On Tue, 2020-03-10 at 23:31 -0400, Daniel Kahn Gillmor wrote:
> Thanks to Ben and Jason for following up here.
> 
> On Wed 2020-03-11 02:52:43 +0000, Ben Hutchings wrote:
> > We definitely can't add a Provides on "real" kernel packages, because
> > this breaks auto-removal of old packages.
> 
> I'm not sure i understand this.  by "real" kernel packages i think you
> mean something like linux-image-5.4.0-4-amd64.
> 
> If linux-image-5.5.0-1-amd64 were installed, with such a Provides:, and
> then linux-image-5.5.0-2-amd64 were installed, also with such a
> Provides:, then why wouldn't autoremoval of linux-image-5.5.0-1-amd64
> still work?  The system would still have the Provides: satisfied.
> 
> Feel free to point me at some piece of Apt or dpkg documentation if i'm
> missing something obvious.

I don't know where/if it's explicitly documented, but it would be with
apt if anywhere.

If some of the packages providing a virtual package are explicitly
installed, and some auto-installed, it could reasonably auto-remove the
latter group (though I don't think it does).  But if all of them are
auto-installed, which will be the case for kernel packages, it can't
tell which should be kept and which removed.

> > We could possibly add it to the meta-packages, but there would have to
> > be a plan for how we can drop it later (and have the Wireguard
> > user-space just assume the kernel supports it).
> 
> When we can just assume that the kernel supports it, we might just drop
> the "wireguard" package entirely, and supply only the "wireguard-tools"
> package (maybe at that point, we make "wireguard-tools" itself Provide:
> wireguard).

But nothing will auto-remove the wireguard package.  So you would have
to keep it as a transitional package for one release cycle, even when
it depends on just wireguard-tools.

Yes, this is annoyingly complicated.  That's why I need there to be a
plan.

Ben.

> At that point, we certainly wouldn't need the Provides: on
> the kernel.
> 
> > We definitely shouldn't accumulate Provides for every component that
> > was previously packaged out-of-tree.
> 
> I can see how that would be problematic :)
> 
>   --dkg
-- 
Ben Hutchings
Humour is the best antidote to reality.

Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to