Bug#727708: Please clarify L options with regard to interfaces
Sam Hartman writes ("Bug#727708: Please clarify L options with regard to interfaces"):
> * Colin said that it would be OK to depend on a stable interface such as
> logind-208 provided that multiple implementations could exist.
Colin, I think you need to clarify this. I think it matters very much
whether multiple implementations _do_ exist.
> * Ian said that this dependency would not be OK.
> I'd like the ballot options to clarify:
> 1) Whether these interface dependencies are acceptable
I don't have an opinion on the technical implementation details such
> 2) Whether they are acceptable in cases where there is only one
My view on that is "no". The key question for me is whether it is
actually possible to use a different init system.
> I'd request the TC consider the following question although I'm not sure
> going into this level of detail on the ballot is appropriate:
The TC ballot texts at the moment talk about "require", not about
dependencies. So it doesn't matter whether a requirement is declared
as a dependency, and if so exactly what the form of that dependency
is. Likewise it doesn't matter whether the dependency (if there is
one) is direct or indirect.
> 3) If we are using virtual packages to define interfaces, what should
> the dependency look like? Would you want a raw virtual dependency such
> as gnome-shell depends on logind-208? If so, isn't that kind of not how
> we currently recommend things? Or a concrete dependency like
> systemd|logind-208? If so, please make sure that if such interface
> dependencies are permitted your policy text actually permits the
I don't think the exact form of the dependency matters very much. I'm
not sure I understand why you think the difference here is important.