Bug#655975: britney: needs to grok multi-arch
On 2012-01-15 17:10, Adam D. Barratt wrote:
> Package: release.debian.org
> User: email@example.com
> Usertags: britney
> britney will need to be taught to understand how to migrate packages in
> a world with cross-architecture dependencies.
> At the very least, things like "foo/i386 depends on bar:amd64" should
> work. Might there be merit to also checking the values of the m-a
> control fields and verifying that the resulting migration makes sense?
A minimal checking of the M-A field is probably in order and should be
trivial to do as well. That being said, do we have any general idea of
how to handle this?
Personally I smell two rather non-trivial issues. The first is that in
order to validate a "foo/i386 depends on bar:amd64" relation, we will
most likely have to give up the "architectures can be checked in
isolation" approach we are doing now. Particularly, "RDEPENDS" (and
"RCONFLICTS"?) can suddenly refer a foreign architecture.
Secondly, the dark corner of Britney (otherwise known as the "C-lib" or
her "installability tester") would need to be updated as well.
On a related note: Currently it only gets one architecture and that
already causes the occasional "AIEEE" issue. If we have to add all
(non-break/non-f***ed) architectures into one big pile, I suspect that
this part might just roll over and die in an endless series of "AIEEE".