Indeed, I am suggesting it should be automatic at this point. Or we genuinely do clean up the whole archive preemptively (and maybe make an overrideable lintian warning about not declaring M-a for arch:all packages). If we do make it automatic Debian would need to go the same way, and we would need to solve the inverse case of the interpretters as well (a new M-a field for the exceptional packages?)
On Mon, Oct 6, 2014 at 7:28 AM, Dimitri John Ledkov <firstname.lastname@example.org> wrote:
Original multi-arch spec did not resolve arch:specific transitiveOn 6 October 2014 03:29, Barry Warsaw <email@example.com> wrote:
> On Oct 05, 2014, at 04:28 PM, Scott Ritchie wrote:
>>Would now also be a good time to look through the archive and assess the
>>various arch:all packages that do not explicitly declare themselves as
>>Multi-Arch:foreign (and thus not installable as dependencies of foreign
>>architecture packages?) Is there a reason for any such arch:all package to
>>not declare this field at this point?
> If that were the case, why would it not just be automatic?
dependencies behind arch:all packages. For example, an arch:all
python-foo which depends on python-pycurl needs not just
any-architecture python-pycurl but the one that matches the
architecture of the interpreter. There were further BOFs, discussions,
proposals and solutions to this but I don't think this has been fully
implemented for each interpreter yet.
But in general, yes, e.g. all fonts, docs, locales, etc. packages
should declare M-A:foreign