Finally with 4.0.2, kernel-2.4 DRM and XF86 DRI for Matrox are compatible
again, but if I start e.g. one of the xscreensaver 3D hacks, it still uses
software rendering. If I start it with LIBGL_DEBUG=1, I can see that it
fails to resolve the symbol XF86DRIOpenFullScreen in
/usr/X11R6/lib/modules/dri/mga_dri.so although it is present in
/usr/X11R6/lib/modules/extensions/libdri.a -- but it isn't external:
--- 8< ---
nils@cognac:~> nm /usr/X11R6/lib/modules/extensions/libdri.a | grep
00000800 t ProcXF86DRIOpenFullScreen
nils@cognac:~> nm --extern-only /usr/X11R6/lib/modules/extensions/libdri.a
| grep XF86DRIOpenFullScreen
--- >8 ---
From the other DRI modules I can see that they all reference that symbol,
so I suspect DRI is broken regardless of the video card.
It worked on r128 last I checked (4.0.1Z era).
Observing people will have noticed that the symbol in libdri.a is
ProcXF86DRIOpenFullScreen, not XF86DRIOpenFullScreen. Duh. Harald Hoyer
pointed out to me that the missing symbol somehow slipped into libGL.so
built with XFree86 (but not packaged).
I guess this should not be the case :-) as libGL is not per se the only means
to access DRI (or am I wrong here?).
for the record, still broken in 4.0.2-0.2 (Jan 3 rawhide update)
Okay, did some work on this last night. The actual problem is in Mesa (hence
changing the component) because Mesa includes a snapshot of the XFree DRI GL
stuff from XFree 4.0.1. Obviously some things have changed since then since DRI
is even still a moving target. I pulled similarly from 4.0.2 and have the
updated Mesa RPM/SRPM at http://katz.linuxpower.org/mesa/. Unfortunately, it's
still not working for me on a r128 but I'm getting the exact problem I had when
I was running DRI CVS, so I think that the Mesa build at least is good and
should enable everything to work properly. At least glxinfo seems to think so.
If someone could verify that, I'd appreciate it so I don't chase ghosts.
Yeah, the new package you put up seems to work (>500 FPS with miniature gears
window, on G400)
*** Bug 22910 has been marked as a duplicate of this bug. ***