Обсуждение: meson: Make test output much more useful on failure (both in CI and locally)

I've gotten totally fed up with the amount of clicking I need to do
every time a test fails in CI to find out what exactly failed. The
output that meson gives in its main output is completely useless
currently. Which means I have to go click through CirrusCI its (pretty
terrible) filebrowser UI searching for log files or regression.diffs.

And locally I have the exact same problem when a perl test is failing.
Meson often shows no useful output at all, which means I have to find
some unknown log file somewhere in my build directory. And then I have
to find the useful spot in that log file that actually tells me what the
error was.

So attached are a few small patches that greatly improve the test output
of the most common test failure reasons (at least the ones I tend to run
into myself).

An example CI run with some intentional failures (attached as the
nocfbot patch) can be found here:
https://cirrus-ci.com/task/6592551433535488

Hello,

This is really useful!

- my $result = run_log($cmd);
- ok($result, $test_name);
+ # Doesn't rely on detecting end of file on the file descriptors,
+ # which can fail, causing the process to hang, notably on Msys
+ # when used with 'pg_ctl start'

The function comments still refer to run_log, for both command_ok and
command_fails

Hi,

On 2026-01-26 11:13:57 +0100, Jelte Fennema-Nio wrote:
> I've gotten totally fed up with the amount of clicking I need to do
> every time a test fails in CI to find out what exactly failed. The
> output that meson gives in its main output is completely useless
> currently.

This doesn't really seem meson specific, right? It just seems about the output
of our own test tooling?

Meson can't really output something that we hide from it...

I agree that what we currently do is pretty unhelpful.


> Subject: [PATCH v1 1/5] meson: Include pg_regress diffs in meson output

This isn't about meson, it's about adding a new argument to pg_regress.


I think we'd need to limit the size of the output here somewhat. If e.g. the
server crashes, you end up with all subsequent tests failing and printing out
a couple hundred kB of pointless diff output.


> @@ -1521,22 +1523,40 @@ results_differ(const char *testname, const char *resultsfile, const char *defaul
>       * append to the diffs summary file.
>       */
>  
> -    /* Write diff header */
> -    difffile = fopen(difffilename, "a");
> +    difffile = fopen(difffilename, "a+");
>      if (difffile)
>      {
> +        fseek(difffile, 0, SEEK_END);
> +        startpos = ftell(difffile);
>          fprintf(difffile,
>                  "diff %s %s %s\n",
>                  pretty_diff_opts, best_expect_file, resultsfile);
> +        fflush(difffile);
> +
> +        /* Run diff */
> +        snprintf(cmd, sizeof(cmd),
> +                 "diff %s \"%s\" \"%s\" >> \"%s\"",
> +                 pretty_diff_opts, best_expect_file, resultsfile, difffilename);
> +        run_diff(cmd, difffilename);
> +
> +        /* Emit diff as TAP diagnostics if requested */
> +        if (print_diffs)
> +        {
> +            char        line[1024];
> +
> +            fseek(difffile, startpos, SEEK_SET);
> +            while (fgets(line, sizeof(line), difffile))
> +            {
> +                size_t        len = strlen(line);
> +
> +                if (len > 0 && line[len - 1] == '\n')
> +                    line[len - 1] = '\0';
> +                diag("%s", line);
> +            }
> +        }
>          fclose(difffile);
>      }

I'm a bit confused. Why are we rerunning diffs and appending to the diff file?


> From c1db07a8bba54e6b903766585f1b521ef526de12 Mon Sep 17 00:00:00 2001
> From: Jelte Fennema-Nio <postgres@jeltef.nl>
> Date: Mon, 26 Jan 2026 09:09:11 +0100
> Subject: [PATCH v1 2/5] perl tap: Show failed command output
> 
> This adds the output of failed commands to the TAP output. Before a
> failed libpq_pipeline test would look like this:
> 
>   Failed test 'libpq_pipeline cancel'
>   at /home/jelte/work/postgres-3/src/test/modules/libpq_pipeline/t/001_libpq_pipeline.pl line 55.
> 
> Now you can actually see the reason of the failure:
> 
>   Failed test 'libpq_pipeline cancel'
>   at /home/jelte/work/postgres-3/src/test/modules/libpq_pipeline/t/001_libpq_pipeline.pl line 55.
> ----- command failed -----
> libpq_pipeline -r 700 cancel port=14309 host=/tmp/htMib451qD dbname='postgres' max_protocol_version=latest
> --------- stderr ---------
> test cancellations...
> libpq_pipeline:315: unexpected number of rows received: 1
> --------------------------
> ---
>  src/test/perl/PostgreSQL/Test/Utils.pm | 36 +++++++++++++++++++++++---
>  1 file changed, 32 insertions(+), 4 deletions(-)
> 
> diff --git a/src/test/perl/PostgreSQL/Test/Utils.pm b/src/test/perl/PostgreSQL/Test/Utils.pm
> index ff843eecc6e..bd1e981c6f0 100644
> --- a/src/test/perl/PostgreSQL/Test/Utils.pm
> +++ b/src/test/perl/PostgreSQL/Test/Utils.pm
> @@ -955,8 +955,22 @@ sub command_ok
>  {
>      local $Test::Builder::Level = $Test::Builder::Level + 1;
>      my ($cmd, $test_name) = @_;
> -    my $result = run_log($cmd);
> -    ok($result, $test_name);
> +    # Doesn't rely on detecting end of file on the file descriptors,
> +    # which can fail, causing the process to hang, notably on Msys
> +    # when used with 'pg_ctl start'
> +    my $stdoutfile = File::Temp->new();
> +    my $stderrfile = File::Temp->new();

I'm afraid that the creation of this many additional tempfiles would slow down
the tests on some platforms...

> +    my $result = IPC::Run::run $cmd, '>' => $stdoutfile, '2>' => $stderrfile;
> +    ok($result, $test_name) or do
> +    {
> +        my $stdout = slurp_file($stdoutfile);
> +        my $stderr = slurp_file($stderrfile);
> +        diag("----- command failed -----");
> +        diag(join(" ", @$cmd));
> +        diag("--------- stdout ---------"), diag($stdout) if $stdout;
> +        diag("--------- stderr ---------"), diag($stderr) if $stderr;
> +        diag("--------------------------");
> +    };

Outputting the entire file seems to again have the danger of completely
swamping the output.

>      return;
>  }
>  
> @@ -972,8 +986,22 @@ sub command_fails
>  {
>      local $Test::Builder::Level = $Test::Builder::Level + 1;
>      my ($cmd, $test_name) = @_;
> -    my $result = run_log($cmd);
> -    ok(!$result, $test_name);
> +    # Doesn't rely on detecting end of file on the file descriptors,
> +    # which can fail, causing the process to hang, notably on Msys
> +    # when used with 'pg_ctl start'

Huh? Any more details?

> +    my $stdoutfile = File::Temp->new();
> +    my $stderrfile = File::Temp->new();
> +    my $result = IPC::Run::run $cmd, '>' => $stdoutfile, '2>' => $stderrfile;
> +    ok(!$result, $test_name) or do
> +    {
> +        my $stdout = slurp_file($stdoutfile);
> +        my $stderr = slurp_file($stderrfile);
> +        diag("-- command succeeded unexpectedly --");
> +        diag(join(" ", @$cmd));
> +        diag("-------------- stdout --------------"), diag($stdout) if $stdout;
> +        diag("-------------- stderr --------------"), diag($stderr) if $stderr;
> +        diag("------------------------------------");
> +    };

I don't think it makes sense to duplicate the logic for this multiple times.


> From c88f1fbf6462da56a1f0739dd83a6284d6bdd389 Mon Sep 17 00:00:00 2001
> From: Jelte Fennema-Nio <postgres@jeltef.nl>
> Date: Mon, 26 Jan 2026 09:32:15 +0100
> Subject: [PATCH v1 3/5] perl tap: Show die reason in TAP output
>...
> diff --git a/src/test/perl/PostgreSQL/Test/Utils.pm b/src/test/perl/PostgreSQL/Test/Utils.pm
> index bd1e981c6f0..df328d0f571 100644
> --- a/src/test/perl/PostgreSQL/Test/Utils.pm
> +++ b/src/test/perl/PostgreSQL/Test/Utils.pm
> @@ -206,6 +206,14 @@ INIT
>      # test may still fail, but it's more likely to report useful facts.
>      $SIG{PIPE} = 'IGNORE';
>  
> +    # Emit die messages as TAP diagnostics so they appear in test output.
> +    $SIG{__DIE__} = sub {
> +        return if $^S;    # Ignore dies inside eval
> +        my $msg = shift;
> +        chomp $msg;
> +        diag("die: $msg");
> +    };
> +

This looked familiar - turns out I had complained about this in the past and
then forgotten about it :(

https://www.postgresql.org/message-id/20220222181924.eehi7o4pmneeb4hm%40alap3.anarazel.de

At least at the time I found it necessary to separately test for $^S, to avoid
triggering the logic for syntax errors.  Not sure why I concluded that though.

In a reply Andrew suggested to use done_testing(), which avoids the pointless
  "Tests were run but no plan was declared and done_testing() was not seen."
which does seem like an improvement.


> From a6b324678293d65f3a417b8f66889b3f3f6a9c29 Mon Sep 17 00:00:00 2001
> From: Jelte Fennema-Nio <postgres@jeltef.nl>
> Date: Mon, 26 Jan 2026 10:04:44 +0100
> Subject: [PATCH v1 4/5] perl tap: Include caller in die messages of psql
> 
> This way by looking at the error output you can actually figure out
> where it occured, instead of just knowing that is in the very often
> called psql function.

I think this is too localized a fix. Right now every die in a .pm file has
this issue.  The easiest fix would probably be to just replace all uses of die
in .pm files with croak (which we already, inconsistently, use).

stderr:
# die: error running SQL: 'psql:<stdin>:1: ERROR:  syntax error at or near "I"
# LINE 1: I am the worst
#         ^'
# while running 'psql --no-psqlrc --no-align --tuples-only --quiet --dbname port=20878 host=/tmp/K5V48mzFx9
dbname='postgres'--file - --variable ON_ERROR_STOP=1' with sql 'I am the worst' at
/home/andres/src/postgresql/src/bin/scripts/t/020_createdb.plline 19.
 

Note that the failure now is reported as coming from the actual test file.


We could also use Test::More->builder->croak(), but that's a heck of a lot
longer, and would require us to be more liberal about sprinkling

    local $Test::Builder::Level = $Test::Builder::Level + 1;
around. I don't see an improvement in the output either.

In contrast Carp.pm just looks for a caller outside of the current package. If
we are worried about errors across Cluster.pm/Utils.pm, we could just fix that
in one place by using @CARP_NOT.

Maybe there's a way to do that more centrally?

Greetings,

Andres Freund

On Mon, 26 Jan 2026 at 17:42, Andres Freund <andres@anarazel.de> wrote:
> This doesn't really seem meson specific, right? It just seems about the output
> of our own test tooling?

Yes, it's general improvements. But the problem is worse in meson than
for because that really only shows the TAP output. The Makefile based
builds will show a bit more information currently.

> I think we'd need to limit the size of the output here somewhat. If e.g. the
> server crashes, you end up with all subsequent tests failing and printing out
> a couple hundred kB of pointless diff output.

Fair enough.

> I'm a bit confused. Why are we rerunning diffs and appending to the diff file?

Now I'm confused. We're running the same amount of diffs as before.
Did you miss the removal of the /* Run diff */ below the block that
you quoted?

> I'm afraid that the creation of this many additional tempfiles would slow down
> the tests on some platforms...
< snip>
> > @@ -972,8 +986,22 @@ sub command_fails
> >  {
> >       local $Test::Builder::Level = $Test::Builder::Level + 1;
> >       my ($cmd, $test_name) = @_;
> > -     my $result = run_log($cmd);
> > -     ok(!$result, $test_name);
> > +     # Doesn't rely on detecting end of file on the file descriptors,
> > +     # which can fail, causing the process to hang, notably on Msys
> > +     # when used with 'pg_ctl start'
>
> Huh? Any more details?

I copy pasted that comment and the pipe to tempfiles from
command_like_safe. I did this after I initially tried to pipe to
variables, but that made the test that we run in SanityCheck get stuck
forever.

I don't get why this is needed either.

> > +     my $stdoutfile = File::Temp->new();
> > +     my $stderrfile = File::Temp->new();
> > +     my $result = IPC::Run::run $cmd, '>' => $stdoutfile, '2>' => $stderrfile;
> > +     ok(!$result, $test_name) or do
> > +     {
> > +             my $stdout = slurp_file($stdoutfile);
> > +             my $stderr = slurp_file($stderrfile);
> > +             diag("-- command succeeded unexpectedly --");
> > +             diag(join(" ", @$cmd));
> > +             diag("-------------- stdout --------------"), diag($stdout) if $stdout;
> > +             diag("-------------- stderr --------------"), diag($stderr) if $stderr;
> > +             diag("------------------------------------");
> > +     };
>
> I don't think it makes sense to duplicate the logic for this multiple times.

It seemed a bit much to make it generic over the input, as it would
need to remove dashes accordingly. Especially since it only has two
copies.

> This looked familiar - turns out I had complained about this in the past and
> then forgotten about it :(
>
> https://www.postgresql.org/message-id/20220222181924.eehi7o4pmneeb4hm%40alap3.anarazel.de
>
> At least at the time I found it necessary to separately test for $^S, to avoid
> triggering the logic for syntax errors.  Not sure why I concluded that though.
>
> In a reply Andrew suggested to use done_testing(), which avoids the pointless
>   "Tests were run but no plan was declared and done_testing() was not seen."
> which does seem like an improvement.

I'll take a look at doing that.

> I think this is too localized a fix. Right now every die in a .pm file has
> this issue.  The easiest fix would probably be to just replace all uses of die
> in .pm files with croak (which we already, inconsistently, use).

I didn't even know that function existed (due to my pretty much
non-existent knowlegde of perl).

On Mon Jan 26, 2026 at 5:42 PM CET, Andres Freund wrote:
> I agree that what we currently do is pretty unhelpful.

Attached is v2, which I think fixes all the issues that you mentioned.

Hi,

Thank you for working on this! I think this is a nice improvement.

On Tue, 27 Jan 2026 at 17:48, Jelte Fennema-Nio <postgres@jeltef.nl> wrote:
>
> On Mon Jan 26, 2026 at 5:42 PM CET, Andres Freund wrote:
> > I agree that what we currently do is pretty unhelpful.
>
> Attached is v2, which I think fixes all the issues that you mentioned.

v2-0001:

This doesn't really read the first 80 lines of the combined diff.
Lines with more than 1023 characters will be counted and printed as
multiple lines. I don't think this is a problem but clearing this
would be helpful. Other than that, it looks good to me.

v2-0002:

I am just curious if test failure can cause file descriptors not being
properly closed and then the IPC::Run:run() call will hang
indefinitely like the pg_ctl case. Other than that, it looks good to
me.

v2-0003 and v2-0004 look good to me.

-- 
Regards,
Nazir Bilal Yavuz
Microsoft

On Tue, 3 Feb 2026 at 11:20, Nazir Bilal Yavuz <byavuz81@gmail.com> wrote:
> v2-0001:
>
> This doesn't really read the first 80 lines of the combined diff.
> Lines with more than 1023 characters will be counted and printed as
> multiple lines. I don't think this is a problem but clearing this
> would be helpful. Other than that, it looks good to me.

I was annoyed by this, so I actually fixed it (manually tested by
changing the buffer to 10 bytes). This meant I needed to introduce
DIAG_DETAIL and DIAG_END, like we already had NOTE_DETAIL and NOTE_END.

> v2-0002:
>
> I am just curious if test failure can cause file descriptors not being
> properly closed and then the IPC::Run:run() call will hang
> indefinitely like the pg_ctl case. Other than that, it looks good to
> me.

No, this issue only happens because for the 'pg_ctl restart' process,
its the child postmaster process outlives the pg_ctl process, WHILE
still having the stdout/stderr file descriptor open. As long as all the
subprocesses exit, the file descriptor will be closed by the OS
automatically. So, this problem really only occurs in these daemon
spawning situations. Which for us in practice is only for pg_ctl
start/restart.

On 2026-02-06 Fr 3:44 AM, Jelte Fennema-Nio wrote:
> On Tue, 3 Feb 2026 at 11:20, Nazir Bilal Yavuz <byavuz81@gmail.com> 
> wrote:
>> v2-0001:
>>
>> This doesn't really read the first 80 lines of the combined diff.
>> Lines with more than 1023 characters will be counted and printed as
>> multiple lines. I don't think this is a problem but clearing this
>> would be helpful. Other than that, it looks good to me.
>
> I was annoyed by this, so I actually fixed it (manually tested by
> changing the buffer to 10 bytes). This meant I needed to introduce
> DIAG_DETAIL and DIAG_END, like we already had NOTE_DETAIL and NOTE_END.
>
>> v2-0002:
>>
>> I am just curious if test failure can cause file descriptors not being
>> properly closed and then the IPC::Run:run() call will hang
>> indefinitely like the pg_ctl case. Other than that, it looks good to
>> me.
>
> No, this issue only happens because for the 'pg_ctl restart' process,
> its the child postmaster process outlives the pg_ctl process, WHILE
> still having the stdout/stderr file descriptor open. As long as all the
> subprocesses exit, the file descriptor will be closed by the OS
> automatically. So, this problem really only occurs in these daemon
> spawning situations. Which for us in practice is only for pg_ctl
> start/restart.
>



I've had a brief look at these. They generally look OK - I didn't see 
any obvious issues.


cheers


andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com

- my $result = run_log($cmd);
- ok($result, $test_name);
+ my ($stdout, $stderr);
+ my $result = IPC::Run::run $cmd, '>' => \$stdout, '2>' => \$stderr;
+ ok($result, $test_name) or do
+ {
+ diag("---------- command failed ----------");
+ _diag_command_output($cmd, $stdout, $stderr);
+ };

Previously this printed a line about running the command even on
success, and the new version removes that. Was this intentional? It's
not mentioned in the commit message and that seems like a useful
feature to me.

For example:

Before:
grep "# Run" build/testrun/initdb/001_initdb/log/regress_log_001_initdb | wc -l
35

After:
grep "# Run" build/testrun/initdb/001_initdb/log/regress_log_001_initdb | wc -l
11

On Wed Mar 4, 2026 at 11:56 PM CET, Zsolt Parragi wrote:
> Previously this printed a line about running the command even on
> success, and the new version removes that. Was this intentional? It's
> not mentioned in the commit message and that seems like a useful
> feature to me.

Fixed this now. I also fixed two off-by-one errors in counting the
number of outputted lines (found by local AI review).

And I also changed the logic to write to the diff file to close and
re-open, because I realized that on Windows the new logic would create
these errors when trying to write to the diffs file:

[09:00:49.251] stderr:
[09:00:49.251] The process cannot access the file because it is being used by another process.
[09:00:49.251] The process cannot access the file because it is being used by another process.
[09:00:49.251] The process cannot access the file because it is being used by another process.
[09:00:49.251] The process cannot access the file because it is being used by another process.
[09:00:49.251] # 4 of 239 tests failed.
[09:00:49.251] # The differences that caused some tests to fail can be viewed in the file
"C:/cirrus/build/testrun/regress/regress/regression.diffs".
[09:00:49.251] # A copy of the test summary that you see above is saved in the file
"C:/cirrus/build/testrun/regress/regress/regression.out".
[09:00:49.251]

Finally I added a new commit which starts to use command_ok in
002_pg_upgrade.pl and 027_stream_regress.pl to get the new nicer output.

On Thu Mar 19, 2026 at 8:44 PM CET, Corey Huinker wrote:
>>
>>
>> Finally I added a new commit which starts to use command_ok in
>> 002_pg_upgrade.pl and 027_stream_regress.pl to get the new nicer output.
>>
>
> I tried this out, and am very pleased with the results.

Thanks for the review.

> nitpick: comment sentence needs a closing '.'

Fixed

> v4-0005: the commit message is probably too bulky to be used as-is, but the
> committer was going to change that anyway and right now too much
> information is way better than not enough.

Shortened it using your example run.

> I can't wait to use this.

Me too :) I've also added the Author/Reviewed-By/Discussion footers to
the commits to make the committers job easier.

Shortened it using your example run.

Glad I was able to help.

 

> I can't wait to use this.

Me too :) I've also added the Author/Reviewed-By/Discussion footers to
the commits to make the committers job easier.

I've held off on doing that in my proposed commits so as not to be presumptuous, but I can see where having it available would be a convenience for the committer. This will be a good test of that.

Applies clean to master, passes tests. Ship it.

Committed with minor tidy up. The main change was to add a @CARP_NOT setting in Utils.pm, so that croak() would look back past Cluster.pm to the TAP script caller.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com

On Thu, 2 Apr 2026 at 14:26, Andrew Dunstan <andrew@dunslane.net> wrote:
> Committed with minor tidy up. The main change was to add a @CARP_NOT setting in Utils.pm, so that croak() would look
backpast Cluster.pm to the TAP script caller.
 

Thanks!