Re: Limiting non-build-time relationships to a set of architectures?
[ CCing #400322 for the additional data. ]
On Tue, 2010-02-09 at 20:25:11 +0100, Cyril Brulebois wrote:
> Frans Pop <firstname.lastname@example.org> (09/02/2010):
> > On Tuesday 09 February 2010, Cyril Brulebois wrote:
> > > Frans Pop <email@example.com> (09/02/2010):
> > > > This format is not (yet) allowed by policy: rootskel-gtk
> > > > (>=0.05) [!s390] (except for build dependencies)
> > >
> > > AFAICT, it just works, and not only for Build-Depends. It can't be
> > > used for an arch: all package, though, since it gets substituted
> > > at build time, so it probably won't do what you would want.
> > I know that it is going to be allowed in the future and because of
> > that I don't doubt that it (mostly?) works. But AFAIK *currently*
> > it's not allowed by policy , except for build deps. And thus it
> > should not yet be used in uploads.
Actually checking now, there does not seem to have been any wording
proposed yet on #400322, I might try to come up with one.
> Oops, indeed. Looks like I forgot about that particular point, thanks
> for pointing this out. It looks like I've been taking it granted for
> quite some time.
Hmm, I also seem to have forgotten about this (I'll call that fair
bias :). I was curious anyway about how long this support has been
around as I thought it had been long, so did some digging the other
* Introduced in dpkg 1.10.11 (Tue, 16 Sep 2003 12:52:11 -0500)
* Regression in dpkg 1.10.14 (Fri, 19 Sep 2003 12:29:34 -0500)
* Fixed again in dpkg 1.13.17 (Mon, 20 Mar 2006 03:33:03 +0200)
Bugs: #252657, #324741, #347819
Also I don't think much tools except for dpkg-dev scripts actually
parse the dependency fields in the binary package stanzas in
debian/control. So this should supposedly not break stuff (but then
I've not checked, etc).
> > A reference to an (official) statement from FTP-masters would be.
> I'd rather have -policy@ folks share their mind about it. I guess
> updating the Policy to allow limiting non-build-time relationships
> (Depends, Recommends, …) to some architectures would be nice to
Right. And I don't see why ftp-masters would have a special say on
this issue either.
> I'm not sure whether warning people about the substitution which
> happens at build time would have its place in the Policy since that
> could be considered an (dpkg-dev) implementation detail (but that can
> cause some headaches).
>  Meaning an Architecture: all package with Depends: foo [bar] will
> have foo, or won't have foo, depending on which architecture it
> will be built upon, rather than the conditional Depends stored in
> the resulting binary.
Yeah, this is something the dpkg scripts should detect and error out
on, I've added it to the TODO list (to be pushed).