Cross-posting to -release as they might be interested. The original thread starts at https://lists.debian.org/debian-qa/2015/04/msg00083.html but there is not much more context context there besides me pointing to the test suite for the current rules that are used by debci to "blame" packages for failures in other packages: http://sources.debian.net/src/debci/0.12/test/test_blame.sh/ On Wed, Apr 29, 2015 at 07:04:19PM +0200, Francesco Poli wrote: > On Tue, 28 Apr 2015 19:53:13 -0300 Antonio Terceiro wrote: > > > Hi, > > Hello Antonio! :-) > > > > > On Mon, Apr 27, 2015 at 08:49:52PM +0100, Joe Mullally wrote: > > > http://ci.debian.net and autopkgtest are awesome. > > > > > > Has anyone thought of writing something like apt-listbugs that checks > > > package upgrades/new installs against the CI data recently? > > > > > > Seems like an obvious and easy win towards getting a safely upgradable > > > "rolling release" debian version based on CI that has been discussed > > > before. > > > > That is interesting, although the original idea was to do that checking > > on the archive side i.e. preventing packages that fail their CI (or that > > make reverse dependencies fail theirs) from migrating to testing. > > > > Such a client-side tool would however be useful anyway, at least for > > people running unstable. apt-listbugs has some logic to pin buggy > > packages if the users says so, and to remove the pinning when the > > corresponding bugs get fixed; it would be nice to not duplicate that. > > I'm CC:ing apt-listbugs maintainer(s?) for input. > > This is somewhat interesting, even though I would have to ask a few > questions about ci.debian.org in order to assess the feasibility of the > idea. > But now I am in a hurry... :p > > First thing off the top of my head: what about implementing an > automatic RC-bug-report filing mechanism in ci.debian.net? > When a package is to be blamed for some CI test failure, ci.debian.net > would automatically file a RC bug report against the package on the > Debian BTS, with appropriate version info and content. > This would prevent the package from migrating from unstable to testing > and would make the test failure visible for apt-listbugs users, at the > same time! > > What do you think about this possible strategy? I am not very confortable with bugs being filed automatically, let alone RC ones. But since the RT had the idea of automatically blocking migration of packages that are correlated¹ with CI failures automatically anyway², maybe that is not an issue. ¹ remember correlation does not imply causation. ² http://lists.debian.org/20131226170803.CA9B1100618@thykier.net Something I can't see very clearly yet is what would be easier/require less effort for long-term maintainance: such a bug filing robot from the CI side, or changing britney to consume CI data. One practical difference with using RC bugs is that maintainers would have the option of manipulating bug severity at their own discretion, while if it was implemented on britney only the RT would be able decide whether a package may be allowed to migrate despite being correlated with CI failures. -- Antonio Terceiro <terceiro@debian.org>
Attachment:
signature.asc
Description: Digital signature