SL 6.X / SLC 6.X / CentOS 6.X / RHEL 6.X

I encountered a strange problem that causes OpenGL (GLX) applications to fail on SL6X, SLC6X, CentOS6X and RHEL6X (where X = 3 for the time being).
Applications die with errors in form (I login to a remote machine, running one of these systems, using “ssh -X -Y”):

[…]$ glxinfo
name of display: localhost:11.0
Error: couldn’t find RGB GLX visual or fbconfig

[…]$ glxgears
Error: couldn’t get an RGB, Double-buffered visual

[…]$ root
(…)
Error in TGLWidget::CreateWindow: No good OpenGL visual found!
terminate called after throwing an instance of 'std::runtime_error’
what(): No good OpenGL visual found!

It seems to me that the problem resides somewhere in current “Mesa” libraries (7.11-5.el6) on these systems.
Note: the appearance of this problem heavily depends on the actual “graphics card” type of the remote machine. For example, one faces it with remote “virtual machines” which have some “virtual graphics cards”, but one does not face it if the remote machine (a “real” / “non-virtual” one) has a “real” NVIDIA graphics card installed.

For the time being, the “brutal fix” is … before you try to run any OpenGL (GLX) application …
[bash]$ export LIBGL_ALWAYS_INDIRECT=1
[tcsh]$ setenv LIBGL_ALWAYS_INDIRECT 1