Re: Bootstrapping alex
On Tue, Dec 22, 2015 at 03:32:53PM +0100, Joachim Breitner wrote:
> Am Dienstag, den 22.12.2015, 14:21 +0000 schrieb Colin Watson:
> > I don't think this is actually a problem in practice because hscolour is
> > in ghc's Build-Depends-Indep, not Build-Depends. Perhaps Adrian forgot
> > to use dpkg-buildpackage -B rather than -b?
>
> I think he refers to the dependency on hscolour by haskell-devscripts.
> It was added there because every Haskell package needs this at
> buildtime, although only when building the arch-all packages.
>
> Unfortunately, there is no way to say that a dependency of haskell-
> devscripts is only then a real dependency when haskell-devscripts
> itself is used for an arch-indep build, and otherwise not.
>
> We could drop it from haskell-devscripts and include it in
> Build-Depends-Indep of _every_ Haskell source package, but that feels
> quite redundant.
Oh, I see. I believe that build profiles can only help with build-time
dependencies rather than run-time dependencies, even if the latter are
transitively build-time. Your proposal is probably the most technically
correct one, but as you say it's rather repetitive.
> > > And are build profiles actually already supported by our auto-building
> > > infrastructure? If not, what is the practical gain today?
> >
> > Our buildds support them to the extent that they correctly build with
> > the empty set of profiles, but not otherwise. sbuild supports building
> > a package with a given set of build profiles, and I believe some tools
> > specialised for bootstrapping support them.
> >
> > However, the practical gain is that it gets the knowledge of how to do
> > this kind of bootstrapping operation out of people's heads and into code
> > where porters can do it fairly easily without needing to ask us for the
> > details.
>
> Hmm. I’m a bit worried about bit-rot then. These are quite delicate
> things, with code in debian/rules etc. If they are not exercised
> regularly, will they not break sooner or later?
We've been averaging new ports at least once a year for quite a while
now, so they'll be exercised from time to time, and I don't think the
code is all that complicated: the main thing is that it doesn't break
the common case, and if a porter should have to make a small adjustment
in the course of bootstrapping then that's not too terrible. We'll
still have made their life easier by pointing them in the right
direction.
There are also efforts such as http://bootstrap.debian.net/ which should
eventually reach the point of being able to test this kind of thing
automatically.
--
Colin Watson [cjwatson@debian.org]
Reply to: