fre 2005-01-28 klockan 09:54 -0700 skrev Ivan Gyurdiev:
On Fri, 2005-01-28 at 17:35 +0100, Peter Backlund wrote:
fre 2005-01-28 klockan 09:17 -0700 skrev Ivan Gyurdiev:
Quite a few of us would like to see software we do personally use pushed to Extras.
Hi, I haven't been following this discussion, but I'd like to say that I don't want to see software I use pushed to extras due to the fact that Extras is not sync-ed to Rawhide. I am a rawhide beta tester, and extras is a pain for me to use. Do you know what kind of hackery is required to get the Nvidia driver from livna properly installed on a rawhide system? That's because Livna, like extras, is not sync-ed to rawhide.
Would you mind sharing exactly what you had to do? Or perhaps open a bug on bugzilla.livna.org.
Well,
If you don't install the livna packages xorg-x11-Mesa-libGL breaks your system at every upgrade, and you can't get rid of it (because it's required).
However you can't install the nvidia-glx package, because it depends on the kernel module, which needs to be recompiled per kernel, and Livna does not sync to rawhide.
You can rebuild the src.rpm like this:
rpmbuild -bc --target i686 --without driverp nvidia-glx.spec
to simply build a new kernel-module-nvidia package when a new kernel is released.
Furthermore the driver distributed is not patched for well-known bugs with the patches here: http://www.minion.de/files/1.0-6629/
What are these patches for?
So, to summarize, my best option is to use the nvidia-distributed installer, extract, patch the driver, install it, generate a fake provides rpm, install the nvidia-glx package on top of that, and disable the startup script for nvidia (because the module as installed by nvidia is not where the livna rpm puts it, so it doesn't work).
Mixing the rpms from livna and the scipt installer is not and will never be supported by livna.org.
Then, there's xorg-x11-devel which owns the libGL.so link, which happens to be a dead link, because the Mesa package is not installed. I have to move the GL devel stuff around and redirect the link to the nvidia files every time xorg-x11-devel is updated, or I can no longer build wine-cvs with openGL support.
Really to resolve this the following is needed:
- updated livna nvidia package to include more patches
If there is a good reason to use a certain patch, it will be included.
- Livna.org sync-ed against Rawhide.
Won't happen anytime soon, but see above for a very simple fix.
- updated xorg-x11 packaging to separate the Mesa GL stuff
- some sort of alternatives system or post-install scripts to find correct provider of libGL.so.1
This already works with the rpms.
That doesn't include the SElinux bug in the strict policy where udev needs to restorecon devices from /usr/etc/devices. I've filed this in bugzilla and I assume it's being resolved.
bz#?
I'm not 100% sure of how runtime linking works, but if you look at a binary like /usr/X11R6/bin/glxgears, it links to the Nvidia libraries when they are installed through the livna.org package, like this:
This is not a runtime issue - it's a compile time issue that I'm talking about. I can't compile stuff against libGL.
If you don't uninstall the Mesa libGL/libGLU rpms, you can compile against those. If you want Nvidia-specific features, use -L/usr/lib/nvidia -I/usr/include/nvidia.
/Peter