Re: Bug#681834: network-manager, gnome, Recommends vs Depends [and 3 more messages]

[Ian Jackson <ijackson@chiark.greenend.org.uk>]

Russ Allbery writes ("Re: Bug#681834: network-manager, gnome, Recommends vs Depends"):
> Tollef Fog Heen <tfheen@err.no> writes:
> > Am I understanding you correctly in that your answer is «A Recommends
> > can never be upgraded to a Depends for a metapackage»?
...
> I think it depends on the purpose of the metapackage.  [...]

I agree with Russ's answer.


Tollef Fog Heen writes ("Re: Bug#681834: network-manager, gnome, Recommends vs Depends"):
> ]] Ian Jackson 
> >     (ii) Users should be able to conveniently install and upgrade
> > 	gnome without network-manager.
> 
> When you write «gnome», are you talking about the «gnome» package or
> some other package?  If so, why?  Why require it to be the «gnome»
> package rather than asking them to have a gnome-but-without-nm
> metapackage?  You're talking about the gnome-core package further down.

I meant the GNOME system, not any particular package.

> Or even if you think that's not enough, why dictate how the gnome
> maintainers solve the problem rather than saying «fix so people can do
> A, B, C in a reasonable manner»?

The purpose of the TC is to review the decisions of maintainers,
particularly technical decisions.  That includes decisions on how to
implement some agreed objectives.

If the implementation strategy is not the best one, and the matter is
sufficiently important, then certainly the TC should intervene to
require that the implementation be done in the best way.

> >   6. The Technical Committee overrules the decision of the gnome-core
> >      metapackage maintainer.  The dependency from gnome-core to
> >      network-manager-gnome should be downgraded to Recommends.
> 
> I think this would be unwise.  I think you're risking upsetting the
> gnome maintainers enough that they'll lose motivation and we'll end up
> with a worse maintained gnome in wheezy and onwards; that's at least the
> feeling I have from how the conversation has gone on the lists, etc.

Personally I don't think this should be a relevant factor.  What you
are saying is that even though someone may be wrong, if they are
stubborn and petulant enough we should refrain from overruling them
for fear that they might throw their toys out of the pram.

Not only would that be setting a bad precedent, it is also insulting
to the gnome maintainers.

Kurt Roeckx writes ("Re: Bug#681834: network-manager, gnome, Recommends vs Depends"):
> This whole "NOT" part is very confusing to me.  I think you're
> trying to say here that enable/disabling Recommends globably
> isn't something you care about, just that there is the possibility
> to remove it if desired by the user.

Yes.

Tollef Fog Heen writes ("Re: Bug#681834: network-manager, gnome, Recommends vs Depends"):
> This looks like a much better statement than the previous one, but I'd
> still like to nitpick a little bit.
> 
> >     The gnome-core metapackage is intended to reflect the core of the
> >     GNOME desktop environment: the basic tools and subsystems that
> >     together constitute GNOME.  The gnome metapackage is intended to
> >     reflect the broader desktop environment, including extra components
> >     and applications.
> 
> gnome-core is also the same as the official core modules of the GNOME
> desktop, as listed in the description.
> 
> I'm questioning whether it's appropriate for the CTTE to decide that the
> purpose of a metapackage is something else than what the maintainer
> decides it is.

Changing the dependency from Depends to Recommends would still serve
that purpose.

> >     If matters are left as they currently stand, users who have the gnome
> >     metapackages installed but do not have network-manager installed will,
> >     in the process of upgrading from squeeze to wheezy (either due to an
> >     explicit decision to remove it or an implicit decision to not install
> >     it by disabling automatic installation of Recommends), end up
> >     installing network-manager on systems where it is currently not
> >     installed.  It will also no longer be possible for users to install
> >     GNOME metapackages in wheezy without installing network-manager.
> 
> This is incorrect, you can still install the gnome-session metapackage,
> which is what the GNOME packagers have repeatedly stated that people who
> want a minimal GNOME desktop and finer package selection should
> install.

This misses the point.  These users do not want a "finer package
selection" in general.  They are quite happy to have the whole of the
rest of the GNOME Core (as defined upstream) to be installed - even in
many cases the whole of the GNOME system (since many of these same
users previously requested, and got, the downgrading of the dependency
from the gnome metapackage).

network-manager is special for the reasons Russ gives.

> >     For most applications and components, the only drawback of this would
> >     be some additional disk space usage, since the application, despite
> >     being installed, wouldn't need to be used.  However, network-manager
> >     assumes that, if it is installed, it should attempt to manage the
> >     system's network configuration.  It attempts to avoid overriding local
> >     manual configuration, but it isn't able to detect all cases where the
> >     user is using some other component or system to manage networking.
> 
> This sounds like a regular bug that should just be fixed.  I've seen
> this repeated here and there, but I haven't seen references to
> still-open bugs.

Even if there are no such bugs right now (which I'm not sure I
believe), there are likely to be such bugs in the future.

It is simply the wrong technical answer to install network-manager
when it is not needed.  It is needlessly exposing the user to risk of
bugs and demanding that they disable it specially.

> >     1. The package takes action automatically because it is installed,
> >        rather than being a component that can either be run or not at the
> >        user's choice.
> 
> Like most all daemons in Debian? :-)

Yes.  This is not a criticism of the network-manager package.  It is
simply a criticism of the idea that it should be installed when it is
not needed.

If we were talking about any other daemon being pulled in with a hard
Depends when it was not strictly needed I would have the same opinion.

> >     3. There is both demonstrable, intentional widespread replacement of
> >        that package by Debian GNOME users and no significant loss of
> >        unrelated GNOME desktop functionality by replacing it with a
> >        different component.
> 
> So if they cripple the GNOME apps to actually require NM, the dependency
> would be fine?  This sounds like a perverse incentive.

That assumes that the GNOME upstream are more interested in forcing NM
down the throats of unwilling users than they are in making software
work.

It also assumes that such behaviour in GNOME wouldn't be considered a
release-critical bug by Debian.  That is a very big assumption.

> I think you should take into consideration that the GNOME maintainers
> have offered interested parties the option of maintaining a
> gnome-core-without-nm package in the pkg-gnome repository.  Somebody who
> cares would have to do that work, though, and I did not see anybody
> stepping forward to do so, so perhaps the demand isn't actually that
> large?

A separate metapackage does not properly fix the problem because it
gets the upgrade from squeeze wrong.

Ian.