Re: make 4.0: archive rebuild resulted in 73 packages broken (help wanted)

[Guillem Jover <guillem@debian.org>]

Hi!

On Tue, 2014-04-29 at 21:53:31 -0700, Russ Allbery wrote:
> Manoj Srivastava <srivasta@ieee.org> writes:
> >         I will cut a normal bug on dpkg, and a serious one on make, and
> >  make the former block the latter while we figure otu what to do. The
> >  options, as I see it are:
> 
> >  1) Do nothing. retain make-3.81 in Debian forever more. Needless to
> >     say, this is not very attractive. Pro: There is no action to
> >     take. Con: Almost every other distro is shipping a more recent
> >     make. We will continue to diverge from everyone else, and already
> >     the featires have diverged enough that people are having to add
> >     special cases in the vuild system for the Debian family of
> >     distributions.
> >  2) Hack dpkg-buildpackage to always load B-D-I, and go back to just
> >     calling ./debian/rules build. This is what we used to do. Pro: it
> >     is pretty easy to do (umm, I would think, but I don't know the dpkg
> >     code base so well anymore). This has the con of the inefficiency we
> >     have tried to eliminate, in that all the build dependencies are
> >     loaded for every build, even when not strictly needed.
> >   3) We state that packages must provide build-arch and build-indep for
> >      Jessie. This should trivially be true for every package using cdbs
> >      or debhelper (or, heaven forbid, my old home brew build system),
> >      and have dpkg-buildpackage call them without testing to see if they
> >      exist.  We would need to do another archive rebuild with the
> >      modified dpkg-buildpackage to see how many packages do not
> >      actually not implement these targets.
> 
> Well, 2 is going back on something that we're trying to transition, and 1
> seems obviously unacceptable.  3 is where we were trying to get to anyway.
> I vote for just biting the bullet and trying to do 3 for jessie.

I think we should first understand why the detection is failing with
the newer make. I'm taking a look now. Once that's done we might just
be able to fix (or workaround) one of:

  * make
  * dpkg-buildpackage
  * affected packages

If none of the above can be fixed then I'm all for just removing the
detection code, although that might be painful, AFAIR the numbers were
a bit scary last time Roger posted them?

Using the debhelper detection logic, which implies parsing make output
looks very much unappealing to me.

Thanks,
Guillem