Re: Cluster.pm psql() undefined $$stderr

From: Andrew Dunstan <andrew(at)dunslane(dot)net>
To: Oleg Tselebrovskiy <o(dot)tselebrovskiy(at)postgrespro(dot)ru>, pgsql-hackers(at)postgresql(dot)org
Subject: Re: Cluster.pm psql() undefined $$stderr
Date: 2025-06-13 14:53:50
Message-ID: 3a142f7b-2da6-4171-9455-f95ca2f09bd7@dunslane.net
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers


On 2025-06-04 We 10:01 AM, Oleg Tselebrovskiy wrote:
> Greetings, everyone!
>
> If you call node->psql in not-array context, with on_error_die => 1,
> but without passing stderr, you will get the following error
> and the test will die, but not the way we expect:
>
>     Use of uninitialized value in concatenation (.) or string at
>     /path/to/source_code/src/test/perl/PostgreSQL/Test/Cluster.pm line
> 2258
>
> This is because $$stderr is not defined in this case
> and warnings became FATAL some time ago.
>
> The code string in question for clarity:
>     ...
>     die
>         "error running SQL: '$$stderr'\n..."
>         if $ret == 3;
>
> Minimal reproduction is:
>
>     $node->psql('postgres', q{SELEC 1}, on_error_die => 1);
>
> This can be reproduced at current master (30c15987)
>
> Undefined $$stderr also should break dying on recieving a signal, here:
>
>     # We always die on signal.
>     if (defined $ret)
>     {
>         ... die (".... $$stderr ...");
>
> One of the ways to fix this is to initialize $$stderr with some value
> to avoid Perl error (Use of uninitialized value) and replace it with
> existing error: "error running SQL: '$$stderr'\n ..."
>
> With this approach we don't lose any useful error messages in
> regress_log_*
>
> The proposed patch is attached (0001)

I think we need to do something slightly earlier than that. Attached is
what I propose.

>
> ---------------------------------------------------------------------------
>
>
> Another question I've stumbled upon when trying to fix the aforementioned
> issue is the following: Does redirecting IPC::Run::run streams work with
> Postgres Perl module SimpleTee? I've tried to use it to "tee" STDERR to
> both $$stderr and to test's regress_log_* file. I'm not sure I'm doing
> this right, but I have not found a way to use SimpleTee with
> IPC::Run::run
>
> Is there a way to use them together?
>
> I've tried something like (0002)
>
>

I think the short answer is no, we've already hijacked STDOUT/STDERR in
Utils.pm to point to the log file, and you shouldn't mess with them. I
don't think you're using SimpleTee correctly anyway, but it's really not
meant for general use, only for the use in Utils.pm.

cheers

andrew

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

Attachment Content-Type Size
fix-cluster-stderr.patch text/x-patch 495 bytes

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Rahila Syed 2025-06-13 15:01:22 Re: add function for creating/attaching hash table in DSM registry
Previous Message John Naylor 2025-06-13 14:12:03 Re: Vacuum ERRORs out considering freezing dead tuples from before OldestXmin