GeezerVison

I don't know if sem_clockwait() timeout argument can be modified by the function, for example if the wait is interrupted by a signal. In case of doubt, I prefer to compute abs_timeout before each call.

In the same file, pthread_cond_timedwait() absolute timeout is only computed once outside the loop. I understand that we can also compute abs_timeout once outside the loop.

The new implementation of time.sleep(), pysleep() function, now also uses an absolute timeout (timespec) computed once outside the loop. It uses clock_nanosleep() or nanosleep(). These functions have an argument for the remaining time, it's not used by pysleep().

sem_clockwait() documentation doesn't say that the absolute timeout is modified, which confirms that it can be only computed once:
https://www.gnu.org/software/libc/manual/html_node/Waiting-with-Explicit-Clocks.html

I don't know if sem_clockwait() can fail with ENOSYS if Python is built with a recent glibc and then run with an older glibc or a different kernel version.

Should we attempt to fallback to sem_timedwait() if sem_clockwait() "doesn't work"? How can we determine if sem_clockwait() "doesn't work"? So far, I failed to find a sem_clockwait() documentation or manual page.

In case of doubt, I prefer to leave the code as it is, and only implement a fallback once we can test it on a real system.

s/absolution/absolute

Nobody expects the Spanish Absolution!