Re: Re[2]: [PATCH] Revamp mouse support in GPE

On Fri, 2006-08-25 at 13:49 +0300, Paul Sokolovsky wrote:
> > Have you tried MACHINE = "qemuarm" in OE?
> 
>   So, far, only to build kernel ;-) .

:)

> > I realise its not the same image but it will be *very* similar, enough
> > for development purposes and the mouse issues should already be already
> > taken care of. I know they are in poky (where qemuarm came from) and if
> > OE .dev needs updates, that should be straightforward.
> > multimachine.bbclass means you won't need to build two whole images as
> > most packages can be reused from one machine in another.
> 
>   Yes, Koen also mentioned that it might be solved there. I'll be
> having look at them (may take some time ;-) ). But that's of course
> true that images are very similar, that's exactly why I would like
> them to just do dynamic detection/adjustment, unless it makes no sense
> too. I'm not sure if OE people has such an aim as priority, but I'd
> guess, it's nice thing, because building 20+ images (count PocketPC
> devices in ports) which differ in less than 1% of content just
> doens't scale. Just have 2-3 functional types of images (even at the
> price of groing each by some mere 10k), plus device-specific
> kernel/modules (and of course, installer being able to combine them
> during install). I'd be interested to work in that direction.

For the specific problem in question, I suspect the best way forward is
to adapt tslib-conf to do runtime detection (like Xserver) instead of
having config files for every machine as we do at the moment. Most of
the 2.6 machines are standardised now anyway. If done properly, we'd
then have variables set that X can then work against.

FWIW, I also have an new qemuarm config that can boot OE's ext2 images
directly rather than using nfs. Both have advantages/disadvantages but
its nice to have the choice! :)

Cheers,

Richard