[Xcb] mesa +xcb = assertion death

[Jeremy Kolb]

Jamey Sharp wrote:
> On Tue, 2005-04-26 at 03:09 -0400, Jeremy Kolb wrote:
> 
>>Okay, so I lied about GL on XCB at the moment since the server-side glx 
>>was being used instead of the client even though it told me otherwise 
>>and now I can't test it.
> 
> 
> I don't understand that: we certainly want GLX indirect to use XCB, and
> we want that just as much as the little bit of XCB that direct rendering
> would use. In fact, I care more about GLX indirect: I understand that X
> servers are moving toward being able to do hardware-accelerated indirect
> rendering, and that a lot of messy code could be discarded by getting
> rid of DRI.
> 

The problem was the server was running the proprietary nvidia glx, even 
though my LD_LIBRARY_PATH was pointing to mesa.  So glxinfo reported 
that the server side glx was Mesa for some odd reason even though it was 
nvidia and all the glx calls went through there.  I'm not entirely sure 
how it works.  However I fixed it by starting an xserver with it's 
default glx, and that seemed to work just fine and now I get mesa's 
indirect rendering code.

> 
>>So now just running the default mesa (freshly compiled with dri support, 
>>otherwise it will not work) and set my LD_LIBRARY_PATH to /opt/fdo/lib 
>>where I keep the Mesa/XCB//X11 --with-xcb libs and now running any GL 
>>program and I get an assertion failing in xcl/xcblock:92.
>>
>>Any ideas?
> 
> 
> That sounds exactly like the bug we fixed a couple weeks ago. You'd
> better not have up-to-date xlibs CVS, or I'm going to be really upset.
> 
> Now, the build of Mesa that you are testing with has your XCB patches,
> right? If rebuilding Xlib doesn't help you, try building Mesa from
> unpatched sources and see if you still see the assertion failure.
> 
This is vanilla Mesa (well, modified but that code isn't enabled) on 
X11.  At first when it happened I updated xlibs because I figured there 
was a fix, but no.  Same error.  It's the latest xlibs.

Jeremy