Buildfarm misses running some contrib TAP tests

I noticed by accident that my buildfarm animal indri stopped running
certain contrib TAP tests this morning. In [1] it's running

contrib-amcheck-check (00:00:12)
contrib-auto_explain-check (00:00:02)
contrib-basebackup_to_shell-check (00:00:01)
contrib-bloom-check (00:00:05)
contrib-oid2name-check (00:00:01)
contrib-pg_stat_statements-check (00:00:02)
contrib-postgres_fdw-check (00:00:07)
contrib-test_decoding-check (00:00:07)
contrib-vacuumlo-check (00:00:01)

but in the very next run [2] we see

contrib-amcheck-check (00:00:12)
contrib-auto_explain-check (00:00:01)
contrib-basebackup_to_shell-check (00:00:02)
contrib-bloom-check (00:00:05)
contrib-oid2name-check (00:00:00)
contrib-pg_stash_advice-check (00:00:03)
contrib-pg_stat_statements-check (00:00:02)
contrib-pg_visibility-check (00:00:03)
contrib-test_decoding-check (00:00:08)

What became of postgres_fdw and vacuumlo? And why wasn't
pg_visibility being run before? And why are dblink and pg_prewarm
visible in neither list? Apparently the addition of pg_stash_advice
changed something here, but what?

Looking at some other BF animals shows that it's not just indri: other
autoconf-based animals are showing misbehavior of this sort as well.

I poked at this by adding some debug printouts, and determined that
what is going wrong is the test to see if we built the module:

# can't test it if we haven't built it
next unless scalar glob("$testdir/*.o $testdir/*.obj");

It's failing, sometimes, on modules that definitely do contain object
files. We've had run-ins with "scalar glob()" before [3], and when
I finally looked at "man perlfunc" what I read is

glob EXPR
glob
In list context, returns a (possibly empty) list of filename
expansions on the value of EXPR such as the standard Unix shell
/bin/csh would do. In scalar context, glob iterates through such
filename expansions, returning undef when the list is exhausted.

I'm not entirely sure what "iterates" means in this context, but
what seems to be happening on my Linux box is that you get undef
unless there is exactly one file matching the glob pattern.

I can't explain why we're not seeing more consistent behavior
out of the buildfarm, like never running postgres_fdw at all.
I wonder if the glob() infrastructure has some buggy internal state.
But the attached patch gives me better results.

regards, tom lane

[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=indri&dt=2026-04-07%2013%3A42%3A46
[2] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=indri&dt=2026-04-07%2014%3A17%3A01
[3] https://www.postgresql.org/message-id/200085.1741127335%40sss.pgh.pa.us