tvl-depot/absl/synchronization/lifetime_test.cc

182 lines
6.7 KiB
C++
Raw Normal View History

// Copyright 2017 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include <cstdlib>
#include <thread> // NOLINT(build/c++11), Abseil test
#include <type_traits>
#include "absl/base/attributes.h"
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
2018-12-13 19:30:03 +01:00
#include "absl/base/const_init.h"
#include "absl/base/internal/raw_logging.h"
#include "absl/base/thread_annotations.h"
#include "absl/synchronization/mutex.h"
#include "absl/synchronization/notification.h"
namespace {
// A two-threaded test which checks that Mutex, CondVar, and Notification have
// correct basic functionality. The intent is to establish that they
// function correctly in various phases of construction and destruction.
//
// Thread one acquires a lock on 'mutex', wakes thread two via 'notification',
// then waits for 'state' to be set, as signalled by 'condvar'.
//
// Thread two waits on 'notification', then sets 'state' inside the 'mutex',
// signalling the change via 'condvar'.
//
// These tests use ABSL_RAW_CHECK to validate invariants, rather than EXPECT or
// ASSERT from gUnit, because we need to invoke them during global destructors,
// when gUnit teardown would have already begun.
void ThreadOne(absl::Mutex* mutex, absl::CondVar* condvar,
absl::Notification* notification, bool* state) {
// Test that the notification is in a valid initial state.
ABSL_RAW_CHECK(!notification->HasBeenNotified(), "invalid Notification");
ABSL_RAW_CHECK(*state == false, "*state not initialized");
{
absl::MutexLock lock(mutex);
notification->Notify();
ABSL_RAW_CHECK(notification->HasBeenNotified(), "invalid Notification");
while (*state == false) {
condvar->Wait(mutex);
}
}
}
void ThreadTwo(absl::Mutex* mutex, absl::CondVar* condvar,
absl::Notification* notification, bool* state) {
ABSL_RAW_CHECK(*state == false, "*state not initialized");
// Wake thread one
notification->WaitForNotification();
ABSL_RAW_CHECK(notification->HasBeenNotified(), "invalid Notification");
{
absl::MutexLock lock(mutex);
*state = true;
condvar->Signal();
}
}
// Launch thread 1 and thread 2, and block on their completion.
// If any of 'mutex', 'condvar', or 'notification' is nullptr, use a locally
// constructed instance instead.
Export of internal Abseil changes. -- 07575526242a8e1275ac4223a3d2822795f46569 by CJ Johnson <johnsoncj@google.com>: Comment cleanup on InlinedVector PiperOrigin-RevId: 221322176 -- 49a5e643f85e34d53c41f5e6cc33357c55c9115d by Matt Kulukundis <kfm@google.com>: Internal cleanup PiperOrigin-RevId: 221309185 -- bb35be87ec9c74244b7d902e7e7d2d33ab139d76 by Abseil Team <absl-team@google.com>: Fix typo in comment. PiperOrigin-RevId: 221145354 -- afd4d7c106919708004e06aeea068a57c28aec44 by Derek Mauro <dmauro@google.com>: Update the debugging log message in CallOnceImpl() PiperOrigin-RevId: 221103254 -- 0b9dace8b88113777bf26a6d38f9bc0bcaf053a1 by Abseil Team <absl-team@google.com>: Workaround an MSVC 2015 bug in compile-time initialization. PiperOrigin-RevId: 220871483 -- ea0a3854511ed26beab827e5a5113766b334db86 by Marek Gilbert <mcg@google.com>: Fix ABSL_HAVE_THREAD_LOCAL when compiling for iOS 8 with Xcode 10. Xcode 10 has moved the check for thread_local to a link time, so clang reports __has_feature(cxx_thread_local) but then linking fails with messages like this: ld: targeted OS version does not support use of thread local variables PiperOrigin-RevId: 220815885 -- 485b6876c158c3dcf37eb32d7e512242d5d4ecc6 by Greg Falcon <gfalcon@google.com>: Make the absl::c_set_xxxx() algorithms refuse to compile when passed an unordered collection from std:: or absl::. These algorithms operate on sorted sequences; passing an unordered container to them is nearly certainly a bug. This change is technically an API break, but it only breaks incorrect code. We could try to be more clever and detect unordered collections from other libraries, but false positives will break legal code, and this would constitute an API break Abseil cannot afford. PiperOrigin-RevId: 220794190 -- c47cff7f9cc70a4c1604eee0131af552f40e46d6 by Jon Cohen <cohenjon@google.com>: MSVC 2017's STL throws a Structured Exception (not a C++ exception, essentially equivalent to SIGSEGV) when variant::emplace calls a throwing constructor when using the debug multithreaded MSVC runtime DLL. This manifests in dbg mode in Bazel builds. Disable tests which trigger this bug. It's impossible to specifically pull out MSVC 2017 -dbg modes because there's no way for Bazel to know when version of MSVC is being used -- you tell Bazel the directory where the MSVC tools live, not which version of MSVC tools to use. Thus the best we can do is switch on _DEBUG, which is set whenever the debug runtime is selected with the /MDd build flag, as in Bazel -dbg modes. See https://msdn.microsoft.com/en-us/library/b0084kay.aspx ctrl-f "_DEBUG" PiperOrigin-RevId: 220706161 -- 43993d4af309d92f4ebff38391dcc245f154ecc7 by Shaindel Schwartz <shaindel@google.com>: Internal change PiperOrigin-RevId: 220688429 -- 2448802972dcc261af153af464f2b022ef54a2a9 by Abseil Team <absl-team@google.com>: Speed up operator* for uint128 in WIN64. PiperOrigin-RevId: 220678790 -- 7b376403dd05ba10152fb52e40b29d8af79b58bb by Abseil Team <absl-team@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 220654834 -- ae08af58111c3f838b8d4de25f501c3559c86002 by Abseil Team <absl-team@google.com>: CMake: Add absl_cc_test function PiperOrigin-RevId: 220603940 GitOrigin-RevId: 07575526242a8e1275ac4223a3d2822795f46569 Change-Id: Iba7f53eb394c8a9de564582a976793f9bb0596d9
2018-11-13 22:22:00 +01:00
void RunTests(absl::Mutex* mutex, absl::CondVar* condvar) {
absl::Mutex default_mutex;
absl::CondVar default_condvar;
Export of internal Abseil changes. -- 07575526242a8e1275ac4223a3d2822795f46569 by CJ Johnson <johnsoncj@google.com>: Comment cleanup on InlinedVector PiperOrigin-RevId: 221322176 -- 49a5e643f85e34d53c41f5e6cc33357c55c9115d by Matt Kulukundis <kfm@google.com>: Internal cleanup PiperOrigin-RevId: 221309185 -- bb35be87ec9c74244b7d902e7e7d2d33ab139d76 by Abseil Team <absl-team@google.com>: Fix typo in comment. PiperOrigin-RevId: 221145354 -- afd4d7c106919708004e06aeea068a57c28aec44 by Derek Mauro <dmauro@google.com>: Update the debugging log message in CallOnceImpl() PiperOrigin-RevId: 221103254 -- 0b9dace8b88113777bf26a6d38f9bc0bcaf053a1 by Abseil Team <absl-team@google.com>: Workaround an MSVC 2015 bug in compile-time initialization. PiperOrigin-RevId: 220871483 -- ea0a3854511ed26beab827e5a5113766b334db86 by Marek Gilbert <mcg@google.com>: Fix ABSL_HAVE_THREAD_LOCAL when compiling for iOS 8 with Xcode 10. Xcode 10 has moved the check for thread_local to a link time, so clang reports __has_feature(cxx_thread_local) but then linking fails with messages like this: ld: targeted OS version does not support use of thread local variables PiperOrigin-RevId: 220815885 -- 485b6876c158c3dcf37eb32d7e512242d5d4ecc6 by Greg Falcon <gfalcon@google.com>: Make the absl::c_set_xxxx() algorithms refuse to compile when passed an unordered collection from std:: or absl::. These algorithms operate on sorted sequences; passing an unordered container to them is nearly certainly a bug. This change is technically an API break, but it only breaks incorrect code. We could try to be more clever and detect unordered collections from other libraries, but false positives will break legal code, and this would constitute an API break Abseil cannot afford. PiperOrigin-RevId: 220794190 -- c47cff7f9cc70a4c1604eee0131af552f40e46d6 by Jon Cohen <cohenjon@google.com>: MSVC 2017's STL throws a Structured Exception (not a C++ exception, essentially equivalent to SIGSEGV) when variant::emplace calls a throwing constructor when using the debug multithreaded MSVC runtime DLL. This manifests in dbg mode in Bazel builds. Disable tests which trigger this bug. It's impossible to specifically pull out MSVC 2017 -dbg modes because there's no way for Bazel to know when version of MSVC is being used -- you tell Bazel the directory where the MSVC tools live, not which version of MSVC tools to use. Thus the best we can do is switch on _DEBUG, which is set whenever the debug runtime is selected with the /MDd build flag, as in Bazel -dbg modes. See https://msdn.microsoft.com/en-us/library/b0084kay.aspx ctrl-f "_DEBUG" PiperOrigin-RevId: 220706161 -- 43993d4af309d92f4ebff38391dcc245f154ecc7 by Shaindel Schwartz <shaindel@google.com>: Internal change PiperOrigin-RevId: 220688429 -- 2448802972dcc261af153af464f2b022ef54a2a9 by Abseil Team <absl-team@google.com>: Speed up operator* for uint128 in WIN64. PiperOrigin-RevId: 220678790 -- 7b376403dd05ba10152fb52e40b29d8af79b58bb by Abseil Team <absl-team@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 220654834 -- ae08af58111c3f838b8d4de25f501c3559c86002 by Abseil Team <absl-team@google.com>: CMake: Add absl_cc_test function PiperOrigin-RevId: 220603940 GitOrigin-RevId: 07575526242a8e1275ac4223a3d2822795f46569 Change-Id: Iba7f53eb394c8a9de564582a976793f9bb0596d9
2018-11-13 22:22:00 +01:00
absl::Notification notification;
if (!mutex) {
mutex = &default_mutex;
}
if (!condvar) {
condvar = &default_condvar;
}
bool state = false;
Export of internal Abseil changes. -- 07575526242a8e1275ac4223a3d2822795f46569 by CJ Johnson <johnsoncj@google.com>: Comment cleanup on InlinedVector PiperOrigin-RevId: 221322176 -- 49a5e643f85e34d53c41f5e6cc33357c55c9115d by Matt Kulukundis <kfm@google.com>: Internal cleanup PiperOrigin-RevId: 221309185 -- bb35be87ec9c74244b7d902e7e7d2d33ab139d76 by Abseil Team <absl-team@google.com>: Fix typo in comment. PiperOrigin-RevId: 221145354 -- afd4d7c106919708004e06aeea068a57c28aec44 by Derek Mauro <dmauro@google.com>: Update the debugging log message in CallOnceImpl() PiperOrigin-RevId: 221103254 -- 0b9dace8b88113777bf26a6d38f9bc0bcaf053a1 by Abseil Team <absl-team@google.com>: Workaround an MSVC 2015 bug in compile-time initialization. PiperOrigin-RevId: 220871483 -- ea0a3854511ed26beab827e5a5113766b334db86 by Marek Gilbert <mcg@google.com>: Fix ABSL_HAVE_THREAD_LOCAL when compiling for iOS 8 with Xcode 10. Xcode 10 has moved the check for thread_local to a link time, so clang reports __has_feature(cxx_thread_local) but then linking fails with messages like this: ld: targeted OS version does not support use of thread local variables PiperOrigin-RevId: 220815885 -- 485b6876c158c3dcf37eb32d7e512242d5d4ecc6 by Greg Falcon <gfalcon@google.com>: Make the absl::c_set_xxxx() algorithms refuse to compile when passed an unordered collection from std:: or absl::. These algorithms operate on sorted sequences; passing an unordered container to them is nearly certainly a bug. This change is technically an API break, but it only breaks incorrect code. We could try to be more clever and detect unordered collections from other libraries, but false positives will break legal code, and this would constitute an API break Abseil cannot afford. PiperOrigin-RevId: 220794190 -- c47cff7f9cc70a4c1604eee0131af552f40e46d6 by Jon Cohen <cohenjon@google.com>: MSVC 2017's STL throws a Structured Exception (not a C++ exception, essentially equivalent to SIGSEGV) when variant::emplace calls a throwing constructor when using the debug multithreaded MSVC runtime DLL. This manifests in dbg mode in Bazel builds. Disable tests which trigger this bug. It's impossible to specifically pull out MSVC 2017 -dbg modes because there's no way for Bazel to know when version of MSVC is being used -- you tell Bazel the directory where the MSVC tools live, not which version of MSVC tools to use. Thus the best we can do is switch on _DEBUG, which is set whenever the debug runtime is selected with the /MDd build flag, as in Bazel -dbg modes. See https://msdn.microsoft.com/en-us/library/b0084kay.aspx ctrl-f "_DEBUG" PiperOrigin-RevId: 220706161 -- 43993d4af309d92f4ebff38391dcc245f154ecc7 by Shaindel Schwartz <shaindel@google.com>: Internal change PiperOrigin-RevId: 220688429 -- 2448802972dcc261af153af464f2b022ef54a2a9 by Abseil Team <absl-team@google.com>: Speed up operator* for uint128 in WIN64. PiperOrigin-RevId: 220678790 -- 7b376403dd05ba10152fb52e40b29d8af79b58bb by Abseil Team <absl-team@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 220654834 -- ae08af58111c3f838b8d4de25f501c3559c86002 by Abseil Team <absl-team@google.com>: CMake: Add absl_cc_test function PiperOrigin-RevId: 220603940 GitOrigin-RevId: 07575526242a8e1275ac4223a3d2822795f46569 Change-Id: Iba7f53eb394c8a9de564582a976793f9bb0596d9
2018-11-13 22:22:00 +01:00
std::thread thread_one(ThreadOne, mutex, condvar, &notification, &state);
std::thread thread_two(ThreadTwo, mutex, condvar, &notification, &state);
thread_one.join();
thread_two.join();
}
void TestLocals() {
absl::Mutex mutex;
absl::CondVar condvar;
Export of internal Abseil changes. -- 07575526242a8e1275ac4223a3d2822795f46569 by CJ Johnson <johnsoncj@google.com>: Comment cleanup on InlinedVector PiperOrigin-RevId: 221322176 -- 49a5e643f85e34d53c41f5e6cc33357c55c9115d by Matt Kulukundis <kfm@google.com>: Internal cleanup PiperOrigin-RevId: 221309185 -- bb35be87ec9c74244b7d902e7e7d2d33ab139d76 by Abseil Team <absl-team@google.com>: Fix typo in comment. PiperOrigin-RevId: 221145354 -- afd4d7c106919708004e06aeea068a57c28aec44 by Derek Mauro <dmauro@google.com>: Update the debugging log message in CallOnceImpl() PiperOrigin-RevId: 221103254 -- 0b9dace8b88113777bf26a6d38f9bc0bcaf053a1 by Abseil Team <absl-team@google.com>: Workaround an MSVC 2015 bug in compile-time initialization. PiperOrigin-RevId: 220871483 -- ea0a3854511ed26beab827e5a5113766b334db86 by Marek Gilbert <mcg@google.com>: Fix ABSL_HAVE_THREAD_LOCAL when compiling for iOS 8 with Xcode 10. Xcode 10 has moved the check for thread_local to a link time, so clang reports __has_feature(cxx_thread_local) but then linking fails with messages like this: ld: targeted OS version does not support use of thread local variables PiperOrigin-RevId: 220815885 -- 485b6876c158c3dcf37eb32d7e512242d5d4ecc6 by Greg Falcon <gfalcon@google.com>: Make the absl::c_set_xxxx() algorithms refuse to compile when passed an unordered collection from std:: or absl::. These algorithms operate on sorted sequences; passing an unordered container to them is nearly certainly a bug. This change is technically an API break, but it only breaks incorrect code. We could try to be more clever and detect unordered collections from other libraries, but false positives will break legal code, and this would constitute an API break Abseil cannot afford. PiperOrigin-RevId: 220794190 -- c47cff7f9cc70a4c1604eee0131af552f40e46d6 by Jon Cohen <cohenjon@google.com>: MSVC 2017's STL throws a Structured Exception (not a C++ exception, essentially equivalent to SIGSEGV) when variant::emplace calls a throwing constructor when using the debug multithreaded MSVC runtime DLL. This manifests in dbg mode in Bazel builds. Disable tests which trigger this bug. It's impossible to specifically pull out MSVC 2017 -dbg modes because there's no way for Bazel to know when version of MSVC is being used -- you tell Bazel the directory where the MSVC tools live, not which version of MSVC tools to use. Thus the best we can do is switch on _DEBUG, which is set whenever the debug runtime is selected with the /MDd build flag, as in Bazel -dbg modes. See https://msdn.microsoft.com/en-us/library/b0084kay.aspx ctrl-f "_DEBUG" PiperOrigin-RevId: 220706161 -- 43993d4af309d92f4ebff38391dcc245f154ecc7 by Shaindel Schwartz <shaindel@google.com>: Internal change PiperOrigin-RevId: 220688429 -- 2448802972dcc261af153af464f2b022ef54a2a9 by Abseil Team <absl-team@google.com>: Speed up operator* for uint128 in WIN64. PiperOrigin-RevId: 220678790 -- 7b376403dd05ba10152fb52e40b29d8af79b58bb by Abseil Team <absl-team@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 220654834 -- ae08af58111c3f838b8d4de25f501c3559c86002 by Abseil Team <absl-team@google.com>: CMake: Add absl_cc_test function PiperOrigin-RevId: 220603940 GitOrigin-RevId: 07575526242a8e1275ac4223a3d2822795f46569 Change-Id: Iba7f53eb394c8a9de564582a976793f9bb0596d9
2018-11-13 22:22:00 +01:00
RunTests(&mutex, &condvar);
}
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
2018-12-13 19:30:03 +01:00
// Normal kConstInit usage
ABSL_CONST_INIT absl::Mutex const_init_mutex(absl::kConstInit);
void TestConstInitGlobal() { RunTests(&const_init_mutex, nullptr); }
// Global variables during start and termination
//
// In a translation unit, static storage duration variables are initialized in
// the order of their definitions, and destroyed in the reverse order of their
// definitions. We can use this to arrange for tests to be run on these objects
// before they are created, and after they are destroyed.
using Function = void (*)();
class OnConstruction {
public:
explicit OnConstruction(Function fn) { fn(); }
};
class OnDestruction {
public:
explicit OnDestruction(Function fn) : fn_(fn) {}
~OnDestruction() { fn_(); }
private:
Function fn_;
};
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
2019-10-30 16:00:44 +01:00
// 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)
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
2018-12-13 19:30:03 +01:00
// kConstInit
// Test early usage. (Declaration comes first; definitions must appear after
// the test runner.)
extern absl::Mutex early_const_init_mutex;
// (Normally I'd write this +[], to make the cast-to-function-pointer explicit,
// but in some MSVC setups we support, lambdas provide conversion operators to
// different flavors of function pointers, making this trick ambiguous.)
OnConstruction test_early_const_init([] {
RunTests(&early_const_init_mutex, nullptr);
});
// This definition appears before test_early_const_init, but it should be
// initialized first (due to constant initialization). Test that the object
// actually works when constructed this way.
ABSL_CONST_INIT absl::Mutex early_const_init_mutex(absl::kConstInit);
// Furthermore, test that the const-init c'tor doesn't stomp over the state of
// a Mutex. Really, this is a test that the platform under test correctly
// supports C++11 constant initialization. (The constant-initialization
// constructors of globals "happen at link time"; memory is pre-initialized,
// before the constructors of either grab_lock or check_still_locked are run.)
extern absl::Mutex const_init_sanity_mutex;
OnConstruction grab_lock([]() ABSL_NO_THREAD_SAFETY_ANALYSIS {
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
2018-12-13 19:30:03 +01:00
const_init_sanity_mutex.Lock();
});
ABSL_CONST_INIT absl::Mutex const_init_sanity_mutex(absl::kConstInit);
OnConstruction check_still_locked([]() ABSL_NO_THREAD_SAFETY_ANALYSIS {
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
2018-12-13 19:30:03 +01:00
const_init_sanity_mutex.AssertHeld();
const_init_sanity_mutex.Unlock();
});
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
2019-10-30 16:00:44 +01:00
#endif // defined(__clang__) || !(defined(_MSC_VER) && _MSC_VER > 1900)
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
2018-12-13 19:30:03 +01:00
// Test shutdown usage. (Declarations come first; definitions must appear after
// the test runner.)
extern absl::Mutex late_const_init_mutex;
// OnDestruction is being used here as a global variable, even though it has a
// non-trivial destructor. This is against the style guide. We're violating
// that rule here to check that the exception we allow for kConstInit is safe.
// NOLINTNEXTLINE
OnDestruction test_late_const_init([] {
RunTests(&late_const_init_mutex, nullptr);
});
ABSL_CONST_INIT absl::Mutex late_const_init_mutex(absl::kConstInit);
} // namespace
int main() {
TestLocals();
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
2018-12-13 19:30:03 +01:00
TestConstInitGlobal();
// Explicitly call exit(0) here, to make it clear that we intend for the
// above global object destructors to run.
std::exit(0);
}