Bug#1128277: Unsupported foreign architectures should be explicitly excluded

David Kalnischkies <[email protected]> Wed, 25 Feb 2026 09:53:51 +0100

Newsgroups	gmane.linux.debian.apt.devel
Message-ID	<aZ6vb-vmUj2i--dn__9664.50493303768$1772010227$gmane$org@gosurf.lan>

(you may want to CC maintainers on reassign as they otherwise only
 get the easily missed Processed mail, see dev-ref §5.8.3.2)

>> Notice: Skipping acquire of configured file 'main/binary-i386/Packages' as repository 'https://packages.microsoft.com/debian/13/prod trixie InRelease' doesn't support architecture 'i386'
[…]
> No. Apt should stop producing warnings for repositories that do not
> carry an architecture that is not enabled on the system.

(But it is enabled on the system ???)


> The fact that apt produces a useless
> warning that the architecture isn't supported is a bug in apt, not
> extrepo.

Well, the idea is to warn the user that this repository might very well
be causing problems as the repository is not intended to be used with
i386. It e.g. wont provide M-A:same variants for it possibly causing
a removal cascade or blocking [security] upgrades indefinitely and
other (not so) subtil problems users will attribute to apt and/or
MultiArch as nobody said there will be problems otherwise.
(appart from the general implicit warning about repositories
 in general being able to cause a lot of problems, that everyone
 ignores of course)

A user can avoid this warning (which is an interactive notice only)
by not requesting the architecture as you noticed already, which
amounts to a "I will keep the pieces if that breaks" admission.

A repository can avoid it by "suppporting" the architecture –
which doesn't mean they have to build all their packages for i386.
In fact, they don't need to build any package, not even ship a
binary-i386/Packages file – all apt cares about is that the
Architectures field in the Release file contains 'i386'.

What that means is that the repository actually supports being
used on i386 – as in, they won't break M-A systems or if they
do they at least consider it a bug: They e.g. can and will ship
M-A: same variants of libraries they overtake or an M-A:foreign
provider will actually support being called by the foreign i386
packages installed on the user system.


So, I am considering this a bug in the repositories.


(On a more practical note, the code has no idea why i386 ended up
 being configured for acquire, so it has to do something, for the
 possibility of it being an explicit Architectures-Add thing…)


Best regards

David Kalnischkies

signature.asc (application/pgp-signature, 833 B)