Bug#1057620: doomsday: segfault in _XFlush() when Qt is using native Wayland

[Adrian Bunk]

On Mon, May 06, 2024 at 10:28:40AM +0100, Simon McVittie wrote:
> Control: retitle 1057620 doomsday: segfault in _XFlush() when Qt is using native Wayland
> 
> On Mon, 06 May 2024 at 00:31:31 +0200, Bernhard Übelacker wrote:
> > Bug 1062969 / Bug 1065714 mentions a workaround
> > to be able to run doomsday with wayland:
> > 
> > SDL_VIDEODRIVER=x11 QT_QPA_PLATFORM=xcb doomsday
> 
> Bug 1062969 is about a problem with SDL_JoystickName(), which is
> fixed in unstable. The clone #1065714 is also about doomsday's use
> of SDL_JoystickName(). Fixing those is unrelated to that workaround:
> the only connection is that I had to use that workaround to avoid the
> crash-on-startup before I was able to reproduce the joystick bug on
> my system.
> 
> Bug 1057620 and its duplicate 1065709 are about the crash on startup that
> can be worked around with "SDL_VIDEODRIVER=x11 QT_QPA_PLATFORM=xcb".
> I'm copying the title from 1065709 to 1057620 to make its scope a bit
> clearer.
> 
> > It looks like upstream removed the relevant code and relies just
> > on SDL functions, but unfortunately did not release a new version yet.
> > 
> > https://github.com/skyjake/Doomsday-Engine/commit/5cc4995861
> 
> I haven't reviewed that change in detail, but letting SDL handle all of the
> functionality within its scope is generally a good direction to go in.
> 
> The root cause of this bug is that doomsday uses both SDL and Qt, each
> of which makes its own independent choice between an X11 backend or a
> native Wayland backend, and then uses X11 directly itself and assumes
> that both SDL and Qt have also chosen to use X11.
> 
> If Doomsday needs to use both SDL and Qt for graphics/windowing, probably
> the right way to implement this would be to let one of those libraries
> choose its backend (X11 or Wayland) according to its usual heuristic,
> and then call configuration functions that force the other library to
> make the same choice.

Is this the same bug as #1113738?

If yes, does the workaround from Bernhard I NMUed there look correct to 
you?

>     smcv

cu
Adrian