xorg HEAD + Mesa HEAD = boom

[Brian Paul]

Kristian Høgsberg wrote:
> Ian Romanick wrote:
> 
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> Kristian Høgsberg wrote:
>>
>>> Benjamin Herrenschmidt wrote:
>>>
>>>> Haven't had time to investigate much yet (and probably won't for a
>>>> couple of weeks) but with a build of today's CVS Mesa, r300 DRI, and X
>>>> against that Mesa version (the whole lot), the server blows up right
>>>> away when launching glxinfo or glxgears.
>>>>
>>>> The latest log I got (with glxgears) shows this backtrace:
>>>
>>> It isn't clear from the stack trace what went wrong, but I've just
>>> checked in a patch which fixes a few issues: GLcore needs to build with
>>> TLS settings matching glx, the ARGB visual is now marked as
>>> non-conformant, and there was a problem with freeing a mesa buffer even
>>> if it never got allocated.
>>
>>
>> What is non-conformant about those visuals?
> 
> 
> Maybe it's not non-conformance, but a driver bug instead.  The radeon 
> driver panics and calls exit() if it sees a visual with no depth buffer 
> (see the switch on line 175 in radeon_state_init.c).

Yikes!  That should be fixed.

The default case should probably be:

	rmesa->state.depth.clear = 0;
	rmesa->state.depth.scale = 0.0;  /* or 1.0? */
	rmesa->staet.stencil.clear = 0;

Can someone try that?

-Brian