Fix use of uninitialized memory in pcntl SIGCHLD handling

[iluuu1994]

psig needs to stay the tail, so that we don't get a dangling element on the end.

[nielsdos]

I'm a bit lost in understanding where the uninitialized memory problem is, and how this fixes it. I need a bit of help understanding this, sorry :/

I have one comment though: the reason I had a goto single_signal; is because I was afraid that a failure of waitpid in the first iteration will cause the psig fields to be uninitialized.

[iluuu1994]

No worries! For context: https://github.com/php/php-src/actions/runs/5372867556/jobs/9746699882

The problem is that previously, we would run psig = psig->next; at the end of the loop for SIGCHLD. This means that, when the loop is finished, we're not pointing to the last element of the used signals, but to the fist unused one. However, the code below it treats psig as the last used one, using its next as the new PCNTL_G(spares). So, we have an unused element at the end of the signal chain, which has its signo uninitialized, thus causing the MSAN failure.

[nielsdos]

Ooooh I get it. Thanks.
Yeah, then this patch looks good, except for the single signal case. It is possible to hit if we get a race where one thread consumes all children in the waitpid loop.

[nielsdos]

Thanks.
Code looks okay. Im happy if CI is happy with this