Usage of semaphore

[Flw5469]

Code:

#include "roc_core/semaphore.h"

TEST(state_tracker, semaphore_test) {
    core::Semaphore sem(0);
    if (sem.timed_wait(50 * core::Second + core::timestamp(core::ClockMonotonic)))
        roc_log(LogDebug, "true");
        else roc_log(LogDebug, "false");
} 

Output:

TEST(state_tracker, semaphore_test)00:42:26.990 [14054] [dbg] roc_test: false

• 401 ms
  OK (183 tests, 1 ran, 0 checks, 0 ignored, 182 filtered out, 401 ms)

The sem.timed_wait function does not wait for 50 seconds, and it returns false indicating that it is timeout. Am I using it wrongly? I was following the practice in src/internal_modules/roc_core/timer.cpp

Thanks!

[gavv]

You've found a bug, thanks! You're on linux, right?

deadline parameter of Semaphore::timed_wait is supposed to be absolute time in CLOCK_MONOTONIC domain. The code that uses it, as well as your example, pass monotonic timestamps to it.

However in roc_core/target_posix_ext/roc_core/semaphore.cpp we pass deadline to sem_timedwait, which expects time in CLOCK_REALTIME domain (i.e. since epoch).

Hence it returns immediately - monotonic timestamps are typically smaller than wallclock timestamps, and it thinks that deadline is already expired.

I've created an issue: #795

You can fix it by yourself (it should be straightforward I guess), otherwise you could wait until I take a look, I think I'll have time next week. (Please let me know if you're waiting for a fix.)

[Flw5469]

You're on linux, right?

yes I am on Ubuntu 20.04.5 LTS in WSL2.

You can fix it by yourself (it should be straightforward I guess), otherwise you could wait until I take a look, I think I'll have time next week. (Please let me know if you're waiting for a fix.)

I will start trying to fix it on Friday, I need this for #749

thanks!

You are welcome :D

[gavv]

Thanks, it'd help if you can create a separate PR for this bug.