Comment 22 for bug 371296

I have no really good solution for this but let me at least give you some more information, maybe you can use that to debug the issue further.

I think, when compiz is running all windows are rendered as 3D textures. In additional, I think compiz also flips itself into INDIRECT rendering mode (which means 3D accel is used but all drawing commands are sent using AIGLX from the app to the X server and then the X server is the process that calls the kernel 3D API). Compiz needs the openGL texture_from_pixmap to be available.

In general, when DIRECT rendering is used (you can tell by running "glxinfo" and grepping the out for "direct"), then you can force a 3D app to use software rendering by making sure that the LIBGL_ALWAYS_SOFTWARE=1 environment variable is set for that process. For example:

    LIBGL_ALWAYS_SOFTWARE=1 glxgears

...will usually give you much lower FPS etc.

You can also control whether DIRECT/INDIRECT rendering should be used with the env var:

    LIBGL_ALWAYS_INDIRECT=1 ./some_app

...however note that if you do this, then the 3D rendering will be done in the X server so if you want to combine that with SOFTWARE rendering then that LIBGL_ALWAYS_SOFTWARE=1 needs to be set inside the X server afaik.

Finally, there is also:

    LIBGL_DEBUG=verbose glxgears

...which allows you to analyze 3D operations in detail.

---

I don't think these environment variables will help you much for OpenOffice though. Basically OO should just work, I hope we can get the bug fixed instead.

Oh and one more thing. The driver option "NoAccel" is like a sledge hammer that turns of _everything_. You might be able to keep some of the 3D accelerated operations and still don't get crashes. Understand which part of the 3D acceleration that is causing the bug might prove useful in narrowing down the bug location.

To test with partial 3D accel enabled for the radeon driver, see the manual page "man radeon". I would in particular recommend that you try using the radeon driver without noaccel but _with_ exactly one of these options at a time:

 Option "EXANoComposite"
 Option "EXANoDownloadFromScreen"
 Option "EXANoUploadToScreen"

If you can repro the bug with DownloadFromScreen then we know that the bug is likely to be hiding in that part of the radeon driver. Also trying different, more recent versions of the driver is useful. For example, if you boot the karmic alpha live CD and the bug is not present, then it's likely that it got fixed somewhere between these versions and then we might be able to search the source diff between these versions for relevant clues.