Xserver driver merging pros & cons

[Michel Dänzer]

On Don, 2011-09-15 at 13:46 -0500, Jeremy Huddleston wrote: 
> On Sep 15, 2011, at 11:02 AM, Michel Dänzer wrote:
> 
> > 3) Out of tree drivers will become second class citizens.
> 
> I don't see that as a con. I see that as a benefit. If something is
> not in the tree, it IS a second class citizen, and users should not
> expect it to work any more. If the trident driver breaks, they can
> always use vesa.

That assumes every driver can reasonably live in the xserver tree.


> > This doesn't apply to proprietary drivers only but also e.g. to the Gallium xorg
> > state tracker, which we may want to use for Radeons at some point (and
> > some nouveau people have been playing with it as well, but I don't know
> > what their plans are for it).
> 
> As someone without much experience with the Gallium xorg state
> tracker, I'm curious what technical hurdles prevent you from using it
> as a library linked against by the x11 driver.

That might be possible, but I'm not sure I see the point in doing all
the extra work converting the nouveau and radeon drivers to use such a
new library, when the xorg state tracker is working now and could
probably be taken to the same level of usability with less effort.

Also, this seems like basically saying "we don't want to maintain this
existing ABI anymore, but surely there's no problem for you defining and
maintaining that new ABI, right?". SEP, anyone?


> > Speaking as a radeon driver developer, merging the driver into the
> > server tree would be unworkable at this point because since the "new
> > development model" has been in effect, it's not possible to get even
> > trivial changes into the server tree without a ridiculous amount of
> > time/effort.
> 
> Can you be more specific?  When we were discussing this yesterday, it
> seemed like the "new development model" was working and that it was no
> longer a barrier to this problem.

Try getting a review for a non-sexy change, e.g. in EXA. And even in
other areas, if it takes more time/effort to get in a bugfix than to fix
the bug in the first place, why should one bother? I'm only bothering
anymore for bugs that affect users really badly.

Generally there's been too many useful patches languishing indefinitely
due to the unrealistic process. Alex said you had slides at XDC showing
that the number of changes went down for each recent xserver release.
That matches my impression that the new process has been killing
momentum, instead of attracting new momentum as had been promised.


-- 
Earthling Michel Dänzer           |                   http://www.amd.com
Libre software enthusiast         |          Debian, X and DRI developer