Re: Libreoffice service fails in OBS (openSUSE)

2017-01-18 9:11 GMT+01:00 Stephan Bergmann <sbergman@redhat.com>:

> On 01/16/2017 01:55 PM, Tomáš Chvátal wrote:
>
> > 2017-01-16 11:55 GMT+01:00 Miklos Vajna <vmiklos@collabora.co.uk
> > <mailto:vmiklos@collabora.co.uk>>:
> >     On Fri, Jan 06, 2017 at 10:56:04AM +0100, Tomáš Chvátal
> >     <tomas.chvatal@gmail.com <mailto:tomas.chvatal@gmail.com>> wrote:
> >     > Ah now i get it. Well it happens always on the same test:
> >     >
> >     > [  649s] trying to instantiate implementation
> >     > "com.sun.star.wizards.agenda.CallWizard"
> >     > [  649s] unknown:0:(anonymous namespace)::Test::test
> >     > [  649s] uncaught exception of type std::bad_alloc
> >     > [  649s] - std::bad_alloc
> >
> >     Here is how I would continue debugging this. Add this after the
> >     SAL_DEBUG() that prints the service name:
> >
> >     if (i.first == "com.sun.star.wizards.agenda.CallWizard")
> >         SAL_DEBUG("foo");
> >
> >     Then in gdb you can put a breakpoint on the SAL_DEBUG("foo"); line,
> >     before the test attempts to instantiate
> >     com.sun.star.wizards.agenda.CallWizard. When you hit the breakpoint,
> > do
> >     "catch throw" and "continue". Hopefully that'll show us where
> >     std::bad_alloc is thrown.
> >
> > Hi, got to it and attaching the trace of the point. Seems like if I just
> > break there and then catch the bad_alloc it stops on the right place
> > finaly :)
>
> Interestingly, <http://ci.libreoffice.org/job/lo_callgrind_linux/2588/>
> started to fail in exactly the same way now, after (among other things) <
> https://cgit.freedesktop.org/libreoffice/core/commit/?id=74
> 3f9fc86f3d3b6e87bf58c0654bcdccab0ab383> "python3: upgrade to release
> 3.5.3".
>
> I assume you are (implicitly) using --enable-python=system in your failing
> build?  If yes, is that a 3.5.3, too?  Maybe that's the direction to search
> in.  Or did you already track the problem down?
>
> Yep we are of course using the system python, but I actually even
backtracked the problem to the issue that python uno failed to load due to
SUSE patch, which I removed now and it seems happily passing over. I still
didn't produce workable binaries so I am waiting on that to say if I am
safely out of water.

Cheers

Tom