Hi Paul, Quoting Paul Gevers (2020-01-05 21:34:07) > On 04-01-2020 22:13, Johannes Schauer wrote: > > yes, I was aware of the introduction of libcrypt, that's why I uploaded > > 0.5.1-3 > > which addresses precisely this change. My naive assumption was, that once > > libcrypt migrates to testing, mmdebstrap would migrate just fine. > > True, if you only look at mmdebstrap. > > > Thanks a lot for opening this bug report because apparently I have a > > misunderstanding about how our CI and migration work together with automatic > > migration and maybe you can help me clear up my understanding of it? :) > > However, there are cases where other packages are blocked due to the > failure of mmdebstrap. Two examples: > > The last couple of days: shadow. shadow somehow triggers britney to add > libcrypt from unstable to the set of packages to test, but NOT > mmdebstrap. The test thus has delayed shadow from migration for about 10 > days. Hence, tests like this are harmful for others. > > Now libcrypt1 migrated, but mmdebstrap did not (due to this bug, hence > downgrading). The old package now fails in testing and that's something > that we consider RC already. However, until the reference test is run, > all packages that run mmdebstrap as test will be believed to cause > regressions. > > Now, you can argue that changes to the base chroot are rare. I don't > know but I could take your word for it. However, *if* they happen, like > now, they do cause issues, like for shadow. If the situation prolongs, > it triggers investigation by others that are not familiar with your test > and it would require an overwrite from the release team if the situation > was critical (e.g. shadow needed to migrate fast) or if it would have > taken longer for libcrypt to finally migrate. > > Please consider your options carefully. I can live with you choosing to > keep the test, but I think its smarter to not have these kind of tests > as potential gating for migration of other package unless it's really > worth it and rare.
I don't think that there is any disagreement. Your arguments make perfect sense to me. Thank you for explaining the situation so thoroughly. I rewrote that test such that this kind of situation should not happen in the future, unless any of the packages dpkg, busybox, libc-bin, base-files, base-passwd or debianutils is no longer part of the archive. Version 0.5.1-4 of mmdebstrap that should fix this issue has been uploaded to unstable already. Thanks a lot! cheers, josch
signature.asc
Description: signature