HPXC initialization causes HPX to hang about hpxc

[hkaiser]

It may be better to store the thread id for hpx_main as a non-reference counting id (i.e., threads_id_type) in this case.

[rtohid]

Thank you, @Pansysk75. If I remember correctly, the issue was in hpxc_setspecific, where the thandle would be null for hpx_main, and what you're referring to was supposed to be the workaround.

[Pansysk75]

@hkaiser I was looking at a fix using hpx::threads::add_thread_exit_callback() to add a callback function to hpx_main. The callback doesn't seem to get called, even though hpx_main terminates :((
Should that work in theory, or am I not supposed to use it like that?

thread_handle(hpx::threads::thread_id_ref_type id)
:
. . .
{
        auto on_completion = [thandle = this]() {
            std::cout << "This print is not printing" << std::endl;
            // Signal completion
            thandle->promise.set_value(nullptr);
            int r = --thandle->refc;
            if (r == 0)
            {
                delete thandle;
            }
        };

        if (!hpx::threads::add_thread_exit_callback(id.noref(), on_completion))  <--------
        {
            // Failed to add callback because the hpx thread has terminated
            // This throw does not happen
            throw;
        }
}

The rationale of doing this, is to retain the ability to have hpxc-thread-data even for threads not launched through the hpxc API, and have that data be deleted when the thread finished (just like we do for the other threads spawned with hpxc_thread_create)

[hkaiser]

@Pansysk75 now I'm confused. I thought the whole problem was that hpx_main was not terminating, in which case it would be logical for the callback to never be called...

[Pansysk75]

@Pansysk75 now I'm confused. I thought the whole problem was that hpx_main was not terminating, in which case it would be logical for the callback to never be called...

Ah, it's terminating, holding the reference just keeps it from getting deleted after termination

[hkaiser]

@Pansysk75 now I'm confused. I thought the whole problem was that hpx_main was not terminating, in which case it would be logical for the callback to never be called...

Ah, it's terminating, holding the reference just keeps it from getting deleted after termination

Well, in this case the callback should be invoked (see: https://github.com/STEllAR-GROUP/hpx/blob/e7c31a49dd4e550a61134e498f3ecfb585d201cb/libs/core/threading_base/include/hpx/threading_base/register_thread.hpp#L72)