Dear debhelper maintaner,

the maintscripts snippets installed by dh_installinit call invoke-rc.d
without --skip-systemd-native, for example to start services in postinst.

This means that, if a package ships both an init-script and a systemd
unit, both will be called, even on systemd systems.

You can find a MR that fixes this issue at

https://salsa.debian.org/debian/debhelper/-/merge_requests/151

Regards,

After a more in-depth inspection, it turned out that the only
maintscripts without `--skip-systemd-native` are those of packages that
use a compat level <= 10.

I modified the linked MR to unconditionally add --skip-systemd-native,
even for packages that use older compat levels.

This is necessary to avoid the bug reported here, now that all packages
in testing have systemd units (thanks to lintian's error
`missing-systemd-service-for-init.d-script` and to the associated
RC-level bugs).

(Alternatively we could declare shipping initscripts in packages with
compat <= 10 a RC bug.)

Regards,

Hi Gioele

Thanks for looking at improving debhelper.

Gioele Barabucci:

Compat 10 was before `dh_installsystemd`, so `dh_installinit` was the
tool (in debhelper) at the time to manage these things. There was an add
on for system `dh_systemd_enable` vs. `dh_systemd_start` on top of that
when dh_installinit.

In other words, the compat level works as it was designed at the time.
Compat 12 was the first compat level to properly support systemd units
(I am deliberately omitting compat 11 since it is actively discourage
due to a weird interaction here). The `--skip-systemd-native` option was
built to support the split.

I am lacking an argument for why this is suddenly a bug in an almost
decade old compat level that was never designed to support this feature.

That all packages has systemd services is not an argument because
packages could have systemd services at that time. Ideally with the
systemd add-on.

This also feels like jumping to conclusions for me (possibly because I
am missing a link you had). At the time, you could have initscripts and
systemd units in the same package on a compat 10 package without it
being an RC bug.

As said, this change would retroactively change a decade old compat
level. While that compat level might have been suboptimal, it worked so
far. I am missing some "very good arguments" for doing a change as older
compat levels are intended to retain feature/bug compatibility for the
features explicitly changed in the later compat level.

Best regards,
Niels