Crash test update

[Jean-Tiare LE BIGOT]

OK, this is the same pattern as the case handled in:

     git show 4bcf1872 oox/source/drawingml/chart/typegroupconverter.cxx

There might similar bugs hanging around. In a nutshell, some fallback 
code in the importer forgot to init some part of the structure. This was 
fine until we moved from relying on the element count to relying on the 
last element index.

Here is a tentative fix: https://gerrit.libreoffice.org/30838

Le 11/11/2016 à 17:20, Caolán McNamara a écrit :
> On Thu, 2016-11-10 at 21:38 +0100, Jean-Tiare LE BIGOT wrote:
>> Thanks for the explanations ! Unfortunately, I can't access
>> https://bugzilla.novell.com/show_bug.cgi?id=403458.
>>
>> Can someone grant access to 'yadutaf' / admin at jtlebi.fr or send me
>> privately a copy of this file to troubleshoot ?
>>
>> I tried to reproduce the crashes with some of the files I have access
>> to:
>>
>> - fdo58197-1.pptx
>
> fdo58197-1.pptx will do fine to reproduce it, they're all the same
> issue.
>
>> They all seem to crash somewhere in libxmlreaderlo.so. It seems
>> unlikely to be related to commit
>> 4bcf1872bbe9db1388769485a7e4c0cbcce3d53c
>
> It crashes in libxmlreaderlo.so because an unexpected std::length_error
> exception is thrown and it dies through std::terminate. My commit
> "fixes" this by making those methods allow std::exception derived
> exceptions to be thrown through them, so it doesn't crash.
>
> if you...
>
> gdb --args ./instdir/program/soffice.bin --headless --convert-to pdf
> fdo58197-1.pptx
> (gdb) catch throw std::length_error
> (gdb) run
> (gdb) up... a bunch of times until you get to chartconverter.cxx
>
> you'll see that chartconverter.cxx:136 is
> Matrix< Any > aMatrix( rDataSeq.mnPointCount, 1 )
> and rDataSeq.mnPointCount is -1, which doesn't make sense as a length
> argument to vector::resize
>
> C.
>

-- 
Jean-Tiare LE BIGOT