[Pkg-dkms-maint] Bug#607903: Fails to build kernel module: missing dependency on libc6-dev

[Phil Miller]

Sorry, that was stupid of me.

That said, dkms's dependencies are weird; it depends on "make |
build-essential | dpkg-dev", which is a rather weird set, since
buildessential depends on make and dpkg-dev, and dpkg-dev itself
depends on make as well. If it really just means 'make', it should
depend only on 'make', and dkms-using packages should depend on things
like libc-dev. If it means other stuff from build-essential as well,
build-essential shouldn't be just one of several options. The
dependency structure of dkms kind of invites occasional brokenness.

On Thu, Dec 23, 2010 at 18:43, Russ Allbery <rra at debian.org> wrote:
> fixed 607903 openafs-modules-dkms/1.5.77-1
> thanks
>
> Phil Miller <pmiller at hmc.edu> writes:
>
>> build-essential depends on "libc6-dev | libc-dev", and is assumed for
>> building any packages, per Policy 4.2.
>
> Yeah, but DKMS modules are something of a special case because the end
> user isn't building the package.  They're just installing a DKMS-enabled
> package, and the building happens as part of the package installation.
> That means that the dependencies do need to be there for the compilation
> to work.
>
> DKMS itself doesn't depend on build-essential because building the average
> kernel module doesn't require userspace development headers, but OpenAFS
> is weird.
>
> This bug has been fixed for a while in Git, but along with other things
> that aren't suitable for the current deep freeze.  (It is fixed in
> experimental, though, so I'll leave it marked closed but with the correct
> version information.)  But there's an upstream security patch that I need
> to upload for squeeze anyway, so I'll fix this along with.  I should be
> able to get to that in the next couple of days.
>
> --
> Russ Allbery (rra at debian.org)               <http://www.eyrie.org/~eagle/>
>