RFC: kernel-modules in Fedora Extras

[Panu Matilainen]

On Sat, 7 Jan 2006, Thorsten Leemhuis wrote:

> Am Samstag, den 07.01.2006, 02:31 -0800 schrieb Panu Matilainen:
>> On Fri, 6 Jan 2006, Thorsten Leemhuis wrote:
>>>
>>>> We can also try to special-case the up/smp/hugemem thing if it's a pain
>>>> to do %foreach in the specfile.
>>>
>>> Is there any example how "%foreach" works?
>>>
>>> I still hope we can get the hardcoded variants up/smp/hugemem/xen-foo
>>> somehow out of the spec file and passed as define by the buildsys to
>>> rpmbuild, too.
>>
>> Yes, please. The up/smp/blaa needs to go away from the spec, it's just
>> hideous. My humble 5 cents :)
>
> Panu (or everybody else of course), any idea how to implement it? I
> searched a bit for the syntax of "%foreach", but wasn't able to find
> anything. We of course could add a lot of "--with smp" "--with foo", put
> we still would have to hardcode the names :-| .

I was thinking of putting the looping into buildsys for known kernel 
variants, basically the equivalent of
for variant in <list of known variants>; do
     rpmbuild -bb --define "variant $variant" kmod.spec
done

...and the spec would only have
kmod_prep %{variant}
kmod_build %{variant}
...etc.

But then I wonder do we really want all variants of all modules - for 
example does SMP Thinkpad-modules make sense (does it even build/work at 
all)? That would require the information of what variants to build for 
each kmod to be stored *somewhere* (outside the spec presumably) which 
makes the whole thing kinda moot I guess. So, taking back the "must go", I 
can live with the proposal. :)

P.S. There *is* a yum plugin for automatically updating livna-style kernel 
module packages along with the kernel in yum-utils cvs, just haven't 
bothered advertising it too much as it's been clear for quite some time 
that FE kernel modules are going to be something completely different.

 	- Panu -