Re: [PATCH feature/new-vcl-scheduler] noop commit: trigger test build for branch feature/new-vcl-s...

Am 13. Juli 2017 21:08:29 MESZ schrieb Miklos Vajna <vmiklos@collabora.co.uk>:
> Hi Jan-Marek,
>
> Jenkins just failed in a vcl/scheduler assertion for me:
>
> https://ci.libreoffice.org/job/lo_gerrit/14865/Config=linux_clang_dbgutil_64/console
>
> It does not seem to be clang, I can reproduce the same with a local gcc
> build.
>
> I guess the relevant line is:
>
> cppunittester:
> /home/tdf/lode/jenkins/workspace/lo_gerrit/Config/linux_clang_dbgutil_64/vcl/source/app/svapp.cxx:540:
> static void Scheduler::ProcessEventsToIdle(): Assertion `!bAnyIdle'
> failed.
>
> Do you have an idea what is the fix for this?

Nothing from my heart. The scheduler was asked to process all pending events, but for whatever 
reason there is still an idle event left (Idle).

My guess is that I broke something with my revert and my quick fix for the missing glib function. I 
just tested the VCL tests before pushing the fix and forgot about the more complex desktop tests.

Interestingly the tinderboxes seem to be fine - very strange. Since you can reproduce the problem, 
I would be interested in a SAL_LOG with vcl.scheduler, to see what is actually happening.

Does the test pass, if you use the gen or kde4 backend? Since my attempted fix was for Gtk, this 
would prove my suspicion about the origin of the unit test failure.

I won't be able to look into this before tomorrow.

Jan-Marek

> Thanks,
>
> Miklos