On Fri, 2005-01-28 at 11:56 -0600, Rex Dieter wrote:
Ivan Gyurdiev wrote:
Why would you split a library off from xorg-x11, but not split the headers?
As I said previously, the MesaGL .so link and headers are required, else you end up with proprietary-ized binaries that work *only* when the nvidia GL libs are installed. MesaGL-linked binaries wor on both MesaGL and Nvidia-GL (supposedly) systems.
I say (supposedly) because I don't have/use NVidia cards myself, but I've heard enough success stories from reputable sources to trust the conclusion.
Allright. I tested this, it works, so I'll admit you're all right, I'm wrong, and from what I've learned from this discussion I even know exactly why - it has to do with symlinks, linker path precedence, and the nvidia script installer, it's not important.
Conclusions: =============
- Mesa-libGL can work alongside nvidia-glx, and should be installed if I am to build against libGL or it doesn't work at all.
- I should stay away from the nvidia installer, use only livna packages, and rebuild against rawhide.
What needs to be done: ======================
- The -devel package should not install dead links on my system - it should require Mesa-libGL.
- I still say the libGL devel stuff and libGLU should be split from xorg-x11-devel.
- livna needs to be patched for memory leak bug, and other bugs. Patches at www.minion.de/files/1.0-6629/.
- There should be a rawhide-sync'ed livna and extras, and not having them makes moving packages to extras evil.
- SElinux strict policy bug with /etc/udev/devices to be fixed...
Anything else I'm missing?