std::thread::sleep does not document its interaction with signals

[GKFX]

Location

std::thread::sleep

Summary

POSIX permits the C function sleep to be implemented using the SIGARLM signal (sleep(3)), meaning it is non-portable to mix use of that signal with sleep. The nanosleep function, which std::thread::sleep uses, is required by POSIX not to interfere with signals (nanosleep(2)), so it should in fact be OK currently to mix std::thread::sleep with SIGALRM. However, this is not actually documented. Given that the Rust function is called "sleep" it is easy to be concerned that there might be a problem.

Could std::thread::sleep make a documented commitment not to interfere with SIGALRM?

[workingjubilee]

The documentation is out of date, at least one platform implements it via usleep.

[workingjubilee]

The same website documents this as unspecified: https://man7.org/linux/man-pages/man3/usleep.3.html#CAVEATS

[workingjubilee]

Interestingly, @ivmarkov notes that ESP-IDF doesn't support signals and thus the interaction of usleep with them is irrelevant, but I feel less confident that such should apply to all platforms past and future?

I also feel less confident that all platforms that claim to implement POSIX correctly implement POSIX.

[workingjubilee]

@GKFX Do you actually need a documented commitment (true going forward) or do you simply want a statement that currently thread::sleep does not affect SIGALRM on any platform which also implements POSIX signals?