Current tinderbox regressions (video drivers)

[Matthias Hopf]

On Feb 25, 09 18:13:07 +0100, Michel Dänzer wrote:
> > Seriously, this change would have been easy to do in a less invasive
> > manner by not nuking the hook, but just not using it any more. And by
> > adding an additional hook for the new buffer types.
> You got that backwards (did anyone actually *look* at the change causing
> this before complaining? :}: I removed the UploadToScratch hook because

Well... Partially :-?

> I had to bump the major version, not the other way around. I had to bump
> the major version (and add the #error) because the EXA_PREPARE_* indices
> are passed to the driver Prepare/FinishAccess hooks, and some drivers
> like radeon(hd) use them as indices into arrays. Without these measures,
> the drivers would have continued to build and appeared to work, but
> could have broken in subtle ways when overflowing the arrays.

I agree that having this #error is much better than letting the drivers
fail subtly. Even if it is the drivers fault by not bound checking (or
had the actual range been strictly defined before?). I explicitly
include radeonhd here (and no, I haven't fixed that yet).

I just fail to see for the moment why the higher indices couldn't be
sent to a different hook - which wouldn't be available if the driver
isn't updated yet. Or is the functionality strictly required for
acceleration to be useful at all?

I guess what we lack is reasonable dynamic (i.e. runtime) API version
validation.

> > Enough ranting, radeonhd should be fixed now.
> Thanks, I hope you checked the Prepare/FinishAccess hooks for the above
> (take a look at my radeon fix).

That was basically the only thing to do. Apart from autoconf magic :-P

CU

Matthias

-- 
Matthias Hopf <mhopf at suse.de>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          mat at mshopf.de
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de