Re: Cleaning up historical portability baggage

On Sat, Aug 6, 2022 at 2:54 AM Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Fri, Aug 5, 2022 at 10:48 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > Hmm ... I agree with you that the end result could be nicer code,
> > but what's making it nicer is a pretty substantial amount of human
> > effort for each and every call site. Is anybody stepping forward
> > to put in that amount of work?
> >
> > My proposal is to leave the call sites alone until someone feels
> > like doing that sort of detail work.
>
> My plan was to nerd-snipe Thomas Munro into doing it.[1]

Alright, well here's a patch for the setsid() stuff following Robert's
plan, which I think is a pretty good plan.

Did I understand correctly that the places that do kill(-pid) followed
by kill(pid) really only need the kill(-pid)?

I checked that user processes should never have pid 0 (that's a
special system idle process) or 1 (because they're always even,
actually it looks like they're pointers in kernel space or something
like that), since those wouldn't play nice with the coding I used
here.

I note that Windows actually *does* have process groups (in one of the
CI threads, we learned that there were at least two concepts like
that). Some of our fake signals turn into messages to pipes, and
others turn into process termination, and in theory we could probably
also take advantage of Windows' support for its native "control C" and
"control break" signals here. It's possible that someone could do
something to make all but the pipe ones propagate to real Windows
process groups. That might be good for things like nuking archiver
subprocesses and the like when taking out the backend, or something
like that, but I'm not planning to look into that myself.