interrupt()ion of threads in timed_wait() of a condition variable has unexpected side effects

[Sebastian Kienzl <seb@…>]

If multiple threads are waiting on a condition variable with timed_wait() then interrupt()ing one them makes timed_wait() return true in another thread. Happens under Linux, not tested under Windows.

I haven't tested it with version later than 1.39.0 but couldn't find a ticket reporting that bug or any mention of it in the changelog.

#include <boost/thread.hpp>
#include <boost/thread/mutex.hpp>
#include <boost/thread/condition_variable.hpp>
#include <iostream>

using namespace std;
using namespace boost;

condition_variable cv;
mutex mut;

void test_thread( const char* name )
{
	try {
		mutex::scoped_lock lock( mut );
		cout << name << " timed_wait: " <<
			cv.timed_wait( lock, posix_time::time_duration( 0, 20, 0 ) ) << endl;
	}
	catch( thread_interrupted& ) {
		cout << name << " interrupted" << endl;
	}
}


int main()
{
	thread thread1( test_thread, "1" );
	thread thread2( test_thread, "2" );


	sleep( 1 );
	cout << "intr 1" << endl;
	thread1.interrupt();
	sleep( 1 );
	cout << "intr 2" << endl;
	thread2.interrupt();
	sleep( 1 );

	thread1.join();
	thread2.join();

	return 0;
}

[Steven Watanabe]

This behavior is expected. Spurious wakeups are permitted for condition variables. This is one of the reasons why you usually need to wait in a loop.

[anonymous]

Thanks for the answer. With this explanation in mind, the documentation and the code-blocks in "Effects" make more sense.

However, the only thing in the written documentation to suggest this behaviour are the words "or spuriously" for both wait() and timed_wait(). If you're aware of POSIX CV-semantics, it's understood. If you aren't (like I was), it's easy to miss this detail. Maybe it would be helpful to point this out more authoritatively.