Re: [RFC] building postgres with meson - v13

On Sun, Oct 02, 2022 at 01:38:37PM -0500, Justin Pryzby wrote:
> On Sun, Oct 02, 2022 at 11:05:30AM -0700, Andres Freund wrote:
> > > Also, you wrote "rm -fr build" between building for gcc and clang, but
> > > since they run in an "always" block, it'd be better to use separate
> > > dirs, to allow seeing logs for the the all (failed) tasks, in case the
> > > last one succeeds.
> >
> > Hm, when are logs important for CompilerWarnings? I don't think we even
> > collect any? Using a different builddir for the "sibling" tests (i.e. the two
> > gcc and the two clang tests) would increase the times a bit because we'd
> > regenerate the bison files etc.
> >
> > I guess it'll look a bit cleaner to use a build-gcc and a build-clang, just to
> > get rid of the irregularity of needing that rm -rf.
>
> The build logs are important when hacking on .cirrus.yml itself.
>
> You're right that we don't normally save logs for CompilerWarnings; one or
> another (unpublished) patch of mine adds that, and then also needed to change
> to use separate dirs in order to debug building while experimenting with your
> patch to use meson.

FYI, this is what led me to make that suggestion.

https://cirrus-ci.com/task/5920691940753408

I had a patch laying around to change the "compiler warnings" task to
use debian "testing", which seems to have added some new flags in -Wall,
which caused me to add (for now) some compiler flags like -Wno-error=...

But when I added them to the task's CFLAGS, it broke "clang" (which
doesn't support the warnings) in an obscure way[0], and no logs
available to show why.

[0] Header "uuid/uuid.h" has symbol "uuid_generate" with dependency
uuid: NO

So, I think it's worth reporting meson's build logs, even though no
tests are run here.

--
Justin