[Xorg] Notes from release call

[Adam Jackson]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Friday 09 July 2004 22:00, Daniel Stone wrote:
> On Fri, Jul 09, 2004 at 12:34:29PM -0700, Eric Anholt wrote:
> > - debrix folks (Daniel Stone, Adam Jackson) have made a lot of progress
> > on autotooling the x server but have found sticky issues with modules
> > involving references to global symbols.  They've been fixed in that
> > tree, mostly, but this may prove to be a hazard for vendors moving to
> > .so modules
>
> Global symbols are OK, but they *must* be resolved at load time; they
> cannot be lazily resolved. This means that the ATI driver, where ati had
> global symbol deps on atimisc/r128/radeon, while the reverse was also
> true, could not be loaded. Ever.
>
> Right now, our compromise is that Load "ati" works, but
> atimisc/r128/radeon doesn't. The current thinking (well, for me anyway),
> is to make a fake libatimisc/libr128/libradeon which just depends on
> libati, and then rename the real ones to libati_atimisc, or whatever.

Not to be contrary, but I'd thought I'd resolved that (ie, that you could load 
r128 on its own without ati first).  If not, perhaps everything except the 
chipset detection could be factored out into, say, aticore, which would then 
be loaded by atimisc/r128/radeon.  'ati' would then be shorthand for "I'm too 
lazy to figure out which chip I have, please load the right one for me".  
Alternatively, if there exists a loader call that can query if a module is 
already loaded, we can stick that at the beginning of the atimisc/r128/radeon 
init routines.

Any of the above works for me, but I think I like the aticore approach best if 
it's feasible.

- - ajax
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)

iD8DBQFA71tMW4otUKDs0NMRAjbGAKC4PtTKioeINBcBy39TtPMQ/ihZ5QCgkq5p
5UuKIY6c9vI+rXKx/IZeduc=
=+/89
-----END PGP SIGNATURE-----