Please unblock package dkms

[ Reason ]
Lots of bugfixes, especially a behavioral regression from bullseye
(inconsistent success/failure depending of the installation order of
foo-dkms and linux-headers-something).

[ Impact ]
Currently there are some unneccessary dkms-related upgrade failures.
(This does not intend to fix kernel postinst aka dkms failing due to
(third-party) dkms modules incompatible with newer kernels. That is not
a regression. All *-dkms packages in bookworm are compatible with the
kernels in bookworm and make use of Testsuite: autopkgtest-pkg-dkms)

[ Tests ]
Most of the patches are cherry-picked from upstream or waiting to be
merged upstream.
Most fixes also have tests added to the upstream testsuite, stripped
while cherry-picking since they depend on additional testing
infrastructure improvements.
For improving the autopkgtest coverage of dkms in Debian CI see
dkms 3.0.10-9 in experimental and #1033900.

[ Risks ]
dkms has been neglected for too long ...
It's quite possible that further improved testing will uncover more bugs
that have been lingering around. Since I've dug into codebase now, I
should be able attempt fixing more ...

[ Checklist ]
  [x] all changes are documented in the d/changelog
  [x] I reviewed all changes and I approve them
  [x] attach debdiff against the package in testing

[ Other info ]
Once all my fixes (and most of the other patches we carry) got merged
upstream, I'll try to get a new upstream release and upload that s.t.
we get even more fixes and the improved testsuite.

unblock dkms/3.0.10-8

+dkms (3.0.10-8) unstable; urgency=medium
+
+  * Disable tests broken by fix-builtin-archive-dkms-coinstallation.patch.
+    (Closes: #1033396)
+  * Run autopkgtest for all architectures using the headers from the
+    linux-headers-generic (virtual) meta-package.
That virtual package does not exist on armel, I'll add an alternative
test dependency for armel in the next upload.

+  * Generate dh_dkms.1 at build time.
+  * Clean up /boot/*.old-dkms on upgrades from bullseye.  (Closes: #717584)
+  * Clean up obsolete /etc/dkms/sign_helper.sh on upgrades. (Closes: #1019563)
+  * dkms-autopkgtest: Try to 'dkms install' its BUILD_DEPENDS before testing a
+    module. (Single level only, without recursion.)
This fixes an autokgtest regression in src:nvidia-cuda-samples (on
behalf of src:nvidia-cuda-toolkit which cannot have autopkgtests on its
own due to the size of the source package).
nvidia-fs-dkms is the only package in the archive using BUILD_DEPENDS in
its dkms.conf and I only recently fixed its support for the bookworm kernel.

+  * Fix fallback value for BUILD_MODULE_NAME[0].  (Closes: #651973)
+  * dkms.8: Clarify 'dkms add' arguments.  (Closes: #704917)
+  * Print a warning if /proc is not mounted.  (Closes: #810665, #968074)
+  * dkms_autoinstaller: Skip autoinstall if headers are missing (like
+    dkms_common.postinst).  (Closes: #1030595)
+  * Use return code 77 if skipping due to BUILD_EXCLUSIVE_*.
This caused some regression in Ubuntu since some script (not present in
Debian) expected to old return code.

+  * Sort kernels by version.
+
+ -- Andreas Beckmann <anbe@debian.org>  Fri, 31 Mar 2023 14:01:26 +0200
+
+dkms (3.0.10-7) unstable; urgency=medium
+
+  [ Dimitri John Ledkov ]
+  * Fix release regression preventing co-installing in-archive dkms
+    packages with kernels that built-in identical or newer versions of the
+    same dkms package. LP: #2008269
+
+ -- Gianfranco Costamagna <locutusofborg@debian.org>  Wed, 22 Mar 2023 00:02:08 +0100


Andreas

unblocked

Cheers