Re; 4KSTACKS again.

Andy Ross andy at plausible.org
Tue Apr 13 23:15:07 UTC 2004


William Hooper wrote:
> This is a test release.  Period.  Yes, forget about the end user,
> because this is a test release.  nVidia's drivers are a black box.
> There is no way to tell if they are broken because of a change that a
> developer made or a wrong assumption nVidia made.  Using developer
> time to probe a black box with a stick is a waste of developer time.

And I was _so_ careful to ask that no one flame over this issue. :)

Just to be clear: no one asked that the patch be reverted.  No one
asked that NVidia drivers be supported.  No one asked for end user
support for a test release.  And most *certainly* no one asked for Red
Hat to to audit NVidia's drivers for stack overflows.

What *I* asked was for someone to explain the reasoning behind the
current FC2 kernel configuration, which is different from the official
2.6.5 kernel in that it has this feature hard wired into the source.

The answer (perhaps delivered with more bile than necessary, but
nonetheless helpful) is that Linus/Andrew et. al. intend that the
4KSTACKS feature go into the mainline in the near term, perhaps with
the 2.6.6 kernel.  So the incompatibility in the Fedora kernel isn't
gratuitous, but simply an "early access" breakage that NVidia would
have had to deal with eventually in any case.

Mind you, I still think that's a poor idea; especially if the feature
gets pushed back to the 2.6.7-8 time frame and Fedora ends up stuck
with a "3D-incompatible" kernel for ~4 months until FC3 comes out.
Other distributions shipping less modified 2.6.5 kernels are not going
to have this problem, placing FC2 at a competitive disadvantage.

I'm happy to accept this issue as-is and compile my own kernels
because I like Fedora for many other reasons.  But quite honestly, I
can't see myself recommending it to novice FlightGear users until
NVidia gets their drivers patched.  Pointing fingers at the
prioprietary vendors may very well be justified, but when the
"official" kernel still works fine out of the box, the argument sounds
kind of empty.

Andy





More information about the fedora-test-list mailing list