Add cost=1100 in epel.repo?

[Paul Howarth]

On Fri, 27 Aug 2010 13:27:52 -0600
Kevin Fenzi <kevin at scrye.com> wrote:

> On Tue, 24 Aug 2010 10:48:17 -0600
> Stephen John Smoogen <smooge at gmail.com> wrote:
> 
> > On Tue, Aug 24, 2010 at 04:23, Paul Howarth <paul at city-fan.org>
> > wrote:
> 
> ...snip...
> 
> > > Any reason why this shouldn't be done?
> > 
> > I believe in the past we have asked the following:
> > 
> > 1) Only  build for the architectures that are excluded (eg build it
> > for ppc64 and not x86_64 or i386)
> > 2) Keep to the same version or make the release number smaller than
> > what is in EL-5
> 
> Yeah, I guess I was thinking we would only be building these on the
> arch that is missing them, but then thinking about that it means we
> have to touch the spec and can't just use the exact same one from
> RHEL. 
> 
> So, I guess I am not opposed to adding a cost in. 
> Should it be 1000? Or does it matter? Does
> centos/rpmforge/dag/dries/iuscommunity use Cost? Can we make sure
> anything we add doesn't cause unneeded problems for them down the
> road?

1000 is the default so it needs to be a number larger than that to have
the desired effect.

I haven't seen anyone else using cost yet, but having a significant gap
between the default cost (1000) and the EPEL cost (e.g. 1100) would
give plenty of scope for other repos to insert themselves wherever they
wanted on the cost scale.

Also bear in mind that cost only comes into play when NEVR+arch are
identical, and many of the other repos are happy to replace base
packages with rpm-newer ones (in fact that is the raison d'être of some
repos), in which case cost wouldn't come into play at all.

Paul.