[Cooker] XFree86-4.0.99.900-2mdk

2001-05-21 Thread Arnd Bergmann

Hi Frederic!

-2mdk looks ok here and fixes the problems that sneaked in between 
my 4.0.99.3 and your 4.0.99.900-1mdk build (empty font files,
duplicate symbols in libtype1.a) as well as most of what I introduced
(3dfx breakage, mesa conflict).

There are still two points left where you did not accept my changes,
maybe with a good reason. I'll just clarify what I did.

1. gccmakedepend creates wrong dependencies for *.S files. This causes
   the build of the xc/extras/Mesa/src/X86/*.S to fail.
   I fixed this by patching gccmakedepend (%patch11), you did the
   compile twice. My approach seems cleaner, but it could have some
   side effects, since I am not sure why they put that strange hack
   in in the first place.

2. The type 1 font loader has a problem when compiled with -ffast-math,
   as I have said before. The current solution in cooker seems to
   be to avoid the use of type 1 fonts in the font server config. I 
   think it should be fixed anyway, so I took out that optimization
   in my spec file. My last demonstration screen shot did not seem
   to be reproducable everywhere. I made a new test with xfontsel
   with both my old 4.0.99.3-1mdk build and your new 4.0.99.900-2mdk
   one in the hope that you can reproduce that and will do something
   about it.
   See my screenshot on http://users.evitech.fi/~arndb/cooker-type1-fonts
   Switching off -ffast-math globally could have some performance
   impact, but maybe you have an idea for how to do it only for
   lib/modules/fonts/Type1/type1.c.

There also needs to be done something about the drm kernel modules.
At least Matrox acceleration does not work without the new module.
My suggestion is to put the modules in an extra source RPM (I already
sent you a .spec), so people like me who build the plain
linus kernels can also use them, but for the distribution, it
would be cleaner to put them in the normal kernel source RPM.

regards,

Arnd 






Re: [Cooker] XFree86-4.0.99.900-2mdk

2001-05-21 Thread John Cavan

Arnd Bergmann wrote:
 1. gccmakedepend creates wrong dependencies for *.S files. This causes
the build of the xc/extras/Mesa/src/X86/*.S to fail.
I fixed this by patching gccmakedepend (%patch11), you did the
compile twice. My approach seems cleaner, but it could have some
side effects, since I am not sure why they put that strange hack
in in the first place.

A similar problem, which may or may not be related, was noted on the DRI
list for the CVS trunk. Mike Harris of Red Hat included the line:

#define PreProcessCmd   cpp

In the associated host.def and apparently the build problems with Mesa
went away. It seems to have the DRI/X guys confused as to why... so YMMV
in this regards.

John