Hi Andreas, On 07-05-2020 20:07, Andreas Beckmann wrote: > *-dkms autopkgtests are currently quite useless without some linux-headers-* > installed to build against (#945594). I think this is solved best in > autodep8 which "defines" the environment for the test. > Even with kernel headers installed, the test may fail early (while > installing test dependencies) if dkms autoinstall fails and not give a > helpful failure log. I therefore sent two patches for dkms (#959910) that > allow disabling autoinstall mode by creating a flag file and then do the > autoinstall step-by-step with verbose error reporting on failure. This > will be done for all installed kernel headers, independent of the > running kernel, s.t. this should work even in a plain chroot. > > This will only work if autoinstall is disabled early enough, therefore > my patch drops the "Depends: @" from the generated test, the > dkms-autopkgtest script will install the *-dkms packages anyway.
We have merge requests pending that also do that: https://salsa.debian.org/ci-team/autodep8/-/merge_requests/21 and some of the discussion in this bug also happened there. > I'm a but unsure about the restrictions of the test: > * isolation-machine should not be needed any more, since the test script > is now independent of the running kernel That's https://salsa.debian.org/ci-team/autodep8/-/merge_requests/20 > * breaks-testbed could be needed, since the test script (which > needs-root) installs/removes packages and 'changes configuration > files' Yes, please. > * skippable is currently not used, but something using it could be > implemented in dkms's dkms-autopkgtest script at some point > * superficial because this only tests whether the module can be built > I tested these changes in dkms and autodep8 on bbswitch-dkms in Debian, > but I have no idea which impact they might have on Ubuntu. So, it's best to have them join in. > At least the list of kernel header packages would need to be different. > Ideally the list of kernel header packages should not be hardcoded but > generated at runtime, but that would have to happen for the correct > release, ... I don't get your last remark. Please also note that the current version of autodep8 is used to cover all suites, so it indeed needs to be generic. Paul
signature.asc
Description: OpenPGP digital signature
