openQA Project (public)

• https://github.com/os-autoinst/openQA/pull/3414 for an extracted method (merged)
• https://github.com/os-autoinst/openQA/pull/3421 t: Remove obsolete test job database entry in all initializations
• https://github.com/os-autoinst/openQA/pull/3416 t: Skip fixtures by default to prevent excessive runtime

okurz wrote:

• https://github.com/os-autoinst/openQA/pull/3414 for an extracted method (merged)
• https://github.com/os-autoinst/openQA/pull/3421 t: Remove obsolete test job database entry in all initializations
• https://github.com/os-autoinst/openQA/pull/3416 t: Skip fixtures by default to prevent excessive runtime

We are down on the use of fixtures in most tests now, and the redundant initialization of the db in Selenium tests is also gone.

That last PR is still an open draft. Are you planning to pursue this further?
Arguably it's a cosmetic fix since it's not removing fixtures but merely adjusting the defaults. So leaving it as-is would make sense to me as well.

cdywan wrote:

That last PR is still an open draft. Are you planning to pursue this further?

yeah, you know how it is with "Low" tickets :) Eventually I would still like to have that in.

https://github.com/os-autoinst/openQA/pull/3421 is merged meanwhile. I did work on https://github.com/os-autoinst/openQA/pull/3416 but then was annoyed by circleCI failing often (handled in other ticket). I looked into this the past days and continued with tiny steps.

Discussing this in the weekly we've identified 3 types of issues:

• Interdependencies of fixtures
• Fixtures from perl hashes are painfully slow - consider straight SQL (suggested by @kraih)
• Subroutine JSON::PP::Boolean::(++ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48. unclear what's causing this

After we discussed the topic today I managed to fix one little issue so added commit https://github.com/os-autoinst/openQA/pull/3416/commits/f2c0ba668d442bf91e2fef4e7eab4f65336e8bf8 onto the PR. Based on test results and the fresh knowledge within the team now I am feeling motivated to bring other issues up to the team more quickly :)

cdywan wrote:

• Subroutine JSON::PP::Boolean::(++ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48. unclear what's causing this

I debugged a little bit.

Mentioned already in https://progress.opensuse.org/issues/38264, also related https://progress.opensuse.org/issues/29048

https://github.com/rurban/Cpanel-JSON-XS/issues/65 (closed but unfixed)

perl -wE 'use Cpanel::JSON::XS (); use JSON::PP ()'
Subroutine JSON::PP::Boolean::(++ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.
Subroutine JSON::PP::Boolean::(0+ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.
Subroutine JSON::PP::Boolean::(-- redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.

Happens only with Devel::Cover and JSON format.

Same with JSON::MaybeXS, which Devel::Cover uses

perl -wE 'use JSON::MaybeXS (); use JSON::PP ()'
Subroutine JSON::PP::Boolean::(0+ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.
Subroutine JSON::PP::Boolean::(-- redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.
Subroutine JSON::PP::Boolean::(++ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.

Can be avoided by loading JSON::PP before everything else:

export PERL5OPT="-MJSON::PP -MDevel::Cover...

Unclear why we only see it in the fullstack tests. Maybe JSON::PP is indirectly used in openQA somewhere before Cpanel::JSON::XS, while in os-autoinst code Cpanel::JSON::XS is used first.
But that would also explain why -MJSON::PP "fixes" it, but a use JSON::PP at the top of the t/full-stack.t does not (because the problem lies in os-autoinst code).

Oh, and just for the record, if you're wondering where that warning comes from, a tiny reproducer:

perl -wE'
package Foo;
use overload "+" => sub { say 23 };
use overload "+" => sub { say 23 };
'
Subroutine Foo::(+ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.

ok, cool. Thanks for the context and the reminder about the other tickets where we have seen this. So in case you think there is anything related to JSON::PP we can do to improve or prevent further future confusing issues then please open a ticket or create a PR or something

tinita wrote:

Same with JSON::MaybeXS, which Devel::Cover uses

perl -wE 'use JSON::MaybeXS (); use JSON::PP ()'
Subroutine JSON::PP::Boolean::(0+ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.
Subroutine JSON::PP::Boolean::(-- redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.
Subroutine JSON::PP::Boolean::(++ redefined at /usr/lib/perl5/5.26.1/overload.pm line 48.

Can be avoided by loading JSON::PP before everything else:

export PERL5OPT="-MJSON::PP -MDevel::Cover...

Unclear why we only see it in the fullstack tests. Maybe JSON::PP is indirectly used in openQA somewhere before Cpanel::JSON::XS, while in os-autoinst code Cpanel::JSON::XS is used first.
But that would also explain why -MJSON::PP "fixes" it, but a use JSON::PP at the top of the t/full-stack.t does not (because the problem lies in os-autoinst code).

What I read there is something uses both JSON::MaybeXS and JSON::PP but if JSON::PP comes first, it's fine.

I would suggest a new ticket about this rather than this one which is about test runtime.

I created https://progress.opensuse.org/issues/90371
I only mentioned it here because the warning was mentioned.