dkms for fc7?

[Denis Leroy]

Jeremy Katz wrote:
> On Mon, 2007-01-22 at 17:10 +0100, Matthias Saou wrote:
>> Jeremy Katz wrote :
>>> Heck, why don't we also stop packaging perl modules as RPMs.  I mean,
>>> there's CPAN, right?  And there's now the cheeseshop + setuptools for
>>> python.
>> Well, we just don't push out a new perl version to the released updates
>> every few weeks :-)
>> If the Fedora kernel never got updated, and only got bugfixes, then
>> maybe kernel module packages would work well enough with weak symbols
>> and such... but it would mean just too much wasted effort.
> 
> If only upstream would work like that :-)
> 
> But seriously, I am pretty vehemently opposed to differing databases for
> tracking installed software.  If the benefits of recompiling modules
> automagically is big enough, then we should make sure that the _output_
> of the recompilation is an rpm that we can install and track just like
> anything else.  There's no reason that dkms couldn't do this.  
> 
> At the same time, I don't think that's the default behavior that we want
> for users who want extra kernel modules.  Instead, they should be able
> to download, install and have them work just like the rest of the
> software that we ship.

dkms is solving a problem that people have *now*; frequent 'yum update' 
breakage (unless you use --skip-broken).

Well, with a integrated Core+Extras build system, we could implement our 
own automated kernel module rebuild system, a repo-level dkms if you 
will. That has the advantage of staying "pure-RPM" and also guarantees 
that the module compiles with the new kernel update. So that yum won't 
let you update your kernel if that would mean losing wireless access, 
for example. We'll have to do this to resolve the firefox/galeon 
breakages automatically also...

-denis