unattended-upgrades 2.9.1+nmu2ubuntu1 failing autopkgtest on arm64

[Dave Jones]

u-u is currently failing autopkgtest on arm64. It does not look like this is a new failure (2.8ubuntu1 appears to have had exactly the same failure) but it's due to the kernel-patterns test assuming that kernel flavors cannot be a suffix of each other. This is true under Debian, and under several architectures on Ubuntu, but the arm64 architecture on Ubuntu has -generic and -generic-64k flavors which cause the test to fail.

The attached debdiff corrects the test for this case.

One may wonder why the test currently passes under armhf, which has -generic and -generic-lpae flavors (I certainly did!). It turns out this is only because Ubuntu autopkgtest runs armhf in a container so the kernel release reported by "uname -r" (used in the test) does not meaningfully match anything in the apt cache.

Given this, the attached debdiff also restricts the kernel-patterns test to isolation-machine as the test assumes a meaningful link between the running kernel and the packages in the archive which is only true in a VM or on the bare metal.

[Steve Langasek (vorlon)]

Have you forwarded this patch to Debian? Debian doesn't *currently* have any flavors that are a substring of another, but that is not a policy in Debian, so the patch is also applicable there.

[Dave Jones (waveform)]

It looks like Debian independently fixed the running_escape_regexp line in January (in 2.9.1+nmu3 [1]). The d/t/control fix isn't *strictly* necessary to fix this and is more a matter of fixing semantics -- but I'll try and forward that change if I can figure out where to submit a PR (doesn't appear u-u is in salsa, and the GH Vcs-Git repo is way behind where it's meant to be).

[1]: https://tracker.debian.org/news/1405664/accepted-unattended-upgrades-291nmu3-source-into-unstable/