Re: [RFC] building postgres with meson

Hi,

On 2021-10-19 15:22:15 -0400, Tom Lane wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
> > On 2021-10-12 01:37:21 -0700, Andres Freund wrote:
> >> As far as I can tell the only OS that postgres currently supports that
> >> meson doesn't support is HPUX. It'd likely be fairly easy to add
> >> gcc-on-hpux support, a chunk more to add support for the proprietary
> >> ones.
>
> > Tom, wrt HPUX on pa-risc, what are your thoughts there? IIRC we gave up
> > supporting HP's compiler on pa-risc a while ago.
>
> Right. I am still testing with gcc on HP-PA. I'd kind of like to
> keep it running just as an edge case for our spinlock support, but
> I'm not sure that I want to do any huge amount of work on meson
> to keep that going.

Makes sense. While that does test an odd special case for our spinlock
implementation, it's also the only supported platform with that edge case, and
it seems extremely unlikely that there ever will be a new platform with such
odd/limited atomic operations.

> I do have a functioning OpenBSD installation on that machine, so
> one alternative if the porting costs look too high is to replace
> gaur with an OpenBSD animal. However, last I checked, OpenBSD
> was about half the speed of HPUX on that hardware, so I'm not
> real eager to go that way. gaur's already about the slowest
> animal in the farm :-(

Yea, that doesn't sound enticing. Seems like we either should keep it running
on hp-ux or just drop parisc support?

> > As I said it'd probably not be too hard to add meson support for hpux on hppa,
> > it's probably just a few branches. But that'd require access somewhere. The
> > gcc compile farm does not have a hppa member anymore...
>
> If you've got an idea where to look, I could add that to my to-do queue.

It might even just work. Looks like meson does have pa-risc detection. While
it doesn't have any specifically for hpux, it just falls back to python's
sys.platform in that case. python3 -c 'import sys;print(sys.platform)'

meson generates output for ninja to execute (basically a faster make that's
partially faster by being much less flexible. Intended to be output by more
user-friendly buildsystems ). Ninja can be built by a minimal python script,
or with cmake. The former doesn't seem to have hpux support, the latter does I
think.
https://github.com/ninja-build/ninja

So it could be interesting to see if ninja builds.

I've not taught the PG meson the necessary stuff for a 32 bit build. So
there's no point is trying whether meson works that much. I'll try to do that,
and let you know.

> I'm more concerned about the effort involved in getting meson going on some
> other old animals, such as prairiedog.

Yea, that's an *old* OS version. One version too old to have support for
@rpath, added in 10.5 :(. Is there a reason to run 10.4 specifically?
According to wikipedia 10.5 is the last version to support ppc.

Looks like python still supports building back to 10.4.

Greetings,

Andres Freund