Re: Weird test mixup

On 15/03/2024 14:10, Heikki Linnakangas wrote:
> On 15/03/2024 13:09, Heikki Linnakangas wrote:
>> I committed a patch to do that, to put out the fire.
> 
> That's turning the buildfarm quite red. Many, but not all animals are
> failing like this:
> 
>> ---
/home/buildfarm/hippopotamus/buildroot/HEAD/pgsql.build/src/test/modules/injection_points/expected/injection_points.out
 2024-03-15 12:41:16.363286975 +0100
 
>> +++
/home/buildfarm/hippopotamus/buildroot/HEAD/pgsql.build/src/test/modules/injection_points/results/injection_points.out
2024-03-15 12:53:11.528159615 +0100
 
>> @@ -1,118 +1,111 @@
>>   CREATE EXTENSION injection_points;
>> +ERROR:  extension "injection_points" is not available
>> +DETAIL:  Could not open extension control file
"/home/buildfarm/hippopotamus/buildroot/HEAD/pgsql.build/tmp_install/home/buildfarm/hippopotamus/buildroot/HEAD/inst/share/postgresql/extension/injection_points.control":
Nosuch file or directory.
 
>> +HINT:  The extension must first be installed on the system where PostgreSQL is running.
>> ...
> 
> Looks like adding NO_INSTALLCHECK somehow affected how the modules are
> installed in tmp_install. I'll investigate..

I think this is a bug in the buildfarm client. In the make_misc_check 
step, it does this (reduced to just the interesting parts):

> # run the modules that can't be run with installcheck
> sub make_misc_check
> {
>     ...
>     my @dirs = glob("$pgsql/src/test/modules/* $pgsql/contrib/*");
>     foreach my $dir (@dirs)
>     {
>         next unless -e "$dir/Makefile";
>         my $makefile = file_contents("$dir/Makefile");
>         next unless $makefile =~ /^NO_INSTALLCHECK/m;
>         my $test = basename($dir);
> 
>         # skip redundant TAP tests which are called elsewhere
>         my @out = run_log("cd $dir && $make $instflags TAP_TESTS= check");
>         ...
>     }

So it scans src/test/modules, and runs "make check" for all 
subdirectories that have NO_INSTALLCHECK in the makefile. But the 
injection fault tests are also conditional on the 
enable_injection_points in the parent Makefile:

> ifeq ($(enable_injection_points),yes)
> SUBDIRS += injection_points gin
> else
> ALWAYS_SUBDIRS += injection_points gin
> endif

The buildfarm client doesn't pay any attention to that, and runs the 
test anyway.

I committed an ugly hack to the subdirectory Makefiles, to turn "make 
check" into a no-op if injection points are disabled. Normally when you 
run "make check" at the parent level, it doesn't even recurse to the 
directories, but this works around the buildfarm script. I hope...

-- 
Heikki Linnakangas
Neon (https://neon.tech)