Re: A new check for debian-installer udebs depending on normal debs?
Jérémy Bobbio <lunar@debian.org> writes:
> I would like to see if a new lintian check could be added that would
> help avoiding an issue affecting the debian-installer which happens
> quite frequently during Debian development cycles.
>
> To give you a current example, the debian-installer daily builds are
> currently broken due to #440731 [1]. This bug is quite simple: the
> debian-installer "udeb" created by brltty depends on a normal "deb"
> package.
>
> [1] http://bugs.debian.org/440731
>
> The debian-installer is actually constituted by "udebs" which are
> just reduced Debian packages. In most of their aspects, they are pretty
> close from normal Debian packages: they have Depends, Provides and
> similar fields. They currently all belong to a separate section of the
> archive (debian-installer/main).
>
> Although, except a few exceptional cases, an "udeb" cannot depend on a
> normal Debian package. Everytime this happens, something breaks in the
> debian-installer.
>
> The brltty issue comes from a new upstream version adding support for a
> new external library that is not available for the debian-installer.
> Thus, dpkg-shlibdeps just picked the normal Debian package to create the
> substvar.
>
> It would be great to have a lintian check that would at least warn
> maintainers of such situation. This would allow them to fix their
> packages by adding the necessary --disable options instead of having to
> go through a serious bug.
>
> Do you feel that this kind of check can be added to lintian within the
> current framework? Is anything unclear? Would anyone more familiar
> with the Lintian code base than people in the debian-installer team be
> inclined to work on adding such check?
I finally had a chance to read this over, and yes, this should be fairly
straightforward. Basically, in checks/fields when we check dependency
fields, see if we're checking a udeb and, if so, make sure that all the
dependencies are on packages ending in -udeb.
Could you file a wishlist bug against lintian so that we don't lose track
of this and, ideally, list the exceptional cases you mention above so that
we can exclude them and not create false positives?
Thanks!
--
Russ Allbery (rra@debian.org) <http://www.eyrie.org/~eagle/>
Reply to: