On 2025-11-09 10.16, Paul Gevers wrote: > Control: block -1 by 1052119 > > Hi Lee, > > On Tue, 13 May 2025 Lee Garrett <[email protected]> wrote: >> While I believe it's a bug in python3-apt or ansible-core, I think we >> should be >> more cautious about setting the pin priority, as this might also >> affect other >> software doing similar tests. AFAICS this pinning shouldn't be needed, >> as the >> package being tested should always have a higher version number than >> the one in >> the apt archive. >> >> So my questions are: Is this really needed? And if yes, could we >> instead do a >> narrower apt pinning at runtime to the specific packages in question? > > Over the years, we've tried multiple things to improve how we're > ensuring that the right packages are tested. We have ideas how to > further improve things, but it's extremely hard to do, because there are > so many things to consider. See e.g. bug #1052119 as an example where we > had to revert an improvement related to pinning, because it didn't play > nicely (yet).
As Paul mentioned, the logic around the pinning that autopkgtest sets is complex and delicate. At first glance "the package being tested should always have a higher version number", but in practice there are many other things to consider. Specifically, the pin you found problematic ensures that the locally built version of a package is the one that will be tested. This allows users to work on autopkgtests for a package without bumping the package version. See also: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1086701 https://salsa.debian.org/ci-team/autopkgtest/-/merge_requests/498/ Cheers, Paride
