Bug#780853: wesnoth: typing chat messages reproducibly crashes the client when using Debian's server package

[Ignacio R. Morelle]

On Sun, 17 May 2015 02:08:48 +0200 Markus Koschany <apo at gambaru.de> wrote:
> On Sat, 16. May 16:44 Vincent Cheng <vcheng at debian.org> wrote:
> > Hi Markus,
> >
> > On Sat, May 16, 2015 at 3:48 PM, Markus Koschany <apo at gambaru.de> wrote:
> > > I installed wesnoth-1.12-dbg and tried to reproduce the issue while
> > > running gdb. Again the client crashed with a segmentation fault. Please
> > > find attached the gdb log file. Hopefully someone else will find some 
clues.
> >
> > Please report this bug upstream at bugs.wesnoth.org. Thanks!
> 
> Hi,
> 
> I'm quite sure that there is something Debian specific involved in here.
> As I stated in my initial bug report, official servers don't crash the
> client when I perform the same actions. This is only reproducible with
> Debian's server package.
> 
> Regards,
> 
> Markus

Hello, I'm one of the upstream Wesnoth developers.

I have investigated the issue extensively (as Wesnoth bugs #23633 and #23599) 
and found that one of the patches applied to libvorbisfile3's source package 
on Debian (in version 1.3.4-2 of the package) is the cause. Turns out this is 
already reported as bug #782831 in Debian [3].

Our upcoming 1.12.3 release will contain a somewhat reliable work-around in 
the form of a larger data/core/music/silence.ogg file (which is used to switch 
the lobby music off, hence the issue here), but I imagine the maintainers of 
Wesnoth on Debian would prefer to wait for #782831 to be fixed.

  1: https://gna.org/bugs/?23633
  2: https://gna.org/bugs/?23599
  3: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=782831

Also, I don't think the Wesnoth server package is directly involved, it's just 
that the client-side bug appears to be highly timing-dependent due to SDL and 
SDL_mixer's involvement. Keyboard or mouse input taking place between attempts 
to set the music to the short silence.ogg seemed to have an important effect 
on my test program, and so did the size of the compiled code (due to 
optimizations I reckon).

-- 
Regards
  Ignacio R. Morelle