Re: [Libreoffice-qa] minutes of ESC call ...

On Mon, Sep 18, 2017 at 6:10 PM, Michael Meeks <michael.meeks@collabora.com>
wrote:

> Hi Markus,
>
> On 18/09/17 16:26, Markus Mohrhard wrote:
> >     * Crash Reporter (Xisco)
> >         + http://crashreport.libreoffice.org/stats/?days=128
> ...
> > I already spent more than a day looking into that issue. It is not a
> > problem in LibreOffice or our tooling. Merged libraries generarted with
> > MSVC 2015 seem to store pdb info differently and DIA (the MS API that
> > breakpad uses to access pdb info) is not able to extract any useful
> > information. The uploaded symbol information don't contain anything
> > useful so that breakpad is not able to generate useful backtraces for
> > the merged library.
>
>         Oh - interesting. So - did I understand right from the above that
> the
> problem is in the client-side up-load of the stack / symbol information ?

Yes. The problem is already during the collection of the symbol information
after the build as part of make symbols. [1] and [2] show two mergedlo.dll
symbol files that show that a significant part is already missing. As a
result there is no way to recover the stacktrace after a crash happens.


>         I'm curious what Mozilla are doing here since they default to
> VS2015 -
> and also have some "put everything in one big DLL" approach.

I have not looked at that. I checked if a newer breakpad version might help
but there was no solution. Additionally I had a test if the incremental
linking flag that we use would be the problem but that was also not
helping. The next step would be to debug the dump_syms executable from
breakpad and try to understand what is going wrong.

Regards,
Markus