On 25/02/2025 08:40, Stéphane Glondu wrote:
Hi,
Le 21/02/2025 à 22:43, Paul Gevers a écrit :
The findlib package is blocked out of testing because of autopkgtest
regressions in ocaml-mdx on arm64 and armel:
https://tracker.debian.org/pkg/findlib
I don't understand why they fail only on these architectures, and, looking at
the logs, the failure is due to a failure in installing dependencies, which I
don't understand.
Me neither, but then I don't fully grasp how Ocaml dependencies (and
particularly their Provides) work. I suspect that if we'd ignore these
failures, the next step of britney2 would tell us. (Or maybe not, see below).
I don't know what to do to make progress on this matter... Could someone more
knowledgeable have a look and explain the situation, please?
I can't, but apparently apt couldn't find a solution, not even with everything
from unstable. Which is weird because in unstable, the test runs fine.
During the phase where it tries with "everything from unstable":
Investigating (0) libmdx-ocaml:arm64 < none -> 2.5.0-2+b3 @un puN Ib >
2.5.0-2+b3 is the version from testing, and it seems to stick to this version
for the rest of the resolution. This is wrong. It should work with libmdx-ocaml
from unstable (2.5.0-2+b4).
I wondered why there was no failure on e.g. amd64, so I rescheduled the test for
amd64 and... it failed! My suspicion is that it fails on all architectures, now.
Then, why did it succeed the first time? I suspect some kind of race condition
where the test was run before the old (pre-binNMUed) packages were removed from
unstable. Is that possible?
We were discussing it last night on #debian-release. This is a regression in
autopkgtest, which no longer "falls back to using all packages from unstable".
It's probably going to be reverted temporarily to unblock transitions which are
unable to migrate atm, though in the long term the debci team may look for a
different solution.
In the meantime, I'll add hints to ignore those failures.
Cheers,
Emilio
