[Xcb] mesa +xcb = assertion death

[Jamey Sharp]

On Wed, 2005-04-27 at 15:27 -0400, Jeremy Kolb wrote:
> If this is any help I threw in a printf and c->out.request_written and 
> c->in.request_read went like:
> 
> 2 - 2
> .....
> .....
> 10 - 10
> 11 - 11
> 12 - 12
> 12 - 65536
> 
> 
> Pretty dramatic jump.  Dunno if this is useful or not, I really don't 
> know how to debug this since I don't know xlib/xcl internals.

I haven't had a chance to investigate yet, but please feel free to keep
poking at it: you're doing great. :-) I can't keep much more of the
relevant architectures in my head than what you've discovered already.
(You've discovered my secret "printf stuff from c->out" debugging
technique: I have nothing more to teach you.)

This at least doesn't look like a regression of the major bug we fixed a
couple of weeks ago, as that would only happen after at least 65530
requests had been sent. It could be a bug in Mesa: does Mesa correctly
call LockDisplay and UnlockDisplay everywhere that it's supposed to?
(The Xlib/XCB patches hook into those functions to get the hard work
done.)

For that matter, is the XTHREADS C pre-processor symbol being defined
anywhere in Mesa's build headers? I patched the modular Xlib source a
little while back to declare this in Xlibint.h if needed, and I suppose
one possibility is that you haven't re-run autoconf and the configure
script in the X11 directory since then. I can't decide if that would
explain your bug, but it seems worth trying.

--Jamey