Bug#727708: On diversity

[Tollef Fog Heen <tfheen@err.no>]

]] Steve Langasek 

> GNOME certainly uses these interfaces already.  Whether they should be
> considered a "dependency" or not is probably something that should be left
> to the maintainers' discretion.  But I think they should certainly be
> handled the same way as logind, generally - with a dependency on some
> virtual package that other implementations could provide (or that can be
> provided by a binary package from systemd source that doesn't depend on pid
> 1 - since these dbus services all work fine on non-systemd systems, and
> there's no technical reason that should ever change given the function of
> the services in question).

I'm willing to look at adding virtual packages for those once we
actually see alternative implementations happening.  Adding them because
somebody might, maybe, possibly add them somewhere down the line sounds
premature.

As for whether they should work with non-pid1 or not, it's also a
question about what we (the systemd maintainers) want to support.  The
daemons currently don't require systemd as pid1, but given all the flak
over maybe making logind require systemd as pid1 there is a very strong
incentive to not make those implementations work with another init.  The
CTTE here and in the past (see NM) views regressions as much more
serious than lacking implementations.

I think this is pretty sad and gives maintainers perverse incentives,
since there's not really any graceful way to say «this will no longer
be/is no longer supported» without risking being struck down.  It's a
somewhat separate discussion from the whole «what should be the default
init system» discussion, but it's one we (as a project) should be having
at some point.

-- 
Tollef Fog Heen
UNIX is user friendly, it's just picky about who its friends are