Bug#843761: invoke-rc.d: Kill 600 birds with one stone (a.k.a. automatic policy-rc.d for init-less chroots)

[Felipe Sateler]

On Thu, 10 Nov 2016 09:15:40 +0100 Martin Pitt <mpitt at debian.org> wrote:
> Hello Andreas,
>
> Andreas Henriksson [2016-11-09 11:08 +0100]:
> > As discussed on IRC we seem to agree that an init-less chroot which does
> > not have a policy-rc.d blocking service actions isn't a sane
> > configuration. This patch auto-detects the situation and skips running
> > the invoke-rc.d action (aka policy-rc.d code 101), unless --force was
> > given. In both situations a warning message is (also) printed.
>
> I must absolutely and loudly protest against mass-killing birds! ☺
>
> However, I do like the patch, it would give us a much saner behaviour
> of package installation in self-created chroots. mk-sbuild and friends
> do install a policy-rc.d already, but I've seen this come up more than
> once already.
>
> In the past where SysV and /etc/init.d/ were "the thing" it could
> still be argued that one doesn't need an explicit "init system" for
> some situations, but with Debian supporting multiple ones (and systemd
> by default) this is entirely moot IMHO.
>
> Michael, any others: Do you see any downside of this?

I also like this patch. invoke-rc.d is a maintainer interface for
starting/stopping services at package installation time (plus a few
other situations like hook scripts). In that context, it makes no
sense for a maintainer script to start a script in an init-less
chroot. If one by any chance actually needs to do so, then it should
not use invoke-rc.d.

Michael, from the discussion on IRC I did not end up entirely clear if
you are OK with this change or not. Could you please comment, if we
can apply this patch?

Saludos