Difference between revisions of "cpp/thread/notify all at thread exit"
(→Example: missing includes) |
(Added LWG issue #2140 DR.) |
||
(19 intermediate revisions by 10 users not shown) | |||
Line 1: | Line 1: | ||
{{cpp/title|notify_all_at_thread_exit}} | {{cpp/title|notify_all_at_thread_exit}} | ||
− | {{cpp/thread/ | + | {{cpp/thread/navbar}} |
− | {{ddcl | header=condition_variable | | + | {{ddcl|header=condition_variable|since=c++11| |
void notify_all_at_thread_exit( std::condition_variable& cond, | void notify_all_at_thread_exit( std::condition_variable& cond, | ||
std::unique_lock<std::mutex> lk ); | std::unique_lock<std::mutex> lk ); | ||
}} | }} | ||
− | + | {{tt|notify_all_at_thread_exit}} provides a mechanism to notify other threads that a given thread has completely finished, including destroying all {{ltt|cpp/keyword/thread_local}} objects. It operates as follows: | |
− | {{ | + | * Ownership of the previously acquired lock {{c|lk}} is transferred to internal storage. |
− | lk.unlock(); | + | |
− | cond.notify_all(); | + | * The execution environment is modified such that when the current thread exits, the condition variable {{c|cond}} is notified as if by {{c multi|lk.unlock();|cond.notify_all();}}. |
− | }} | + | |
+ | The implied {{c|lk.unlock()}} is [[cpp/atomic/memory_order|sequenced after]] the destruction of all objects with [[cpp/keyword/thread_local|thread local storage duration]] associated with the current thread. | ||
− | + | If any of the following conditions is satisfied, the behavior is undefined: | |
+ | * {{c|lk}} is not locked by the calling thread. | ||
+ | * If some other threads are also waiting on {{c|cond}}, {{c|lk.mutex()}} is different from the mutex unlocked by the waiting functions ({{tt|wait}}, {{lc|wait_for}} and {{lc|wait_until}}) called on {{c|cond}} by those threads. | ||
===Notes=== | ===Notes=== | ||
− | + | An equivalent effect may be achieved with the facilities provided by {{lc|std::promise}} or {{lc|std::packaged_task}}. | |
− | + | The supplied lock {{c|lk}} is held until the thread exits. Once this function has been called, no more threads may acquire the same lock in order to wait on {{c|cond}}. If some threads are waiting on this condition variable, ensure that the condition being waited for is satisfied while holding the lock on {{c|lk}}, and that this lock is not released and reacquired prior to calling {{tt|notify_all_at_thread_exit}} to avoid confusion from spurious wakeups in other threads. | |
− | + | ||
− | The supplied lock {{ | + | |
In typical use cases, this function is the last thing called by a detached thread. | In typical use cases, this function is the last thing called by a detached thread. | ||
===Parameters=== | ===Parameters=== | ||
− | {{ | + | {{par begin}} |
− | {{ | + | {{par|cond|the condition variable to notify at thread exit}} |
− | {{ | + | {{par|lk|the lock associated with the condition variable {{c|cond}}}} |
− | {{ | + | {{par end}} |
===Return value=== | ===Return value=== | ||
Line 34: | Line 35: | ||
===Example=== | ===Example=== | ||
− | {{example | + | {{example |
− | + | |This partial code fragment illustrates how {{tt|notify_all_at_thread_exit}} can be used to avoid accessing data that depends on thread locals while those thread locals are in the process of being destructed: | |
− | + | |code= | |
+ | #include <cassert> | ||
+ | #include <condition_variable> | ||
#include <mutex> | #include <mutex> | ||
+ | #include <string> | ||
#include <thread> | #include <thread> | ||
− | + | ||
std::mutex m; | std::mutex m; | ||
std::condition_variable cv; | std::condition_variable cv; | ||
− | + | ||
bool ready = false; | bool ready = false; | ||
− | + | std::string result; // some arbitrary type | |
− | + | ||
void thread_func() | void thread_func() | ||
{ | { | ||
+ | thread_local std::string thread_local_data = "42"; | ||
+ | |||
std::unique_lock<std::mutex> lk(m); | std::unique_lock<std::mutex> lk(m); | ||
− | result = | + | |
+ | // assign a value to result using thread_local data | ||
+ | result = thread_local_data; | ||
ready = true; | ready = true; | ||
+ | |||
std::notify_all_at_thread_exit(cv, std::move(lk)); | std::notify_all_at_thread_exit(cv, std::move(lk)); | ||
− | } // destroy thread_locals | + | |
− | + | } // 1. destroy thread_locals; | |
+ | // 2. unlock mutex; | ||
+ | // 3. notify cv. | ||
+ | |||
int main() | int main() | ||
{ | { | ||
std::thread t(thread_func); | std::thread t(thread_func); | ||
t.detach(); | t.detach(); | ||
+ | |||
// do other work | // do other work | ||
+ | // ... | ||
+ | |||
+ | // wait for the detached thread | ||
std::unique_lock<std::mutex> lk(m); | std::unique_lock<std::mutex> lk(m); | ||
− | + | cv.wait(lk, []{ return ready; }); | |
− | + | ||
− | + | // result is ready and thread_local destructors have finished, no UB | |
− | + | assert(result == "42"); | |
} | } | ||
− | + | |output= | |
}} | }} | ||
+ | |||
+ | ===Defect reports=== | ||
+ | {{dr list begin}} | ||
+ | {{dr list item|wg=lwg|dr=2140|std=C++11|before=the call to {{tt|notify_all_at_thread_exit}}<br>synchronized with calls to functions waiting on {{c|cond}}|after=updated the synchronization<br>requirement}} | ||
+ | {{dr list end}} | ||
===See also=== | ===See also=== | ||
− | {{ | + | {{dsc begin}} |
− | {{ | + | {{dsc inc|cpp/thread/promise/dsc set_value_at_thread_exit}} |
− | {{ | + | {{dsc inc|cpp/thread/packaged_task/dsc make_ready_at_thread_exit}} |
− | {{ | + | {{dsc end}} |
+ | |||
+ | {{langlinks|de|es|fr|it|ja|pt|ru|zh}} |
Latest revision as of 23:39, 13 March 2024
Defined in header <condition_variable>
|
||
void notify_all_at_thread_exit( std::condition_variable& cond, std::unique_lock<std::mutex> lk ); |
(since C++11) | |
notify_all_at_thread_exit
provides a mechanism to notify other threads that a given thread has completely finished, including destroying all thread_local objects. It operates as follows:
- Ownership of the previously acquired lock lk is transferred to internal storage.
- The execution environment is modified such that when the current thread exits, the condition variable cond is notified as if by lk.unlock();
cond.notify_all();.
The implied lk.unlock() is sequenced after the destruction of all objects with thread local storage duration associated with the current thread.
If any of the following conditions is satisfied, the behavior is undefined:
- lk is not locked by the calling thread.
- If some other threads are also waiting on cond, lk.mutex() is different from the mutex unlocked by the waiting functions (
wait
, wait_for and wait_until) called on cond by those threads.
Contents |
[edit] Notes
An equivalent effect may be achieved with the facilities provided by std::promise or std::packaged_task.
The supplied lock lk is held until the thread exits. Once this function has been called, no more threads may acquire the same lock in order to wait on cond. If some threads are waiting on this condition variable, ensure that the condition being waited for is satisfied while holding the lock on lk, and that this lock is not released and reacquired prior to calling notify_all_at_thread_exit
to avoid confusion from spurious wakeups in other threads.
In typical use cases, this function is the last thing called by a detached thread.
[edit] Parameters
cond | - | the condition variable to notify at thread exit |
lk | - | the lock associated with the condition variable cond |
[edit] Return value
(none)
[edit] Example
This partial code fragment illustrates how notify_all_at_thread_exit
can be used to avoid accessing data that depends on thread locals while those thread locals are in the process of being destructed:
#include <cassert> #include <condition_variable> #include <mutex> #include <string> #include <thread> std::mutex m; std::condition_variable cv; bool ready = false; std::string result; // some arbitrary type void thread_func() { thread_local std::string thread_local_data = "42"; std::unique_lock<std::mutex> lk(m); // assign a value to result using thread_local data result = thread_local_data; ready = true; std::notify_all_at_thread_exit(cv, std::move(lk)); } // 1. destroy thread_locals; // 2. unlock mutex; // 3. notify cv. int main() { std::thread t(thread_func); t.detach(); // do other work // ... // wait for the detached thread std::unique_lock<std::mutex> lk(m); cv.wait(lk, []{ return ready; }); // result is ready and thread_local destructors have finished, no UB assert(result == "42"); }
[edit] Defect reports
The following behavior-changing defect reports were applied retroactively to previously published C++ standards.
DR | Applied to | Behavior as published | Correct behavior |
---|---|---|---|
LWG 2140 | C++11 | the call to notify_all_at_thread_exit synchronized with calls to functions waiting on cond |
updated the synchronization requirement |
[edit] See also
sets the result to specific value while delivering the notification only at thread exit (public member function of std::promise<R> )
| |
executes the function ensuring that the result is ready only once the current thread exits (public member function of std::packaged_task<R(Args...)> )
|