id	summary	reporter	owner	description	type	status	milestone	component	version	severity	resolution	keywords	cc
5034	boost::this_thread::sleep either returns instantaneously or spins	Jamie Allsop	Anthony Williams	"From the docs I'd expect to be able to write this:

{{{
#!cpp
std::cout
    << ""Time Before Sleep: ""
    << boost::posix_time::microsec_clock::local_time()
    << std::endl;

boost::this_thread::sleep( boost::posix_time::seconds(10) );

std::cout
    << ""Time After Sleep : ""
    << boost::posix_time::microsec_clock::local_time()
    << std::endl;
}}}

and have the 'this' thread sleep for approximately 10 seconds. Instead sleep returns instantaneously:

{{{
Time Before Sleep: 2010-Dec-25 01:38:50.949160000
Time After Sleep : 2010-Dec-25 01:38:50.949436000
}}}

I've tried this with boost 1.45 and boost 1.43 using gcc 4.4 on debian x86_64. This used to work for me on 32-bit debian unstable using gcc 4.3.

In addition to sleep returning instantaneously (happens every time) I have observed with more complex scenarios involving multiple threads sleep enter a busy spin, never returning (process needs killed). I am trying to create a minimal test case for this.

Rather obviously the code where this hits is here in
source:/boost/tags/release/Boost_1_45_0/libs/thread/src/pthread/thread.cpp

{{{
#!cpp
        void sleep(const system_time& st)
        {
            detail::thread_data_base* const thread_info=detail::get_current_thread_data();
        
            if(thread_info)
            {
                unique_lock<mutex> lk(thread_info->sleep_mutex);
                while(thread_info->sleep_condition.timed_wait(lk,st));
            }
            else
            {
                xtime const xt=get_xtime(st);
            
                for (int foo=0; foo < 5; ++foo)
                {
#   if defined(BOOST_HAS_PTHREAD_DELAY_NP)
                    timespec ts;
                    to_timespec_duration(xt, ts);
                    BOOST_VERIFY(!pthread_delay_np(&ts));
#   elif defined(BOOST_HAS_NANOSLEEP)
                    timespec ts;
                    to_timespec_duration(xt, ts);
                
                    //  nanosleep takes a timespec that is an offset, not
                    //  an absolute time.
                    nanosleep(&ts, 0);
#   else
                    mutex mx;
                    mutex::scoped_lock lock(mx);
                    condition cond;
                    cond.timed_wait(lock, xt);
#   endif
                    xtime cur;
                    xtime_get(&cur, TIME_UTC);
                    if (xtime_cmp(xt, cur) <= 0)
                        return;
                }
            }
        }

}}}

Specifically something appears to go wrong here:

{{{
            if(thread_info)
            {
                unique_lock<mutex> lk(thread_info->sleep_mutex);
                while(thread_info->sleep_condition.timed_wait(lk,st));
            }
}}}

I came across this while writing a test for something else and so have not been able to follow up in any more detail yet at this time.
"	Bugs	closed	To Be Determined	thread	Boost 1.45.0	Showstopper	worksforme		cnewbold@…