I'm running the nvidia-beta drivers 1.0.9626 and everything works fine, everything but a small bug concerning the compilation of applications.
Now compiling "real" opengl programs is not a problem, as ppracer or celestia compile well. However when trying kipi, gwenview or digikam I get the same error message, which looks like this:

Now I tried it with different versions of the driver, but it remains also with 1.0.8774 and 1.0.8776, which clears the beta-cause.
The library is located in /usr/lib/, and there's a symlink in /usr/X11R6/lib/.
Also the common bug in libGL.la with "for use with libtool" is solved.

Now I've already tried a lot and searched the internet from bottom to top. I can't find a solution, or nothings works. It's definitely related to the nvidia-driver, cause with normal X files everything compiles fine (except the point, that I have no libGL.la installed with x.org)

I would appreciate any kind of help!

it has been my experience and i might be wrong for your version that the files needed to link against the nvidia libs are not installed by the nvidia installer.
seems like i had to extract from the .run with a -x command then compile then put headers and .la in place by hand and possibley edit .la with some better path info.
.la file is just a text file with some info and paths in it don't be afraid to edit.

i'm kind of thinking if libGL.la is in /usr/lib then the script won't need or try to use __LIBGL_PATH__ but i might be quite wrong about that one.

I personally prefer installing by hand, that's what my first thought was about the problem, that I missed something to change, what the nvidia-installer does. But when I tried the installer, nothing changed. Now nowhere in the libGL.la-file itself is __LIBGL_PATH__ written, but it seems that there's still some file around refering to that place!
By the way, what do you mean by compiling, the only thing to be compiled is the kernel module and I don't think that's the problem.

Thanks anyway

the nvidia installer builds and installs the library and creates the .la file
do --extract-only on the run thingy and then do a make and look around inside.
important to see the error as libtool: line 2931 so that is where the variable is.
like i said before it must just be a last ditch effort by the script because it didn't find what it needed.

also very problematic to try to link against nvidia libs.

try running ./autogen.sh before configure

very complex issue -- i wish i could think of all the posibilities.
once i tried to bbuild all of Qt and KDE against nvidia libs and ended up with such a mess i had to uninstall nvidia and start over.

first when you build the kde apps you use the version from within the larger package ? yes

otherwise if you are using out of cvs you might need to do
make -f admin/Makefile.common cvs
before configure

Thanks for your advice, I already solved the problem!
It was somehow "basic".
The problem was, that I installed the nvidia-drivers by hand, and discovered just lately, that the libGL.la file contains parts, which need to be changed.
Unfortunately I had compiled libkipi before I recognized that problem and that __LIBGL_PATH__ thing was hardcoded into that libs. And all three apps, which failed to compile needed those libs.
So all I had to do to get it working, was recompile libkipi with the changed libGL.la file and all worked out fine!
But thanks for your advice!