[Xcb] Progress

[Jeremy A. Kolb]

On Mon, 5 Dec 2005, Jamey Sharp wrote:

> On Mon, Dec 05, 2005 at 01:13:53PM -0500, Jeremy Kolb wrote:
> > It's been pretty quiet around here.  There's still a lot to do on the
> > todo list and I know that there are some technical issues holding back
> > some extensions from being fully implemented.  Is there any plan for
> > addressing these issues?
> 
> No, there's no plan. I'm not much of a planner. :-) So... let's make
> one?
> 
> It's probably been obvious that I've been distracted again for the last
> couple of months. I've been working a bit on making the X Test Suite
> pass when running on my version of Xlib that uses XCB, so that's where
> all of my XCB time has gone recently. I'll probably continue putting my
> time in that direction for a while longer.
> 

Sweet.

> I remember that your port of Mesa to XCB needs one of these "technical
> issues" fixed for performance reasons, but if I recall correctly, you
> can do a complete but performance-impaired port with an unmodified XCB.
> Is that true, and do you want to get back to that?
> 

Yeah, I do need to do that.  One thing we talked about on #dri-devel was 
porting the software x11 backend that draws using the X protocol to xcb.  
I haven't started that yet, most of my work has been on the indirect 
rendering side.

> Fixes for these technical issues mostly require people who can work with
> XSLT. Now, XSLT is not all that hard as long as you remember to treat it
> like a programming language, although there's an awful lot of stuff in
> the current XSLT stylesheet for XCB. So I think that anybody who can
> spare a couple of weeks to ramp up can contribute these fixes and let
> extension development move forward again.
> 

It's mostly funky lists in structures that's holding back things like 
xprint and xv.  Unfortunately my comments suck so I don't really remember.

> I seem to recall that Josh was considering replacing his XSLT bits with
> something in a more traditional language -- Python, was it? If ramping
> up on the current stuff is too daunting, somebody might try replacing it
> with something that people can improve more easily. Any takers?
> 

I know that Ian Romanick uses python to generate code for all of the GLX 
requests in Mesa which is pretty cool.  We may want to talk to him if 
someone wants to do it in python.

> At the moment I have to mostly rely on you folks to follow through on
> to-do items, though I'd love to get questions and I'll provide as much
> help as I can.
> 
> --Jamey
>