any impending news on video drivers for f12?

[Adam Williamson]

On Mon, 2009-08-17 at 19:57 -0400, Bill McGonigle wrote:
> On 08/17/2009 02:50 PM, Adam Williamson wrote:
> > Bill's post isn't wrong in any respect, but it's just a description of a
> > situation that everyone knows damn well about already. Everyone working
> > on X is aware that the entire thing is a gigantic moving target, but
> > restating the problem doesn't help anyone move forward...
> 
> Let me pose a few questions to illustrate some of my unknowns.  I'll try
> to be concise, and I recognize these are somewhat interdependent:
> 
> 1) is there a known scope to the limit of work that needs to be done to
> make xorg generally useful for desktop users?  That is, is this an
> arduous interval with a known outcome for fixing the current drivers or
> is the future purely speculative? (again, with utmost respect for those
> putting in the hard development work)

Arduous interval.

> 1a) are we hitting expected problems, illustrating unfinished work, or
> exposing poorly considered architecture/decisions?  Is this process
> proceeding as expected or is it much worse (or better)?

Expected problems. Required reading:
http://keithp.com/blogs/Sharpening_the_Intel_Driver_Focus/

at present, the X and kernel devs are rewriting huge chunks of the
drivers (and smaller chunks of the server, kernel, libdrm and various
other bits) in order to fundamentally change how some rather important
things are done: 2D _and_ 3D acceleration, memory management,
modesetting to name a few. This is obviously very disruptive, but also
necessary and will be ultimately beneficial. It's proceeding broadly as
expected, AIUI.

> 1b) do we stand a chance of catching up to the pace of hardware
> development at the current driver development rate?

Intel? Yes. Intel's rate of hardware development is actually rather slow
from the driver's point of view, because they don't change the
driver/hardware interface much these days. Basically the only changes
that have had to be added to the intel driver to add support for the
last two or three 'generations' of intel graphics hardware are simply to
add their PCI IDs to the detection code.

> 2) assuming 1 is bounded, once intel/radeon are whipped into shape, are
> we out of the woods?  More specifically, are ATI & Intel actually
> providing hardware spec or are we still having to reverse engineer?

Intel provide specs and have done for years, for all chipsets except the
GMA 500 (for which see my blog, if you can hack the dullness). I'm not
so clear on the ATI situation; they say they're providing specs, I'm not
sure how full the specs in question are. fcami might know more there, if
he's reading. NVIDIA don't provide any specs, nouveau is reverse
engineered.

> This gets to open source being able to keep up with the hardware
> revision cycle.  It used to be this was forever doomed to uncertainty,
> hardware availability to reverse engineers and HCL's.  I thought it had
> gotten better, and certainly an open future is much brighter.
> 
> 3) is xorg making the right decisions?  Software management is littered
> with the bodies of those who decided a ground-up rewrite is the only way
> to go.  Yet, sometimes it is.

I'm not sufficiently expert to answer this, but to _me_, the
explanations make sense and don't reek of 'let's do this because it's
more fun than fixing bugs'. I don't honestly know who you could ask who
would know enough to pass judgment but _also_ be independent.

> 4) in retrospect, did Fedora adopt the new drivers at the right point in
> their lifetimes?  This is tricky - on one hand we want Fedora to be a
> great environment for development of new technologies.  But, to be held
> in balance is that certain things, like video not working, can bring
> huge other parts of the development process to a halt (e.g. I've got an
> OpenGL app that can't be developed on a current Fedora on any of the
> several hardware devices I have at hand, so that proceeds on another
> distro).  

Judgment call. I don't think there was a 'right' answer to this; I could
argue it either way depending on how I was feeling that morning :) I
don't think that going with the newer versions of intel and raden in F11
was definitively wrong, put it that way.

> I'm reminded of the recent effort to move opensync from .3
> back to .2 because the upstream acknowledges widespread brokenness and,
> as such, asks distros to not package .3.

it's not entirely analogous, because that _was_ a case of speculative
development. This is an area where I know whereof I speak :). The
opensync developers will openly admit, if you ask them, that the
development process for 0.3 is/was an excessively screwed pooch, it was
badly handled from the start. The re-development of intel and radeon at
least has a clear roadmap and convincing start and end points, and is
generally hitting those marks so far.

-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Fedora Talk: adamwill AT fedoraproject DOT org
http://www.happyassassin.net