithTimeout/Deadline()
bool AwaitCommon(const Condition &cond,
synchronization_internal::KernelTimeout t);
// Attempt to remove thread s from queue.
void TryRemove(base_internal::PerThreadSynch *s);
// Block a thread on mutex.
void Block(base_internal::PerThreadSynch *s);
// Wake a thread; return successor.
base_internal::PerThreadSynch *Wakeup(base_internal::PerThreadSynch *w);
friend class CondVar; // for access to Trans()/Fer().
void Trans(MuHow how); // used for CondVar->Mutex transfer
void Fer(
base_internal::PerThreadSynch *w); // used for CondVar->Mutex transfer
#endif
// Catch the error of writing Mutex when intending MutexLock.
Mutex(const volatile Mutex * /*ignored*/) {} // NOLINT(runtime/explicit)
Mutex(const Mutex&) = delete;
Mutex& operator=(const Mutex&) = delete;
};
// -----------------------------------------------------------------------------
// Mutex RAII Wrappers
// -----------------------------------------------------------------------------
// MutexLock
//
// `MutexLock` is a helper class, which acquires and releases a `Mutex` via
// RAII.
//
// Example:
//
// Class Foo {
//
// Foo::Bar* Baz() {
// MutexLock l(&lock_);
// ...
// return bar;
// }
//
// private:
// Mutex lock_;
// };
class ABSL_SCOPED_LOCKABLE MutexLock {
7 years ago
public:
explicit MutexLock(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu) : mu_(mu) {
7 years ago
this->mu_->Lock();
}
MutexLock(const MutexLock &) = delete; // NOLINT(runtime/mutex)
MutexLock(MutexLock&&) = delete; // NOLINT(runtime/mutex)
MutexLock& operator=(const MutexLock&) = delete;
MutexLock& operator=(MutexLock&&) = delete;
~MutexLock() ABSL_UNLOCK_FUNCTION() { this->mu_->Unlock(); }
7 years ago
private:
Mutex *const mu_;
};
// ReaderMutexLock
//
// The `ReaderMutexLock` is a helper class, like `MutexLock`, which acquires and
// releases a shared lock on a `Mutex` via RAII.
class ABSL_SCOPED_LOCKABLE ReaderMutexLock {
7 years ago
public:
explicit ReaderMutexLock(Mutex *mu) ABSL_SHARED_LOCK_FUNCTION(mu) : mu_(mu) {
7 years ago
mu->ReaderLock();
}
ReaderMutexLock(const ReaderMutexLock&) = delete;
ReaderMutexLock(ReaderMutexLock&&) = delete;
ReaderMutexLock& operator=(const ReaderMutexLock&) = delete;
ReaderMutexLock& operator=(ReaderMutexLock&&) = delete;
~ReaderMutexLock() ABSL_UNLOCK_FUNCTION() { this->mu_->ReaderUnlock(); }
7 years ago
private:
Mutex *const mu_;
};
// WriterMutexLock
//
// The `WriterMutexLock` is a helper class, like `MutexLock`, which acquires and
// releases a write (exclusive) lock on a `Mutex` via RAII.
class ABSL_SCOPED_LOCKABLE WriterMutexLock {
7 years ago
public:
explicit WriterMutexLock(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
7 years ago
: mu_(mu) {
mu->WriterLock();
}
WriterMutexLock(const WriterMutexLock&) = delete;
WriterMutexLock(WriterMutexLock&&) = delete;
WriterMutexLock& operator=(const WriterMutexLock&) = delete;
WriterMutexLock& operator=(WriterMutexLock&&) = delete;
~WriterMutexLock() ABSL_UNLOCK_FUNCTION() { this->mu_->WriterUnlock(); }
7 years ago
private:
Mutex *const mu_;
};
// -----------------------------------------------------------------------------
// Condition
// -----------------------------------------------------------------------------
//
// As noted above, `Mutex` contains a number of member functions which take a
Export of internal Abseil changes. -- ba4dd47492748bd630462eb68b7959037fc6a11a by Abseil Team <absl-team@google.com>: Work around nvcc 9.0 compiler bug for open-source Tensorflow build. With the current implementation, when I (unintentionally and transitively) include absl/types/optional.h in a CUDA compilation unit, I get the following nvcc error message: INFO: From Compiling tensorflow/core/kernels/crop_and_resize_op_gpu.cu.cc: external/com_google_absl/absl/types/optional.h: In member function 'void absl::optional_internal::optional_data_dtor_base<T, <anonymous> >::destruct()': external/com_google_absl/absl/types/optional.h:185:50: error: '__T0' was not declared in this scope data_.~T(); I've also seen similar compilation failures online, for flat_hash_map: https://devtalk.nvidia.com/default/topic/1042599/nvcc-preprocessor-bug-causes-compilation-failure/ The bug is always around unnamed template parameters. Therefore, the workaround is to make them named. PiperOrigin-RevId: 219208288 -- dad2f40cb2e8d5017660985ef6fb57f3c3cdcc80 by CJ Johnson <johnsoncj@google.com>: Adds internal macros for catching and throwing unknown exception types PiperOrigin-RevId: 219207362 -- 0a9840328d2d86e8420b853435fdbf1f7a19d931 by Abseil Team <absl-team@google.com>: Fix typo in mutex.h comments. PiperOrigin-RevId: 219199397 -- 0d576dc7597564210bfdf91518075064756f0bf4 by Matt Calabrese <calabrese@google.com>: Internal change. PiperOrigin-RevId: 219185475 -- 66be156095571959fb19a76da8ad0b53ec37658e by Abseil Team <absl-team@google.com>: Fix alignment conformance for VS 2017 >= 15.8 (fix #193) PiperOrigin-RevId: 219129894 -- a6e1825a12587945f8194677ccfdcaba6f7aad1d by Abseil Team <absl-team@google.com>: Reapply PR #173 PiperOrigin-RevId: 219129361 -- cf72ade4881b25acc6ccaea468f69793a0fdce32 by Abseil Team <absl-team@google.com>: Update .gitignore PiperOrigin-RevId: 219127495 -- 0537490c6348a2cb489abe15638928ac5aa6982a by Jon Cohen <cohenjon@google.com>: Small refactor and reformat of error messages from the exception safety test framework. PiperOrigin-RevId: 218927773 -- 4c556ca45fa25698ad12002a00c713aeceefab73 by CJ Johnson <johnsoncj@google.com>: Updates the inlined vector swap tests to check for number of moves that took place if available PiperOrigin-RevId: 218900777 -- dcbfda0021a1e6dfa9586986b1269c06ec394053 by Mark Barolak <mbar@google.com>: Add parens around calls to std::numeric_limits<>::min and std::numeric_limits<>::max to prevent compilation errors on Windows platforms where min and max are defined as macros. PiperOrigin-RevId: 218888700 GitOrigin-RevId: ba4dd47492748bd630462eb68b7959037fc6a11a Change-Id: I0e393958eb8cb501b85f6114979f6d4d86ed996c
6 years ago
// `Condition` as an argument; clients can wait for conditions to become `true`
7 years ago
// before attempting to acquire the mutex. These sections are known as
// "condition critical" sections. To use a `Condition`, you simply need to
// construct it, and use within an appropriate `Mutex` member function;
// everything else in the `Condition` class is an implementation detail.
//
// A `Condition` is specified as a function pointer which returns a boolean.
// `Condition` functions should be pure functions -- their results should depend
// only on passed arguments, should not consult any external state (such as
// clocks), and should have no side-effects, aside from debug logging. Any
// objects that the function may access should be limited to those which are
// constant while the mutex is blocked on the condition (e.g. a stack variable),
// or objects of state protected explicitly by the mutex.
//
// No matter which construction is used for `Condition`, the underlying
// function pointer / functor / callable must not throw any
// exceptions. Correctness of `Mutex` / `Condition` is not guaranteed in
// the face of a throwing `Condition`. (When Abseil is allowed to depend
// on C++17, these function pointers will be explicitly marked
// `noexcept`; until then this requirement cannot be enforced in the
// type system.)
//
// Note: to use a `Condition`, you need only construct it and pass it within the
// appropriate `Mutex' member function, such as `Mutex::Await()`.
//
// Example:
//
// // assume count_ is not internal reference count
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
5 years ago
// int count_ ABSL_GUARDED_BY(mu_);
7 years ago
//
// mu_.LockWhen(Condition(+[](int* count) { return *count == 0; },
7 years ago
// &count_));
//
// When multiple threads are waiting on exactly the same condition, make sure
// that they are constructed with the same parameters (same pointer to function
// + arg, or same pointer to object + method), so that the mutex implementation
// can avoid redundantly evaluating the same condition for each thread.
class Condition {
public:
// A Condition that returns the result of "(*func)(arg)"
Condition(bool (*func)(void *), void *arg);
// Templated version for people who are averse to casts.
//
// To use a lambda, prepend it with unary plus, which converts the lambda
// into a function pointer:
// Condition(+[](T* t) { return ...; }, arg).
//
// Note: lambdas in this case must contain no bound variables.
//
// See class comment for performance advice.
template<typename T>
Condition(bool (*func)(T *), T *arg);
// Templated version for invoking a method that returns a `bool`.
//
// `Condition(object, &Class::Method)` constructs a `Condition` that evaluates
// `object->Method()`.
//
// Implementation Note: `absl::internal::identity` is used to allow methods to
// come from base classes. A simpler signature like
// `Condition(T*, bool (T::*)())` does not suffice.
template<typename T>
Condition(T *object, bool (absl::internal::identity<T>::type::* method)());
// Same as above, for const members
template<typename T>
Condition(const T *object,
bool (absl::internal::identity<T>::type::* method)() const);
// A Condition that returns the value of `*cond`
explicit Condition(const bool *cond);
// Templated version for invoking a functor that returns a `bool`.
// This approach accepts pointers to non-mutable lambdas, `std::function`,
// the result of` std::bind` and user-defined functors that define
// `bool F::operator()() const`.
//
// Example:
//
// auto reached = [this, current]() {
// mu_.AssertReaderHeld(); // For annotalysis.
// return processed_ >= current;
// };
// mu_.Await(Condition(&reached));
// See class comment for performance advice. In particular, if there
// might be more than one waiter for the same condition, make sure
// that all waiters construct the condition with the same pointers.
// Implementation note: The second template parameter ensures that this
// constructor doesn't participate in overload resolution if T doesn't have
// `bool operator() const`.
template <typename T, typename E = decltype(
static_cast<bool (T::*)() const>(&T::operator()))>
explicit Condition(const T *obj)
: Condition(obj, static_cast<bool (T::*)() const>(&T::operator())) {}
// A Condition that always returns `true`.
static const Condition kTrue;
// Evaluates the condition.
bool Eval() const;
// Returns `true` if the two conditions are guaranteed to return the same
// value if evaluated at the same time, `false` if the evaluation *may* return
// different results.
//
// Two `Condition` values are guaranteed equal if both their `func` and `arg`
// components are the same. A null pointer is equivalent to a `true`
// condition.
static bool GuaranteedEqual(const Condition *a, const Condition *b);
private:
typedef bool (*InternalFunctionType)(void * arg);
typedef bool (Condition::*InternalMethodType)();
typedef bool (*InternalMethodCallerType)(void * arg,
InternalMethodType internal_method);
bool (*eval_)(const Condition*); // Actual evaluator
InternalFunctionType function_; // function taking pointer returning bool
InternalMethodType method_; // method returning bool
void *arg_; // arg of function_ or object of method_
Condition(); // null constructor used only to create kTrue
// Various functions eval_ can point to:
static bool CallVoidPtrFunction(const Condition*);
template <typename T> static bool CastAndCallFunction(const Condition* c);
template <typename T> static bool CastAndCallMethod(const Condition* c);
};
// -----------------------------------------------------------------------------
// CondVar
// -----------------------------------------------------------------------------
//
// A condition variable, reflecting state evaluated separately outside of the
// `Mutex` object, which can be signaled to wake callers.
// This class is not normally needed; use `Mutex` member functions such as
// `Mutex::Await()` and intrinsic `Condition` abstractions. In rare cases
// with many threads and many conditions, `CondVar` may be faster.
//
// The implementation may deliver signals to any condition variable at
// any time, even when no call to `Signal()` or `SignalAll()` is made; as a
// result, upon being awoken, you must check the logical condition you have
// been waiting upon.
7 years ago
//
// Examples:
//
// Usage for a thread waiting for some condition C protected by mutex mu:
// mu.Lock();
// while (!C) { cv->Wait(&mu); } // releases and reacquires mu
// // C holds; process data
// mu.Unlock();
//
// Usage to wake T is:
// mu.Lock();
// // process data, possibly establishing C
// if (C) { cv->Signal(); }
// mu.Unlock();
//
// If C may be useful to more than one waiter, use `SignalAll()` instead of
// `Signal()`.
//
// With this implementation it is efficient to use `Signal()/SignalAll()` inside
// the locked region; this usage can make reasoning about your program easier.
//
class CondVar {
public:
CondVar();
~CondVar();
// CondVar::Wait()
//
// Atomically releases a `Mutex` and blocks on this condition variable.
// Waits until awakened by a call to `Signal()` or `SignalAll()` (or a
// spurious wakeup), then reacquires the `Mutex` and returns.
//
7 years ago
// Requires and ensures that the current thread holds the `Mutex`.
void Wait(Mutex *mu);
// CondVar::WaitWithTimeout()
//
// Atomically releases a `Mutex` and blocks on this condition variable.
// Waits until awakened by a call to `Signal()` or `SignalAll()` (or a
// spurious wakeup), or until the timeout has expired, then reacquires
// the `Mutex` and returns.
7 years ago
//
// Returns true if the timeout has expired without this `CondVar`
// being signalled in any manner. If both the timeout has expired
// and this `CondVar` has been signalled, the implementation is free
// to return `true` or `false`.
//
// Requires and ensures that the current thread holds the `Mutex`.
bool WaitWithTimeout(Mutex *mu, absl::Duration timeout);
// CondVar::WaitWithDeadline()
//
// Atomically releases a `Mutex` and blocks on this condition variable.
// Waits until awakened by a call to `Signal()` or `SignalAll()` (or a
// spurious wakeup), or until the deadline has passed, then reacquires
// the `Mutex` and returns.
7 years ago
//
// Deadlines in the past are equivalent to an immediate deadline.
//
// Returns true if the deadline has passed without this `CondVar`
// being signalled in any manner. If both the deadline has passed
// and this `CondVar` has been signalled, the implementation is free
// to return `true` or `false`.
//
// Requires and ensures that the current thread holds the `Mutex`.
bool WaitWithDeadline(Mutex *mu, absl::Time deadline);
// CondVar::Signal()
//
// Signal this `CondVar`; wake at least one waiter if one exists.
void Signal();
// CondVar::SignalAll()
//
// Signal this `CondVar`; wake all waiters.
void SignalAll();
// CondVar::EnableDebugLog()
//
// Causes all subsequent uses of this `CondVar` to be logged via
// `ABSL_RAW_LOG(INFO)`. Log entries are tagged with `name` if `name != 0`.
// Note: this method substantially reduces `CondVar` performance.
void EnableDebugLog(const char *name);
private:
#ifdef ABSL_INTERNAL_USE_NONPROD_MUTEX
synchronization_internal::CondVarImpl *impl() { return impl_.get(); }
synchronization_internal::SynchronizationStorage<
synchronization_internal::CondVarImpl>
impl_;
#else
bool WaitCommon(Mutex *mutex, synchronization_internal::KernelTimeout t);
void Remove(base_internal::PerThreadSynch *s);
void Wakeup(base_internal::PerThreadSynch *w);
std::atomic<intptr_t> cv_; // Condition variable state.
#endif
CondVar(const CondVar&) = delete;
CondVar& operator=(const CondVar&) = delete;
};
// Variants of MutexLock.
//
// If you find yourself using one of these, consider instead using
// Mutex::Unlock() and/or if-statements for clarity.
// MutexLockMaybe
//
// MutexLockMaybe is like MutexLock, but is a no-op when mu is null.
class ABSL_SCOPED_LOCKABLE MutexLockMaybe {
7 years ago
public:
explicit MutexLockMaybe(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
: mu_(mu) {
if (this->mu_ != nullptr) {
this->mu_->Lock();
}
}
~MutexLockMaybe() ABSL_UNLOCK_FUNCTION() {
7 years ago
if (this->mu_ != nullptr) { this->mu_->Unlock(); }
}
7 years ago
private:
Mutex *const mu_;
MutexLockMaybe(const MutexLockMaybe&) = delete;
MutexLockMaybe(MutexLockMaybe&&) = delete;
7 years ago
MutexLockMaybe& operator=(const MutexLockMaybe&) = delete;
MutexLockMaybe& operator=(MutexLockMaybe&&) = delete;
7 years ago
};
// ReleasableMutexLock
7 years ago
//
// ReleasableMutexLock is like MutexLock, but permits `Release()` of its
// mutex before destruction. `Release()` may be called at most once.
class ABSL_SCOPED_LOCKABLE ReleasableMutexLock {
7 years ago
public:
explicit ReleasableMutexLock(Mutex *mu) ABSL_EXCLUSIVE_LOCK_FUNCTION(mu)
7 years ago
: mu_(mu) {
this->mu_->Lock();
}
~ReleasableMutexLock() ABSL_UNLOCK_FUNCTION() {
7 years ago
if (this->mu_ != nullptr) { this->mu_->Unlock(); }
}
void Release() ABSL_UNLOCK_FUNCTION();
7 years ago
private:
Mutex *mu_;
ReleasableMutexLock(const ReleasableMutexLock&) = delete;
ReleasableMutexLock(ReleasableMutexLock&&) = delete;
7 years ago
ReleasableMutexLock& operator=(const ReleasableMutexLock&) = delete;
ReleasableMutexLock& operator=(ReleasableMutexLock&&) = delete;
7 years ago
};
#ifdef ABSL_INTERNAL_USE_NONPROD_MUTEX
Export of internal Abseil changes. -- 636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>: Add move semantics to absl::container_internal::CompressedTuple PiperOrigin-RevId: 225394165 -- 43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>: Add a constructor to allow for global absl::Mutex instances. This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration. What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally. Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern. PiperOrigin-RevId: 225389447 -- 7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 225373389 -- fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>: Update absl/time/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225246853 -- 9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>: Update absl/synchronisation/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225237980 -- a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>: Internal cleanup PiperOrigin-RevId: 225226813 -- 48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>: Use a shim #define for wchar_t in msvc in int128. On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type. This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header. https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem. PiperOrigin-RevId: 225223756 GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8 Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074
6 years ago
inline constexpr Mutex::Mutex(absl::ConstInitType) : impl_(absl::kConstInit) {}
Export of internal Abseil changes. -- bdce7e57e9e886eff1114d0266781b443f7ec639 by Derek Mauro <dmauro@google.com>: Change {Get|Set}EnvironmentVariable to {Get|Set}EnvironmentVariableA for compatibility with /DUNICODE. PiperOrigin-RevId: 239229514 -- 2276ed502326a044a84060d34eb19d499e3a3be2 by Derek Mauro <dmauro@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 239228622 -- a462efb970ff43b08a362ef2343fb75ac1295a50 by Derek Mauro <dmauro@google.com>: Adding linking of CoreFoundation to CMakeLists in absl/time. Import https://github.com/abseil/abseil-cpp/pull/280. Fix #283 PiperOrigin-RevId: 239220785 -- fc23327b97f940c682aae1956cf7a1bf87f88c06 by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of gcc (8.3.0 today) with libstdc++ and bazel. PiperOrigin-RevId: 239220448 -- 418c08a8f6a53e63b84e39473035774417ca3aa7 by Derek Mauro <dmauro@google.com>: Disable part of the variant exeception safety test on move assignment when using versions of libstd++ that contain a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87431#c7 PiperOrigin-RevId: 239062455 -- 799722217aeda79679577843c91d5be62cbcbb42 by Matt Calabrese <calabrese@google.com>: Add internal-only IsSwappable traits corresponding to std::is_swappable and std::is_nothrow_swappable, which are used with the swap implementations of optional and variant. PiperOrigin-RevId: 239049448 -- aa46a036038a3de5c68ac5e5d3b4bf76f818d2ea by CJ Johnson <johnsoncj@google.com>: Make InlinedVectorStorage constructor explicit PiperOrigin-RevId: 239044361 -- 17949715b3aa21c794701f69f2154e91b6acabc3 by CJ Johnson <johnsoncj@google.com>: Add absl namesapce to internal/inlined_vector.h PiperOrigin-RevId: 239030789 -- 834628325953078cc08ed10d23bb8890e5bec897 by Derek Mauro <dmauro@google.com>: Add test script that uses Docker to build Abseil with gcc-4.8, libstdc++, and cmake. PiperOrigin-RevId: 239028433 -- 80fe24149ed73ed2ced995ad1e372fb060c60427 by CJ Johnson <johnsoncj@google.com>: Factors data members of InlinedVector into an impl type called InlinedVectorStorage so that (in future changes) the contents of a vector can be grouped together with a single pointer. PiperOrigin-RevId: 239021086 -- 585331436d5d4d79f845e45dcf79d918a0dc6169 by Derek Mauro <dmauro@google.com>: Add -Wno-missing-field-initializers to gcc compiler flags. gcc-4.x has spurious missing field initializer warnings. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36750 PiperOrigin-RevId: 239017217 -- 94602fe4e33ee3a552a7f2939c0f57a992f55075 by Abseil Team <absl-team@google.com>: Formatting fixes. PiperOrigin-RevId: 238983038 -- a1c1b63c08505574e0a8c491561840cecb2bb93e by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libc++ and bazel. PiperOrigin-RevId: 238669118 -- e525f8d20bc2f79a0d69336b902f63858f3bff9d by Derek Mauro <dmauro@google.com>: Disable the test optionalTest.InPlaceTSFINAEBug until libc++ is updated. PiperOrigin-RevId: 238661703 -- f99a2a0b5ec424a059678f7f226600f137b4c74e by Derek Mauro <dmauro@google.com>: Correct the check for the FlatHashMap-Any test bug (list conditions instead of platforms when possible) PiperOrigin-RevId: 238653344 -- 777928035dbcbf39f361eb7d10dc3696822f692f by Jon Cohen <cohenjon@google.com>: Add install rules for Abseil CMake. These are attempted to be limited to in-project installation. This serves two purposes -- first it's morally the same as using Abseil in-source, except you don't have to rebuild us every time. Second, the presence of an install rule makes life massively simpler for package manager maintainers. Currently this doesn't install absl tests or testonly libraries. This can be added in a follow-up patch. Fixes #38, Fixes #80, Closes #182 PiperOrigin-RevId: 238645836 -- ded1c6ce697c191b7a6ff14572b3e6d183117b2c by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libstdc++ and bazel. PiperOrigin-RevId: 238517815 GitOrigin-RevId: bdce7e57e9e886eff1114d0266781b443f7ec639 Change-Id: I6f745869cb8ef63851891ccac05ae9a7dd241c4f
6 years ago
7 years ago
#else
inline Mutex::Mutex() : mu_(0) {
ABSL_TSAN_MUTEX_CREATE(this, __tsan_mutex_not_static);
}
Export of internal Abseil changes. -- bdce7e57e9e886eff1114d0266781b443f7ec639 by Derek Mauro <dmauro@google.com>: Change {Get|Set}EnvironmentVariable to {Get|Set}EnvironmentVariableA for compatibility with /DUNICODE. PiperOrigin-RevId: 239229514 -- 2276ed502326a044a84060d34eb19d499e3a3be2 by Derek Mauro <dmauro@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 239228622 -- a462efb970ff43b08a362ef2343fb75ac1295a50 by Derek Mauro <dmauro@google.com>: Adding linking of CoreFoundation to CMakeLists in absl/time. Import https://github.com/abseil/abseil-cpp/pull/280. Fix #283 PiperOrigin-RevId: 239220785 -- fc23327b97f940c682aae1956cf7a1bf87f88c06 by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of gcc (8.3.0 today) with libstdc++ and bazel. PiperOrigin-RevId: 239220448 -- 418c08a8f6a53e63b84e39473035774417ca3aa7 by Derek Mauro <dmauro@google.com>: Disable part of the variant exeception safety test on move assignment when using versions of libstd++ that contain a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87431#c7 PiperOrigin-RevId: 239062455 -- 799722217aeda79679577843c91d5be62cbcbb42 by Matt Calabrese <calabrese@google.com>: Add internal-only IsSwappable traits corresponding to std::is_swappable and std::is_nothrow_swappable, which are used with the swap implementations of optional and variant. PiperOrigin-RevId: 239049448 -- aa46a036038a3de5c68ac5e5d3b4bf76f818d2ea by CJ Johnson <johnsoncj@google.com>: Make InlinedVectorStorage constructor explicit PiperOrigin-RevId: 239044361 -- 17949715b3aa21c794701f69f2154e91b6acabc3 by CJ Johnson <johnsoncj@google.com>: Add absl namesapce to internal/inlined_vector.h PiperOrigin-RevId: 239030789 -- 834628325953078cc08ed10d23bb8890e5bec897 by Derek Mauro <dmauro@google.com>: Add test script that uses Docker to build Abseil with gcc-4.8, libstdc++, and cmake. PiperOrigin-RevId: 239028433 -- 80fe24149ed73ed2ced995ad1e372fb060c60427 by CJ Johnson <johnsoncj@google.com>: Factors data members of InlinedVector into an impl type called InlinedVectorStorage so that (in future changes) the contents of a vector can be grouped together with a single pointer. PiperOrigin-RevId: 239021086 -- 585331436d5d4d79f845e45dcf79d918a0dc6169 by Derek Mauro <dmauro@google.com>: Add -Wno-missing-field-initializers to gcc compiler flags. gcc-4.x has spurious missing field initializer warnings. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36750 PiperOrigin-RevId: 239017217 -- 94602fe4e33ee3a552a7f2939c0f57a992f55075 by Abseil Team <absl-team@google.com>: Formatting fixes. PiperOrigin-RevId: 238983038 -- a1c1b63c08505574e0a8c491561840cecb2bb93e by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libc++ and bazel. PiperOrigin-RevId: 238669118 -- e525f8d20bc2f79a0d69336b902f63858f3bff9d by Derek Mauro <dmauro@google.com>: Disable the test optionalTest.InPlaceTSFINAEBug until libc++ is updated. PiperOrigin-RevId: 238661703 -- f99a2a0b5ec424a059678f7f226600f137b4c74e by Derek Mauro <dmauro@google.com>: Correct the check for the FlatHashMap-Any test bug (list conditions instead of platforms when possible) PiperOrigin-RevId: 238653344 -- 777928035dbcbf39f361eb7d10dc3696822f692f by Jon Cohen <cohenjon@google.com>: Add install rules for Abseil CMake. These are attempted to be limited to in-project installation. This serves two purposes -- first it's morally the same as using Abseil in-source, except you don't have to rebuild us every time. Second, the presence of an install rule makes life massively simpler for package manager maintainers. Currently this doesn't install absl tests or testonly libraries. This can be added in a follow-up patch. Fixes #38, Fixes #80, Closes #182 PiperOrigin-RevId: 238645836 -- ded1c6ce697c191b7a6ff14572b3e6d183117b2c by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libstdc++ and bazel. PiperOrigin-RevId: 238517815 GitOrigin-RevId: bdce7e57e9e886eff1114d0266781b443f7ec639 Change-Id: I6f745869cb8ef63851891ccac05ae9a7dd241c4f
6 years ago
Export of internal Abseil changes. -- 636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>: Add move semantics to absl::container_internal::CompressedTuple PiperOrigin-RevId: 225394165 -- 43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>: Add a constructor to allow for global absl::Mutex instances. This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration. What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally. Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern. PiperOrigin-RevId: 225389447 -- 7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 225373389 -- fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>: Update absl/time/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225246853 -- 9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>: Update absl/synchronisation/CMakeLists.txt to use new functions i.e. absl_cc_(library|test) PiperOrigin-RevId: 225237980 -- a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>: Internal cleanup PiperOrigin-RevId: 225226813 -- 48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>: Use a shim #define for wchar_t in msvc in int128. On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type. This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header. https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem. PiperOrigin-RevId: 225223756 GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8 Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074
6 years ago
inline constexpr Mutex::Mutex(absl::ConstInitType) : mu_(0) {}
7 years ago
inline CondVar::CondVar() : cv_(0) {}
#endif
// static
template <typename T>
bool Condition::CastAndCallMethod(const Condition *c) {
typedef bool (T::*MemberType)();
MemberType rm = reinterpret_cast<MemberType>(c->method_);
T *x = static_cast<T *>(c->arg_);
return (x->*rm)();
}
// static
template <typename T>
bool Condition::CastAndCallFunction(const Condition *c) {
typedef bool (*FuncType)(T *);
FuncType fn = reinterpret_cast<FuncType>(c->function_);
T *x = static_cast<T *>(c->arg_);
return (*fn)(x);
}
template <typename T>
inline Condition::Condition(bool (*func)(T *), T *arg)
: eval_(&CastAndCallFunction<T>),
function_(reinterpret_cast<InternalFunctionType>(func)),
method_(nullptr),
arg_(const_cast<void *>(static_cast<const void *>(arg))) {}
template <typename T>
inline Condition::Condition(T *object,
bool (absl::internal::identity<T>::type::*method)())
: eval_(&CastAndCallMethod<T>),
function_(nullptr),
method_(reinterpret_cast<InternalMethodType>(method)),
arg_(object) {}
template <typename T>
inline Condition::Condition(const T *object,
bool (absl::internal::identity<T>::type::*method)()
const)
: eval_(&CastAndCallMethod<T>),
function_(nullptr),
method_(reinterpret_cast<InternalMethodType>(method)),
arg_(reinterpret_cast<void *>(const_cast<T *>(object))) {}
// Register a hook for profiling support.
//
// The function pointer registered here will be called whenever a mutex is
// contended. The callback is given the absl/base/cycleclock.h timestamp when
// waiting began.
//
// Calls to this function do not race or block, but there is no ordering
// guaranteed between calls to this function and call to the provided hook.
// In particular, the previously registered hook may still be called for some
// time after this function returns.
void RegisterMutexProfiler(void (*fn)(int64_t wait_timestamp));
// Register a hook for Mutex tracing.
//
// The function pointer registered here will be called whenever a mutex is
// contended. The callback is given an opaque handle to the contended mutex,
// an event name, and the number of wait cycles (as measured by
// //absl/base/internal/cycleclock.h, and which may not be real
// "cycle" counts.)
//
// The only event name currently sent is "slow release".
//
// This has the same memory ordering concerns as RegisterMutexProfiler() above.
void RegisterMutexTracer(void (*fn)(const char *msg, const void *obj,
int64_t wait_cycles));
// TODO(gfalcon): Combine RegisterMutexProfiler() and RegisterMutexTracer()
// into a single interface, since they are only ever called in pairs.
// Register a hook for CondVar tracing.
//
// The function pointer registered here will be called here on various CondVar
// events. The callback is given an opaque handle to the CondVar object and
// a string identifying the event. This is thread-safe, but only a single
7 years ago
// tracer can be registered.
//
// Events that can be sent are "Wait", "Unwait", "Signal wakeup", and
// "SignalAll wakeup".
//
// This has the same memory ordering concerns as RegisterMutexProfiler() above.
void RegisterCondVarTracer(void (*fn)(const char *msg, const void *cv));
// Register a hook for symbolizing stack traces in deadlock detector reports.
//
// 'pc' is the program counter being symbolized, 'out' is the buffer to write
// into, and 'out_size' is the size of the buffer. This function can return
// false if symbolizing failed, or true if a null-terminated symbol was written
// to 'out.'
//
// This has the same memory ordering concerns as RegisterMutexProfiler() above.
//
// DEPRECATED: The default symbolizer function is absl::Symbolize() and the
// ability to register a different hook for symbolizing stack traces will be
// removed on or after 2023-05-01.
ABSL_DEPRECATED("absl::RegisterSymbolizer() is deprecated and will be removed "
"on or after 2023-05-01")
7 years ago
void RegisterSymbolizer(bool (*fn)(const void *pc, char *out, int out_size));
// EnableMutexInvariantDebugging()
//
// Enable or disable global support for Mutex invariant debugging. If enabled,
// then invariant predicates can be registered per-Mutex for debug checking.
// See Mutex::EnableInvariantDebugging().
void EnableMutexInvariantDebugging(bool enabled);
// When in debug mode, and when the feature has been enabled globally, the
// implementation will keep track of lock ordering and complain (or optionally
// crash) if a cycle is detected in the acquired-before graph.
// Possible modes of operation for the deadlock detector in debug mode.
enum class OnDeadlockCycle {
kIgnore, // Neither report on nor attempt to track cycles in lock ordering
kReport, // Report lock cycles to stderr when detected
kAbort, // Report lock cycles to stderr when detected, then abort
};
// SetMutexDeadlockDetectionMode()
//
// Enable or disable global support for detection of potential deadlocks
// due to Mutex lock ordering inversions. When set to 'kIgnore', tracking of
// lock ordering is disabled. Otherwise, in debug builds, a lock ordering graph
// will be maintained internally, and detected cycles will be reported in
// the manner chosen here.
void SetMutexDeadlockDetectionMode(OnDeadlockCycle mode);
} // namespace absl
// In some build configurations we pass --detect-odr-violations to the
// gold linker. This causes it to flag weak symbol overrides as ODR
// violations. Because ODR only applies to C++ and not C,
// --detect-odr-violations ignores symbols not mangled with C++ names.
// By changing our extension points to be extern "C", we dodge this
// check.
extern "C" {
void AbslInternalMutexYield();
} // extern "C"
Export of internal Abseil changes. -- bdce7e57e9e886eff1114d0266781b443f7ec639 by Derek Mauro <dmauro@google.com>: Change {Get|Set}EnvironmentVariable to {Get|Set}EnvironmentVariableA for compatibility with /DUNICODE. PiperOrigin-RevId: 239229514 -- 2276ed502326a044a84060d34eb19d499e3a3be2 by Derek Mauro <dmauro@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 239228622 -- a462efb970ff43b08a362ef2343fb75ac1295a50 by Derek Mauro <dmauro@google.com>: Adding linking of CoreFoundation to CMakeLists in absl/time. Import https://github.com/abseil/abseil-cpp/pull/280. Fix #283 PiperOrigin-RevId: 239220785 -- fc23327b97f940c682aae1956cf7a1bf87f88c06 by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of gcc (8.3.0 today) with libstdc++ and bazel. PiperOrigin-RevId: 239220448 -- 418c08a8f6a53e63b84e39473035774417ca3aa7 by Derek Mauro <dmauro@google.com>: Disable part of the variant exeception safety test on move assignment when using versions of libstd++ that contain a bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87431#c7 PiperOrigin-RevId: 239062455 -- 799722217aeda79679577843c91d5be62cbcbb42 by Matt Calabrese <calabrese@google.com>: Add internal-only IsSwappable traits corresponding to std::is_swappable and std::is_nothrow_swappable, which are used with the swap implementations of optional and variant. PiperOrigin-RevId: 239049448 -- aa46a036038a3de5c68ac5e5d3b4bf76f818d2ea by CJ Johnson <johnsoncj@google.com>: Make InlinedVectorStorage constructor explicit PiperOrigin-RevId: 239044361 -- 17949715b3aa21c794701f69f2154e91b6acabc3 by CJ Johnson <johnsoncj@google.com>: Add absl namesapce to internal/inlined_vector.h PiperOrigin-RevId: 239030789 -- 834628325953078cc08ed10d23bb8890e5bec897 by Derek Mauro <dmauro@google.com>: Add test script that uses Docker to build Abseil with gcc-4.8, libstdc++, and cmake. PiperOrigin-RevId: 239028433 -- 80fe24149ed73ed2ced995ad1e372fb060c60427 by CJ Johnson <johnsoncj@google.com>: Factors data members of InlinedVector into an impl type called InlinedVectorStorage so that (in future changes) the contents of a vector can be grouped together with a single pointer. PiperOrigin-RevId: 239021086 -- 585331436d5d4d79f845e45dcf79d918a0dc6169 by Derek Mauro <dmauro@google.com>: Add -Wno-missing-field-initializers to gcc compiler flags. gcc-4.x has spurious missing field initializer warnings. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36750 PiperOrigin-RevId: 239017217 -- 94602fe4e33ee3a552a7f2939c0f57a992f55075 by Abseil Team <absl-team@google.com>: Formatting fixes. PiperOrigin-RevId: 238983038 -- a1c1b63c08505574e0a8c491561840cecb2bb93e by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libc++ and bazel. PiperOrigin-RevId: 238669118 -- e525f8d20bc2f79a0d69336b902f63858f3bff9d by Derek Mauro <dmauro@google.com>: Disable the test optionalTest.InPlaceTSFINAEBug until libc++ is updated. PiperOrigin-RevId: 238661703 -- f99a2a0b5ec424a059678f7f226600f137b4c74e by Derek Mauro <dmauro@google.com>: Correct the check for the FlatHashMap-Any test bug (list conditions instead of platforms when possible) PiperOrigin-RevId: 238653344 -- 777928035dbcbf39f361eb7d10dc3696822f692f by Jon Cohen <cohenjon@google.com>: Add install rules for Abseil CMake. These are attempted to be limited to in-project installation. This serves two purposes -- first it's morally the same as using Abseil in-source, except you don't have to rebuild us every time. Second, the presence of an install rule makes life massively simpler for package manager maintainers. Currently this doesn't install absl tests or testonly libraries. This can be added in a follow-up patch. Fixes #38, Fixes #80, Closes #182 PiperOrigin-RevId: 238645836 -- ded1c6ce697c191b7a6ff14572b3e6d183117b2c by Derek Mauro <dmauro@google.com>: Add hermetic test script that uses Docker to build with a very recent version of clang with libstdc++ and bazel. PiperOrigin-RevId: 238517815 GitOrigin-RevId: bdce7e57e9e886eff1114d0266781b443f7ec639 Change-Id: I6f745869cb8ef63851891ccac05ae9a7dd241c4f
6 years ago
7 years ago
#endif // ABSL_SYNCHRONIZATION_MUTEX_H_