Re: Bug#1131311: autopkgtest: Bad versions comparison

David Kalnischkies <[email protected]> Fri, 20 Mar 2026 11:41:57 +0100

Newsgroups	gmane.linux.debian.apt.devel
Message-ID	<[email protected]>

Am Fri, Mar 20, 2026 at 10:27:58AM +0100, schrieb Xavier:
> - resolver should take the higher version

fwiw that is not possible.

Yes, using version provides can produce these funny situations
in which multiple seemingly conflicting statements about a version
are true, but we can't just pick one of the versions and run with it.

If we would always pick the highest as you suggest:
node-rollup-plugin-terser build-depends on debhelper-compat = 13.
Now, that is the highest available debhelper-compat ~ for now,
14 is around the corner making your package FTBFS on its release.
Never mind all the packages that happen to still = 12 or lower.


Now, why not for Breaks? Well, we don't know what the interface is,
but it might very well be file-based, so something providing version 1
provides a file in a certain format in a certain location. If your
package can't deal with that it Breaks version 1 (well, its broken by,
so ideally the other package would have it, but that is not always
possible without access to a time machine). That there is also a
version 2 somewhere else is of no concern for the breakage caused
(so a simple Depends >= 2 doesn't work). Yes, in this situation
a "Depends >= 2" is satisfied, while also a "Breaks <= 1", which is
one of the simpler funny situations I alluded to earlier.


That we haven't yet fully descended into madness is simply a result
of Provides usually being on purely virtual packages and not that
common in general. The various node, rust, … ecosystems with their
auto-generated packages and mega-provides violate both assumptions,
so their maintainers run with some regularity into this fun that is
usually just nightmare fuel for the select few reasoning about
dependencies on the regular for us (like apt).


Best regards

David Kalnischkies

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