Short summary about selecting a glx implementation
==================================================
To switch between the current and legacy NVIDIA drivers, use
update-glx --config nvidia
To switch between the GLX implementation from MESA/glvnd and NVIDIA
use the command
update-glx --config glx
This will also trigger an update of the initrd with a possibly changed
kernel module blacklist.
Detailed description of the diversions/alternatives for libGL.so.1 etc.
=======================================================================
NVIDIA provides accelerated OpenGL libraries libGL.so.* which are shipped
in the libgl1-nvidia-glx (or a corresponding libgl1-nvidia-legacy*-glx)
package. Furthermore NVIDIA maintains several legacy driver releases that
support older GPUs and ship files with the same names (but of course they
may not be mixed between different versions).
As these libraries and modules "replace" system libraries (libGL.so* from
MESA) and Xorg modules, care has to be taken to allow clean upgrades of
both the drivers and the system libraries.
Several additional requests have risen over time, e.g. from live systems:
* allow parallel installation of several drivers (even from different
vendors)
* allow driver selection at runtime
* allow switching back to the free Xorg/MESA implementation without having
to uninstall the accelerated driver and libraries
The vendor replacement libraries are only meant to be used as a complete
set as partial usage usually lead to hard to detect errors. So we only
support switching to a vendor implementation completely or using it not at
all.
Therefore we have chosen the following approach using diversions and
alternatives.
The system libraries are being diverted (to /usr/lib/mesa-diverted), so they
are moved out of the way of possible replacements and further updates to
them will happen at the diversion location. This is implemented in the
package glx-diversions which is a general solution to be used by all vendor
implementations.
All vendor implementations install their files in a private library directory
so there are no file conflicts happening. Then they register an alternative
'glx' which allows one to select the implementation. In case it is enabled,
all libraries (and other files) belonging to the implementation will be enabled
via slave alternatives, i.e. appropriate symlinks will be put into the system
locations. Triggers are being used to detect the installation or removal of
any file belonging to the slave set of an alternative and to update the
alternative accordingly.
User configuration can be done with the command
update-glx --config glx
which allows one to select a glx provider from mesa, nvidia.
(Available choices depend on the installed drivers.)
For NVIDIA, there is a second alternative that allows one to switch between
the current driver version and eventually installed legacy ones:
update-glx --config nvidia
libGL.so
========
The libGL.so link is managed by a dpkg trigger as an alternative, too.
But there are no alternative solutions available besides the diverted link
from the libgl1-mesa-dev package (if this package is installed), so this
cannot be reconfigured. The intention behind this is to always link an
application at compile time to the MESA implementation of libGL.so.1 in
order to produce portable binaries, but to use the accelerated libGL.so.1
when the application is being executed.
-- Andreas Beckmann <debian@abeckmann.de> Wed, 22 Jun 2011 12:44:11 +0200
