kill(process) should throw an exception rather than return -errno?

[stevengj]

kill(p) currently returns 0 on success, -errno on failure, and -1 if the process was not running. (This is undocumented, by the way.) Shouldn't it throw an exception on failure instead?

[StefanKarpinski]

Proposal from @Keno:

• return true/false indicating if the process was running.
• raise an error if the process failed.

I would add that we'd probably want a force=false keyword argument.

@JeffBezanson proposes renaming to sigsend or sendsignal (or something).

@vtjnash suggested notify as a potential function name.

[JeffBezanson]

To continue quoting each other, @StefanKarpinski proposes something like kill(p) ensures the process stops in whatever way, e.g. succeeding if the process is already dead, while another function has lower-level signal-sending semantics.

[Keno]

Final proposal: Throw an error on EPERM and EINVAL, but silently allow ESRCH. On linux we should never get that because the kernel gives success if the process is a zombie. On windows it's racy, so we might get it.

[StefanKarpinski]

Can't you still get ESRCH if you've already successfully waited on the zombie previously? I.e. this should always work the first time, but could fail if you do it again.