Expand waitid support, add waitid tests

[valpackett]

• add a bunch of stuff to WaitidStatus
• detect successful return with no process more carefully according to documentation
  • Linux manpage:

    If WNOHANG was specified in options and there were no children in a waitable state, then waitid() returns 0 immediately and the state of the siginfo_t structure pointed to by infop is unspecified. To distinguish this case from that where a child was in a waitable state, zero out the si_pid field before the call and check for a nonzero value in this field after the call returns.

  • BSD manpages don't really mention the unspecified thing e.g.:

    If WNOHANG was specified and waitid() didn't find a matching process, si_signo and si_pid will be set to zero.

• add tests
• would it be ok to underive Copy on WaitidStatus? Technically an API break that should go to 0.38 release planning #578? But WaitidStatus was currently completely unusable with nothing on it…

[valpackett]

wow, siginfo_t is such a royal mess… at least in the libc crate there are accessors, just some missing on some platforms, but for the raw linux backend… I guess we'll have to make an extension trait with accessors in backend::linux_raw::c?

[valpackett]

Oofff okay, now it builds everywhere and isn't actually too bad, tho some cfg()'s are quite silly. And stuff like the linux_raw accessor for si_status should probably be pushed down into the backend modules (?)

[sunfishcode]

Is the reason for explicit zeroing here the comment in the Linux man page about zeroing out the si_pid field? If so, could you quote that documentation here, so that it's recorded in the code why this uses zeroed memory?

[sunfishcode]

Rustix does have a few places where the public API for linux_raw differs from the public API for the libc backend on Linux, but we're working on eliminating them, as it makes code less portable. Is it complicated to add Linux support using the libc backend as well? Also, yeah, your suggestion to push these details into src/backend/* sounds good; perhaps in new src/backend/*/wait.rs files?

[valpackett]

It should already work with the libc backend (CI should test this, right?) — it's the final not(any(…)) version. The libc crate does make an effort to offer the same interface here on all platforms, just lacking in a few places such as rust-lang/libc#3185 and OpenBSD/NetBSD currently.

[sunfishcode]

Ah, I see now. I was confused by the cfgs :-).

[valpackett]

Okay, now it looks decent I guess…

[sunfishcode]

Cool, thanks!

[pawelchcki]

Thanks for this change. I was just looking for the missing waitidstatus methods! 🎉

[sunfishcode]

This is now released in rustix 0.37.8.