Boost C++ Libraries: Ticket #9708: boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite

owner, status changed

viboes — Tue, 25 Feb 2014 06:52:12 GMT

owner changed from Anthony Williams to viboes
status new → assigned

have you tried with

                if (boost::cv_status::timeout == scv.wait_for(l, boost::chrono::duration<int, std::ratio<1,1>>::max()))
                {
                        // wait_for return timeout. it means that time period has elapsed.
                        std::terminate();
                }

nikolay@… — Tue, 25 Feb 2014 07:27:18 GMT

bcv.wait_for(l, boost::chrono::duration<int, boost::ratio<1,1>>::max()) works good.

But the issue with timed_wait.

I have some code like:

void foo(boost::posix_time::time_duration& td)
{
...
cv.timed_wait(..., td);
...
}
void f1()
{
...
foo(boost::posix_time::time_duration(boost::posix_time::pos_infin));
...
}
void f2()
{
...
foo(boost::posix_time::minutes(2));
...
}
void f3()
{
...
foo(boost::posix_time::milliseconds(500));
...
}

viboes — Thu, 27 Feb 2014 22:23:47 GMT

I suspect that boost::posix_time::pos_infin can be seen as a negative duration. Could you confirm?

anonymous — Fri, 28 Feb 2014 03:59:38 GMT

boost::posix_time::pos_infin - positive infinity
boost::posix_time::neg_infin - negative infinity

time_duration td(pos_infin);
td.is_pos_infinity(); --> true
td.is_negative(); --> false

As I see possible this issue may be caused by some changes inside Boost.DateTime library. I found new note in the Boost.DateTime documentation:

When a time_duration is a special value, either by construction or other means, the following accessor functions will give unpredictable results: hours(), minutes(), seconds(), ticks(), fractional_seconds(), total_nanoseconds(), total_microseconds(), total_milliseconds(), total_seconds()

So it seems that Boost.Thread incorrectry call total_milliseconds() inside timed_wait on duration which is special value. Possible code in timed_wait should be like:

        template<typename duration_type>
        bool timed_wait(unique_lock<mutex>& m,duration_type const& wait_duration)
        {
            if (wait_duration.is_pos_infinity())
            {
                wait(m); // or do_wait(m,detail::timeout::sentinel());
                return true;
            }
            if (wait_duration.is_special_value())
            {
                        //throw some error
            }
            return do_wait(m,wait_duration.total_milliseconds());
        }

viboes — Fri, 28 Feb 2014 22:19:29 GMT

timed_value is deprecated. Even if the patch is simple, it is not complete. If you provide a complete patch with tests and the needed documentation I would apply it.

type changed

viboes — Sat, 01 Mar 2014 07:45:57 GMT

type Bugs → Feature Requests

summary changed

viboes — Sat, 01 Mar 2014 07:46:39 GMT

summary regression: boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite → Align to new Boost.DtaeTime behavior (was regression: boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite)

type, summary changed

viboes — Sat, 01 Mar 2014 08:16:06 GMT

type Feature Requests → Bugs
summary Align to new Boost.DtaeTime behavior (was regression: boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite) → was regression: boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite

summary changed

viboes — Sat, 01 Mar 2014 08:16:39 GMT

summary was regression: boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite → boost::condition_variable::timed_wait unexpectedly wakes up while should wait infinite

nikolay@… — Mon, 03 Mar 2014 03:54:08 GMT

What is "timed_value"? If you mean "timed_wait" then I have failed to find any notes in the Boost.Thread documentation that this function is deprecated.

Anyway currently situation looks like following:

there was bug in Boost.Thread in incorrect usage of Boost.DateTime
in previous releases it was lucky that code in Boost.Thread works
in latest releases this issue (incorrect usage of Boost.DateTime) is not hidden and Boost.Thread code does not work as expected
Boost.Thread users have an regression after update to the latest boost releases
I have suggested some changes to fix the problem in the Boost.Thread. I know that it is not final code, but I hope it is a good start point to Boost.Thread authors/sustainers to make an correct and complete changes
you can fix this issue by any other changes (even with breaking changes), for example, you can always throw something like std::invalid_arg in case if wait_duration.is_special_value(). It is your responsibility to chose correct fix from the view of the Boost.Thread library

P.S. I am not familiar with git so now I have lose an ability to generate any formal patches to Boost...

milestone changed

viboes — Wed, 12 Mar 2014 22:50:35 GMT

milestone To Be Determined → Boost 1.56.0

https://github.com/boostorg/thread/commit/68dc454a665a54ab505336860401195dc356fd48

status changed; resolution set

viboes — Wed, 12 Mar 2014 23:01:20 GMT

status assigned → closed
resolution → fixed