On Thu, Apr 08, 2021 at 02:58:02AM -0400, Tom Lane wrote:
> Julien Rouhaud <rjuju123@gmail.com> writes:
> > On Wed, Apr 07, 2021 at 11:33:20PM -0700, Andres Freund wrote:
> >> Nothing special, really. Surprised the BF doesn't see it:
>
> > Is think this is because the buildfarm client is running installcheck for the
> > contribs rather than check, and pg_stat_statements/Makefile has:
> > # Disabled because these tests require "shared_preload_libraries=pg_stat_statements",
> > # which typical installcheck users do not have (e.g. buildfarm clients).
> > NO_INSTALLCHECK = 1
>
> No, because if that were the explanation then we'd be getting no
> buildfarm coverage at all for for pg_stat_statements. Which aside
> from being awful contradicts the results at coverage.postgresql.org.
Is there any chance that coverage.postgresql.org isn't backed by the buildfarm
client but a plain make check-world or something like that?
> I think Andres has the right idea that there's some more-subtle
> variation in the test conditions, but (yawn) too tired to look
> into it right now.
I tried to look at some force-parallel-mode animal, like mantid, and I don't
see any evidence of pg_stat_statements being run by a "make check", and only a
few contrib modules seem to have an explicit check phase. However, looking at
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=mantid&dt=2021-04-08%2007%3A07%3A05
I see
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=mantid&dt=2021-04-08%2007%3A07%3A05&stg=contrib-install-check-C:
[...]
make -C pg_stat_statements installcheck
make[1]: Entering directory `/u1/tac/build-farm-11/buildroot/HEAD/pgsql.build/contrib/pg_stat_statements'
make[1]: Nothing to be done for `installcheck'.