Re: Cleaning up historical portability baggage

On Sun, Aug 14, 2022 at 10:36 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
> On 2022-08-14 10:03:19 +1200, Thomas Munro wrote:
> > I hadn't paid attention to our existing abstract Unix socket support
> > before and now I'm curious: do we have a confirmed sighting of that
> > working on Windows?
>
> I vaguely remember successfully trying it in the past. But I just tried it
> unsuccessfully in a VM and there's a bunch of other places saying it's not
> working...
> https://github.com/microsoft/WSL/issues/4240

I think we'd better remove our claim that it works then. Patch attached.

We could also reject it, I guess, but it doesn't immediately seem
harmful so I'm on the fence. On the Windows version that Cirrus is
running, we happily start up with:

2022-08-13 20:44:35.174 GMT [4760][postmaster] LOG: listening on Unix
socket "@c:/cirrus/.s.PGSQL.61696"

... and then client processes apparently can't see it, which is
confusing but, I guess, defensible if we're only claiming it works on
Linux. We don't go out of our way to avoid this feature on a per-OS
basis in general, though at least on a typical Unix system it fails
fast. For example, my FreeBSD system here barfs:

2022-08-15 13:26:13.483 NZST [29956] LOG: could not bind Unix address
"@/tmp/.s.PGSQL.5432": No such file or directory

... because the kernel just sees an empty string and can't locate the
parent directory.