On Sam, 2002-08-31 at 00:44, Alan Hourihane wrote:
> On Sat, Aug 31, 2002 at 12:13:08 +0200, Michel D=E4nzer wrote:
> > On Don, 2002-08-29 at 23:47, Alan Hourihane wrote:=20
> > >=20
> > > A more architectural view needs to be taken on how this operates
> > > with other chipsets and remove the whole dependency of doing this
> > > to every driver.
> > After some thinking, something along these lines might work:
> > If a driver can support reinit (i.e. it can deal with the DRI getting
> > disabled and enabled again), it calls a DRI support function to registe=
> > hooks for disabling and enabling. The DRI wraps the EnterVT() and
> > LeaveVT() functions and calls those hooks in its wrapper functions.
> > Would that be any better than the current implementation? It would stil=
> > require about the same amount of code to be added to each driver, but o=
> > top of that it would introduce a dependency on a certain version of the
> > DRI library, and it's not guaranteed to work with all drivers (they
> > might have different ordering constraints in the VT switching
> > functions).
> > So, do we absolutely have to wait until one day, someone figures out an
> > elegant way to unify all of suspend/resume, DRI on multiple displays
> > (both with clients running everywhere at the same time) and reinit? Or
> > can't we have the latter now, and then fold it into the rest on that
> > fine day? It satisfies my needs now, and I can hardly believe that they
> > are my needs alone.
> The problem is, once you let one driver get out of control it takes
> some factoring out to re-review this at a later stage.
'out of control', 'strong impact' (Jens) - this sounds like FUD to me.
I'd appreciate if you showed me it isn't that and elaborated. Bear in
mind this is essentially about a couple dozen lines of code, which
implement a complete feature and are unlikely to change, exploiting the
current infrastructure without requiring any changes to it.
> There's no rush, 4.3.0 isn't around the corner. Take some time and think
> about the EnterVT/LeaveVT alternative you mentioned and present
> something more formal.
Not sure what you expect, neither do I see the point really, as
Again, this satisfies my needs now. I'd rather see those criticizing and
aiming for a more 'general' solution putting forward something.
> See how we can use the current APM code to help too, if you need it.
I don't, it's only relevant for suspend/resume AFAICS.
Earthling Michel D=E4nzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member / CS student, Free Software enthusiast