(In reply to Michal Luscon from comment #0)
> Please add installonlypkg(kernel) provide into kernel-pae subpackage.

Please remind me what isntallonlypkg means, and why this would be needed.  Also, why only the kernel-pae subpackage and not all main kernel packages?

It is meant to be a hint for dnf that a package contains kernel and therefore it should be treated as installonly (http://dnf.readthedocs.org/en/latest/conf_ref.html#installonlypkgs-label). You are right that all kernel subpackages providing kernel which are not explicitly named kernel should contain installonlypkg(kernel) provide.

I mentioned kernel-pae just because of https://bugzilla.redhat.com/show_bug.cgi?id=1306057 for the big picture .

(In reply to Michal Luscon from comment #2)
> It is meant to be a hint for dnf that a package contains kernel and
> therefore it should be treated as installonly
> (http://dnf.readthedocs.org/en/latest/conf_ref.html#installonlypkgs-label).
> You are right that all kernel subpackages providing kernel which are not
> explicitly named kernel should contain installonlypkg(kernel) provide.

So, it seems we keep piling on hacks here.  DNF is hardcoded to look for kernel, and that's it.  Why would we keep adding installonlypkg to every kernel subpackage that gets added just to work around DNF's hardcoding?

Wouldn't it be better to simply add installonlypkg to all the kernel packages and just rely on that?  It seems less error prone in the long run.

Also, from an implementation standpoint we're going to have to do that anyway due to the way that we generate the main kernel packages from the spec file.

This has been added to rawhide for the moment, it should trickle down as releases get rebased, but since it changes the way updates are handled, I don't want to push it everywhere immediately.

We did this as Justin said, but there are related issues in dnf because of this.  That is being tracked in bug 1333591