Packaging/Review Guidelines change

[Paul Howarth]

On Thu, 2006-01-05 at 14:42 -0600, Tom 'spot' Callaway wrote:
> On Thu, 2006-01-05 at 21:04 +0100, Nicolas Mailhot wrote:
> > Tom 'spot' Callaway wrote:
> > > As approved by FESCO today, the following change to the Fedora Extras
> > > Packaging and Review Guidelines is now in effect:
> > > 
> > > MUST: Packages must not own files or directories already owned by other
> > > packages. The rule of thumb here is that the first package to be
> > > installed should own the files or directories that other packages may
> > > rely upon.
> > 
> > What about directories owned by already existing packages, when the 
> > already existing packages are not needed for the new package (and may 
> > not be present on the target system) ?
> > 
> > Should a dep be added just to make sure the other existing package is 
> > installed before the new one ?
> > 
> > Or do we allow packages to install files in directories which are not 
> > owned by anything on the live system ?
> > 
> > Or should the shared directories be migrated to a central package like 
> > filesystem (or the equivalent for extras) which will own them ?
> 
> I'm pretty sure all of the FHS directories are owned by filesystem, so I
> don't think we'd need to add too much there...
> 
> The only exception case is where two packages use the same directory
> structure (outside of the FHS), but neither relies on the other one
> (aka, either one could be installed first, or without the other). Then,
> and only then, its acceptable for both packages to own the directory, as
> they both have the potential to be "first". This is a corner case,
> though.

It's not that much of a corner case; it happens all the time with perl
modules.

For example, perl-A-B and perl-A-C (if both noarch packages) may be
completely unrelated packages and both will probably want to own
%{perl_vendorlib}/A/

Another one: perl-A-B and perl-A-B-C, where the latter is an enhancement
of the former, may both currently own %{perl_vendorlib}/A/B/. Now it
could be argued that since perl-A-B-C depends on perl-A-B, it shouldn't
own %{perl_vendorlib}/A/B/, but since rpm will remove the packages in an
unspecified order if removed together (as I've been told in a package
review recently), if perl-A-B is removed before perl-A-B-C then there
will be an unowned directory left over.

Paul.