Adding Provides: strictdep-kernel[-flavour]-arch = kernelevr to kernels (was: Nameing guideline ...)

Axel Thimm Axel.Thimm at physik.fu-berlin.de
Fri Dec 19 00:01:29 UTC 2003


On Thu, Dec 18, 2003 at 03:43:54PM +0100, Nicolas Mailhot wrote:
> Le jeu 18/12/2003 à 15:11, Axel Thimm a écrit :
> > On Wed, Dec 17, 2003 at 04:59:53PM +0100, Nicolas Mailhot wrote:
> > > Le mer 17/12/2003 à 15:51, Chris Ricker a écrit :
> > > > On Wed, 17 Dec 2003, Nicolas Mailhot wrote:
> > > > 
> > > > > So make the SMP one require the UP.
> > > > > Frankly SMP alsa users are the minority, they'd better have an UP kernel
> > > > > as fallback, so they can live with two alsa modules without polluting
> > > > > normal user systems.
> > > > 
> > > > Why? My SMP boxes don't have a UP kernel on them.
> > > 
> > > Well, better :
> > > - SMP users shall have SMP and UP code
> > > than :
> > > - everyone shall have SMP and UP code
> > > 
> > > And anyway who is going to use 2.4 alsa on smp now ?
> > 
> > The naming guidelines should serve more than alsa on 2.4 up machines
> > ;)
> > 
> > I had gone through various models and finaly arrived at the same
> > solution as Fernando. It is the only maintainable way for supporting
> > all the released kernels.
> 
> Well I'm not arguing against strict rules but forcing SMP and  UP in the
> same package. Seems way to kludgy for me, when splitting in two packages
> can be done easily.

Yes, splitting into two or even three (including bigmem) is
mandatory. Sombody argued in favor of all-in-one because of lazyness
to check for matching kernel and kernel modules. When having strict
dependencies you don't run into the danger of having wrong
kernel/module matching.

> As far as I'm concerned we should have module requires exact kernel for
> which it was build, and if someone could add a gcc check somewhere this
> probably be all to the better.
> 
> (with 2.6 if we can get the config used exported in /proc/config.gz I'm
> all for requiring a config check at install time, even if it means being
> unable to install modules for a not-live kernel. Mismatching modules and
> kernel can be *nasty*)

You can use /boot/config-<kernelevr> already now, and the kernel needs
not be live. A possible way to do it would be to match with

Provides: kernelconfig-md5sum = 0bee89b07a248e27c83fc3d5951213c1
and
Requires: kernelconfig-md5sum = 0bee89b07a248e27c83fc3d5951213c1

in kernel and kmdls respectively. That would be the utimate check and
(almost) superseeds the strictdep-kernel idiom (still need to match
the underlying kernel sources, this will probably have to stay the
kernel's evr).

Nice idea, Nicolas!
-- 
Axel.Thimm at physik.fu-berlin.de
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.fedoraproject.org/pipermail/devel/attachments/20031219/0d532d79/attachment-0002.bin 


More information about the devel mailing list