On May 21, 08 16:36:37 +0200, Marvin wrote:
I got OpenGL renderer string: Mesa DRI R300 20060815 x86/MMX+/3DNow!+/SSE2 NO-TCL
So you're compiling Mesa w/o optimization? Probably good idea for debugging, I don't see how this should change behavior here.
this may be a difference - are you using x86 or x86-64 ?
Uh-oh. Yes, it could be a 64bit issue, wouldn't be the first. I'll install that on this machine when I get to work on Friday. Luckily I've got some spare partitions.
I use x86-64 with: make -j2 linux-dri-x86-64 DRI_DRIVER_INSTALL_DIR=/usr/lib64/dri DRI_DIRS="r300"
I wonder why this effectively disabled some of the optimizations...
the funny thing is, that the glxinfo of Factory reports the error in the second call, but the self compiled version from the mesa branch does not. Sadly, the self compiled version of glxgears still aborts.
Even more weird :-/
I compared the two runs using ltrace -Sf, but couldn't see any difference (except the failing glxChooseVisual). What could be the reason for this behavior? I don't know much about X, but I'm able to use a debugger...
I'll try to reproduce on 64bit first.
If I fail, I'll guide you through.
Using gdb on the Xserver is still really nontrivial. E.g. here I cannot
start X in the debugger (it hangs during keyboard initialization), but
have to attach it to a running process. This is bad if you want to add
watchpoints on variables initialized during startup...
Matthias
--
Matthias Hopf