Re: build-arch and autobuilders ?
On Thu, Sep 12, 2002 at 11:57:25AM -0500, Manoj Srivastava wrote:
> >> I thought (as outlined in a related bugreport, although my words in
> >> this report were a bit confused) that the policy should have made the
> >> binary-arch target mandatory, so that the atobuilders could know from
> >> the declared standard-version whether the target was expected or not.
>
> Well, since policy currently lists build-arch and build-indep
> as optional, and a large number of packages provide neither, we would
> need to work through a transition period if we were to mandate
> it. Also, since targets are not really something that helper packages
> provide (as far as I am aware), every rules file would have to be
> tweaked manually; and this is likely to take time. So, mandatory
> would happen in a couple of releases; but I would think that we would
> need a solution faster.
If policy >= X.Y.Z mandates build-arch, then autobuilders could check
standard-version to decide. If a package with newer standard-version does
not provide it, then it's considered a serious bug, and the autobuilders
will even take care of not letting those propagate into testing.
Anyway, if/when a new policy version requires that, and then a package
appears referencing this new version, I find reasonable to require a small
edition of the rules file.
> Julian> There was a long flame^Wdiscussion a while back about the possibility
> Julian> of doing something like the following:
>
> Julian> ret=$(set +e; debian/rules -q build-arch >/dev/null 2>&1; echo $?)
> Julian> if [ $ret -eq 2 ]; then
> Julian> debian/rules build
> Julian> else
> Julian> debian/rules build-arch
> Julian> fi
>
> This sounds workable, does not require most package's rules
> files to be edited, and works today.
1. this does not help to give provision for non-makefiles debian/rules
2. there may be a buggy make in sid, or I miss something. Given a makefile
(not a rules file, but that should not matter) containing:
|boot:
| @ for p in $(BACKENDS); do \
| echo ">>> $$p"; \
| $(MAKE) boot-$$p || exit 1; \
| done
When the target has already successfully built, and I check that running
just "make boot" succeeds with exit code 0, with -q I get:
biglook-alpha[1092]$ make -q boot; echo $?
>>> gtk
make: *** [boot] Erreur 1
2
=> obviously some command is run
=> make.info says for exit code 2: "It will print messages describing the
particular errors.". Hm.
Or did I miss something ?
--
Yann Dirson <ydirson@altern.org> | Why make M$-Bill richer & richer ?
Debian-related: <dirson@debian.org> | Support Debian GNU/Linux:
Pro: <yann.dirson@fr.alcove.com> | Freedom, Power, Stability, Gratuity
http://ydirson.free.fr/ | Check <http://www.debian.org/>
Reply to: