rawhide report: 20050819 changes

[Florian La Roche]

On Fri, Aug 19, 2005 at 03:14:59PM +0200, Matthias Saou wrote:
> Jeremy Katz wrote :
> 
> > On Fri, 2005-08-19 at 13:31 +0100, Paul Howarth wrote:
> > > Are there any advantages in *not* having the Requires:?
> > 
> > If there aren't requirements on kernel packages, then buildroots don't
> > need to have a kernel available/installed which speeds up the creation
> > time for them.  
> 
> So we're back to the "missing" feature of rpm that would be :
> 
> "_If foo is installed_, then require it with this version dependency"

A review of te packages in question and their runtime behaviour
should be enough. In general we wouldn't support running a 2.4 anymore,
probably also with packages which don't list this requirement.
For a few items during the 2.4 -> 2.6 transition I could see how
lvm2 would need a "Requires: kernel >= 2.6", but now we should just
put a runtime check into the relevant places and be done.

More critical is to support updates from older releases and in that
case not listing the requirements for the kernel does help getting things
right. Also buildroots or now xen installs might benefit from less data,
but we don't optimize for those cases too much in general. They still
benefit a little bit.


> These, as well as "suggests" and such have been discussed many time, and
> the general conclusion is that rpm won't ever be able to do that. At least
> not without major rewrite/breakage/step-forward.
> 
> If only it was as easy as "Requires: ?kernel >= 2.6.12" :-(

Yeah, we don't optimize for fine-grained deps and I think that is the
right decision. Adding those other parts wouldn't be that hard, but
then changing all packages to get this right plus also all tools who
work ontop of rpmlib would generate more work than we would get benefits
from it. Enough other work left where we can improve rpm packages and
details of how updates work.

greetings,

Florian La Roche