On Wednesday 08 November 2006 14:55, Lubos Lunak wrote:
> On Wednesday 08 November 2006 02:15, Philip Falkner
wrote:
>
> I'll check how AIGLX works here with NVidia. I think
Compiz needs to be
> run with --strict-binding that somehow slightly changes
how it binds the
> pixmap to the texture (Beryl tries to detect AIGLX and
should turn it on
> automatically).
I can move over the strict-binding code, but I can't test
it. The failure to
make ctxbuffer current right at the start with AIGLX means
it won't be able
to do anything with the code when it gets there.
> Well, assuming it just needs to be destroyed slightly
later, the
> bound_pixmap approach is still ok. If it crashes even
with that, then I
> guess that's a rather bad bug that needs to be worked
around somehow. I'm
> afraid just leaving out the glXDestroyPixmap() call
would rather soon make
> the leak quite big.
Making pixmap's lifespan imitate bound_pixmap's (make it
part of Window,
destroy it in discardTexture(), also destroy it if it's not
None before
creating it in bindTexture()) does allow multiple menus to
open/close in
konsole, but new windows+decos, or highlighting a menu item
still crashes X.
With what I've tried so far, that's the closest I've gotten.
I'll keep
trying, though.
For what it's worth, this problem may be related to
htt
ps://bugs.freedesktop.org/show_bug.cgi?id=4696, along
with the linked Red
Hat bug report. Certainly, the ATI crash log attached to
the freedesktop bug
looks similar enough to mine.
> > Also, kwin now crashes with the saturation code.
int glversionminor =
> > glversioninfo[1].toInt() causes a QList assert,
with an out-of-range
> > index. I've not dug any deeper yet, since I only
just updated, but
> > forcing it to 2 allows me not to crash, at least.
I'm assuming that
> > glversioninfo isn't quite what's expected, but
I'll look into that
> > tomorrow.
>
> That code looks perfectly fine to me, both the old one
and the current one
> in SVN. OpenGL docs say the version is
<version><space><whatever>, where
> version is either major.minor or major.minor.release .
I don't see how it
> could break with the version reported by your glxinfo.
I realised it's only breaking in AIGLX mode: with no
context, glGetString()
returns nothing. In regular X, it works perfectly. Sorry
for the mistake.
--
Philip Falkner
_______________________________________________
Kwin mailing list
Kwin kde.org
https://ma
il.kde.org/mailman/listinfo/kwin
|