rpms/fence-agents/F-11 fence-agents.spec,1.13,1.14

[Michael Schwendt]

On Tue, 27 Oct 2009 12:17:30 +0100, Fabio wrote:

> >> +%if 0%{?fedora} >= 12
> >> +Requires: libvirt-client
> >> +%else
> >> +Requires: libvirt
> >> +%endif
> >> +
> > 
> > What is this explicit dependency on a package name supposed to achieve?
> 
> > There is the automatic arch-specific dependency on the libvirt SONAME
> > already, and it is tons better than a non-arch-specific and version-less
> > dependency on a package name.
> 
> The dependency on the library is pulled in via fence_xvmd that might or
> might be not build (depending on ./configure invocation).
> 
> virsh used to be part of libvirt in any release before F12. It´s now
> moved to libvirt-client.
> 
> So while rpm resolver does the right thing for fence_xvmd and pulls in
> the right soname Requires, it cannot detect the usage of virsh within
> fence_virsh.

It's good practise to add a comment to the .spec file that explains
this explicit dependency.
 
> If there are better ways to handle it, I am absolutely happy to change
> the spec file but I don´t think it is correct either to break
> fence_virsh because somebody is not building fence_xvmd* (that is going
> to be deprecated upstream btw in not too long future).
 
> I also considered a specific file Requires: /usr/bin/virsh, but policy
> suggests to avoid that for different reasons.

Really? What policy is that? Programs in bin paths are covered by the
primary metadata. Such a dependency would be more accurate.