[Yum-devel] [PATCH/RFC] Follow arch a bit better with obsoletes

[seth vidal]

On Tue, 2007-10-09 at 20:43 -0400, Jeremy Katz wrote:
> On Tue, 2007-10-09 at 20:27 -0400, seth vidal wrote:
> > On Tue, 2007-10-09 at 17:07 -0400, Jeremy Katz wrote:
> > > Currently on multiarch systems, you can end up getting a new arch of a
> > > package pulled in if it's done via an obsolete rather than an update
> > > (rh#301661).  The attached patch is a first crack at making the behavior
> > > for obsoletes match that of updates.  Best way to make clearer is
> > > probably an example...
> > > 
> > > On your x86_64 system, you removed all i386 packages.  Now, there is a
> > > new NetworkManager package which obsoletes dhcdbd.  The repository
> > > contains NetworkManager.i386 and NetworkManager.x86_64.  With the
> > > current code, you end up with both x86_64 and i386 packages installed.
> > > With the attached patch, you'll only get NetworkManager.x86_64.
> > > 
> > > I've double checked that arch changing works (ie, NetworkManager.i386 is
> > > the only thing in the repo and it obsoletes dhcdbd.x86_64) and I can't
> > > see any reason this would break.
> > 
> > Just at first glance this looks like that foo.noarch won't be able to
> > obsolete bar.i386, though. I'll do a test in a few but have you already
> > checked that?
> 
> foo.noarch won't obsolete bar.i386 only if there's also a foo.i386.  Now
> for dinner.

so isn't this a non-starter, then? If we don't let noarch ->arch and
arch->noarch obsoletes then we'll be regressing vs current behavior.

-sv