[Mesa3d-users] LIBGL_ALWAYS_INDIRECT ignored / glxgears: Error: couldn't get an RGB, Double-buffered visual. [solved or circumvented?]

[Reuti <re...@st...>]

Hi all,

I searched the archive for something related to the subject of this  
post. But although I found other users also having this problem on  
the Internet, there was nothing related in the archive here. Maybe  
this is even the wrong list for this issue, then maybe someone can  
please point me to the correct one for further investigation.


*** Setup:

I have a powerful server running Linux with several cores (but  
without any sophisticated graphics card, just a minimum **) ), to  
which users connect from the offices. On their desks they use a  
Linux, Mac OS X and Linux.


*** Behavior of failing programs:

Starting programs via SSH's X11 forwarding which use OpenGL I get  
errors like:

- glxgears: Error: couldn't get an RGB, Double-buffered visual.
- glxinfo: Error: couldn't find RGB GLX visual
- others startup up display only a black screen w/o any content

Even setting LIBGL_ALWAYS_INDIRECT or LIBGL_ALWAYS_SOFWARE doen't  
change anything.


*** Behavior of working programs:

Some applications like GaussView 4 from Gaussian or Maestro from  
Schrodinger have a switch -soft, respecitvely -sgi, which enables  
them to work in this configuration. Unfortunately not all software we  
want to use have such an option.


*** Obersavtion between workstation:

When I try to "emulate" the server by a workstation which has an  
Nvidia card installed, and connect to this machine from another  
workstation, all is working fine. I can see of course, that in this  
case other libraries are used. On the server:

$  ldd /usr/bin/glxgears
...
         libGL.so.1 => /usr/lib64/libGL.so.1 (0x00007f3f8273d000)
...
         libXxf86vm.so.1 => /usr/lib64/libXxf86vm.so.1  
(0x00007f3f8181d000)
         libXdamage.so.1 => /usr/lib64/libXdamage.so.1  
(0x00007f3f8161a000)
         libXfixes.so.3 => /usr/lib64/libXfixes.so.3  
(0x00007f3f81414000)
         libdrm.so.2 => /usr/lib64/libdrm.so.2 (0x00007f3f8120b000)
...

But on an "emulated" server on a workstation:

$ ldd /usr/bin/glxgears
...
         libGL.so.1 => /usr/X11R6/lib64/libGL.so.1 (0x00007f952cf92000)
...
         libGLcore.so.1 => /usr/X11R6/lib64/libGLcore.so.1  
(0x00007f952b1d6000)
         libnvidia-tls.so.1 => /usr/lib64/tls/libnvidia-tls.so.1  
(0x00007f952d286000)
         libXext.so.6 => /usr/lib64/libXext.so.6 (0x00007f952afc4000)
....


*** Solution.

Now I found a hint on the Internet *), to copy the files from the  
workstation from /usr/X11R6/lib and lib64 to the server machine. I  
chose to copy them to /opt/nvidia not to mess up with the default  
installation. When I now set LD_LIBRARY_PATH all applications are  
running fine (one still listed the old libGL.so and I had to use  
LD_PRELOAD for it).


*** Question:

Why is this working? My assumption would be, that the original  
libGL.so supplied by openSUSE 11.1 and openSUSE 11.2 (ubuntu karmic  
in the link *) ) checks the local graphics card first, before it can  
even decide to honor LIBGL_ALWAYS_INDIRECT/SOFTWARE at all. Hence an  
error is given to the application and fails. When this behavior is  
the same across three dristributions with different versions of  
libGL.so (7.2 and 7.5), I would assume that this behavior is intended  
- but why? Was is falling back to software rendering silently in  
former versions of libGL.so?

And why is the libGL.so installed with Nvidia working, although there  
is no Nvidia card at all in the server? Or is the problem in the X11- 
Client libraries, which run on the server?


-- Reuti

*) http://forum.compiz.org/viewtopic.php?f=86&t=12271

**) It's a server from SUN with only the graphics card from the ILOM  
(Integrated Lights Out Management for remote administration and KVM).