Re: [Glibc-bsd-commits] r5595 - branches/experimental/kfreebsd-10/debian

[Steven Chamberlain <steven@pyro.eu.org>]

On 22:34, Christoph Egger wrote:
> hmm are we going to have a kernel without its own fuse.ko in jessie?

I don't think so.  Only if perhaps this was a sid chroot on a wheezy
9.0 kernel, in which case, fuse isn't expected to work anyway?

> are we going to remove 9 completely?

Of course, I assumed it was going away completely;  if it was staying in
jessie, it should have been updated to 9.3, weeks ago.  I don't expect
we have the resources to support two kernels in stable again (at the
same time as oldstable).

We do have to support sid/jessie chroots on a wheezy 9.0 kernel (for the
buildds) but they won't be using fuse.

> If not maybe fuse4bsd should only be
> changed to depend on either a 10 kernel

This seemed like a good idea...  it addresses the upgrade case so that
a kfreebsd-10 kernel package definitely gets installed.  But this is
still not right:
  * until reboot, the old kernel is still running
  * the user can still choose to boot an older kernel, package
dependencies don't consider this
  * on kfreebsd-i386, we'd need to install version 10+1 of either
kfreebsd-image-486 | kfreebsd-image-686 | kfreebsd-image-xen but
APT isn't likely to know which one is correct, because wheezy d-i
didn't use that metapackage.

So I'm back to thinking userland should never express dependencies on
a kernel package, at all.

I still think kfreebsd-10 images could Provides: + Conflicts: fuse4bsd,
so that reverse-dependencies are satisfied and fuse4bsd can go away.
It means fuse-using packages won't be installable in chroots but maybe
that's correct.

> or fuse4bsd-dkms (why is that a
> recommends btw?)

Don't know.

Regards,
-- 
Steven Chamberlain
steven@pyro.eu.org