Comment 16 for bug 1996950

Revision history for this message
Julian Andres Klode (juliank) wrote :

So I want to say verification failed on bionic because arm64 failed grub-install due to missing efibootmgr, but it's not actually a regression. Otherwise autopkgtests passed everywhere:

sqlite> select package, arch, release, triggers, exitcode from test, result where test_id==test.id AND triggers like "%grub2-unsigned/2.06-2ubuntu14%" AND triggers like "%grub2-signed%" AND requester="juliank" ORDER BY release;
package arch release triggers exitcode
------- ----- ------- ---------------------------------------------------------- --------
gzip amd64 bionic grub2-signed/1.187.2~18.04.1 grub2-unsigned/2.06-2ubuntu14 0
gzip arm64 bionic grub2-signed/1.187.2~18.04.1 grub2-unsigned/2.06-2ubuntu14 12
hello arm64 focal grub2-signed/1.187.2~20.04.2 grub2-unsigned/2.06-2ubuntu14 0
hello amd64 focal grub2-signed/1.187.2~20.04.2 grub2-unsigned/2.06-2ubuntu14 0
hello amd64 jammy grub2-signed/1.187.2 grub2-unsigned/2.06-2ubuntu14 0
hello arm64 jammy grub2-signed/1.187.2 grub2-unsigned/2.06-2ubuntu14 0
hello arm64 kinetic grub2-signed/1.187.2 grub2-unsigned/2.06-2ubuntu14 0
hello amd64 kinetic grub2-signed/1.187.2 grub2-unsigned/2.06-2ubuntu14 0

To fix grub-install needing efibootmgr on bionic, we could add a Depends: efibootmgr [amd64 arm64] to grub2-common there. This is not entirely accurate of course because it's only needed if you have -efi bits installed; but: this allows grub2-unsigned and grub2-signed to remain identical between releases which severely reduces workload for updates.

Though to be fair, we can also add a Depends: grub2-common (>= 2.02+dfsg1-15) | efibootmgr to grub2-signed in all releases. Still missing Depends in grub2-unsigned. We probably should not block on that though.

It's not clear to me how autopkgtest arm64 images ended up with grub-efi-arm64-signed installed successfully in the first place, peculiar, it should not have built an image that way.