RFC: kernel-modules in Fedora Extras

[Ralf Corsepius]

On Sat, 2006-01-07 at 14:20 +0100, Thorsten Leemhuis wrote:
> Am Samstag, den 07.01.2006, 07:25 +0100 schrieb Ralf Corsepius:
> > On Fri, 2006-01-06 at 18:17 +0100, Thorsten Leemhuis wrote:
> > > Am Freitag, den 06.01.2006, 18:04 +0100 schrieb Ralf Corsepius:
> > > > On Fri, 2006-01-06 at 17:26 +0100, Thorsten Leemhuis wrote:
> > > > > Am Freitag, den 06.01.2006, 11:12 -0500 schrieb Dan Williams:
> > > > Consider the current nvidia disaster people are facing with the kernel
> > > > drivers on livna.
> > > 
> > > You mean this "disaster" that is around for ~two years already?
> > If you mean that kernel-module updates with livna modules have never
> > worked smoothly with yum, yes.
> 
> That's why we discuss this stuff here and work on a proposal that works
> better.
> 
> We know that we need support from yum (or a yum-plugin) to make it work
> smoothly. 
> 
> > >  Seems people ignored it for a long time and it works mostly fine for
> > > a lot of people. 
> > It doesn't work for anybody.
> 
> We know. 

I didn't necessarily have this impression :(

> > The only way I see, is to add the missing packages to the repos, i.e. to
> > rebuild the kernel-modules and module-userspace-libs for all kernels
> > having ever been shipped for a Fedora release: 
> 
> This was discussed earlier. People didn't like the idea - they only
> wanted to build for the latest kernel.
Sorry, but these people have no idea about what they are talking.

You are shipping a broken distribution. Therefore you can't avoid fixing
it. This time it affects livna and nvidia, but the problem actually is
much more general and at least affects all kernel-modules which are
accompanied by userspace libraries/tools

> I disagree -- we IMHO should build for the latest kernel and the one
> that was shipped in core.

... and at least the (latest-1)  kernel, because this is the kernel
people which regularly (but not daily basis) update are using.

>  Why? There are people with modems or
> bandwidth-per-month limitations that don't update the kernel. Yes,
> that's dangerous, but people do it. 

Another aspect you seem to ignore: It's not uncommon that kernels are
broken on a particular HW - Therefore users, can't avoid resorting to
using older kernel.
 
> But we can't build for all kernels.
I don't buy that.

Write yourself a script to generate the specs, and let them be built on
a built system.

>  This wastes to much time IMHO.
Building will probably take will take some hours, but even some
individual rpms probably are more demanding than building this bunch of
packages.

> And
> old kernels are deleted for the updates-repo anyway 
Then building for all those in updates + core would be a reasonable
compromise.

Ralf