Bug#688772: gnome Depends network-manager-gnome
On Thu, Sep 27, 2012 at 03:01:23AM +0200, Michael Biebl wrote:
> Ian Jackson wrote:
> > What you are proposing is a compromise between doing the right thing
> > for our users, and upholding the autonomy of the maintainer.
>
> Changing the Depends to Recommends was never the right solution to the
> real issue, imo. The underlying problem is that:
>
> a/ some users will always be unhappy with the choices we make for those
> meta-packages. Some want A instead of B, some want no C at all, some
> want D being added and such. It is simply impossible to please everyone.
> My simple recommendation for such users is, to juse not use those
> meta-packages then and pick and chose what you want instead.
> You do that *once* during a dist-upgrade and you can take the
> meta-package we provide as input. So this isn't a lot effort.
Hmm. This is a common statement regarding metapackages - I've made it
myself in the past - but my experience is that it isn't really as true
as we'd like to think. If you aren't familiar with the details of a
subsystem, removing its metapackage just for the sake of avoiding a
single troublesome package means that you end up taking on the task of
identifying all that subsystem's essential components in future
upgrades; and if you miss one, your system may well behave oddly after
the next release and it's quite possible that the maintainers won't have
a great deal of sympathy since you brought it upon yourself. It can be
a quick road to a world of pain. If you're a developer, maybe it's not
so hard, but there are people who have good and valid reasons to be
uncomfortable with one choice or another who aren't deeply familiar with
the whole subsystem.
Our practice in Ubuntu is to try to subdivide metapackages into
essential components - those where we simply aren't prepared to call the
system, say, an "Ubuntu desktop" without them - and ones that are merely
strongly recommended, where we can reasonably envisage alternatives.
network-manager has long fallen into the second category, despite its
usefulness; as do many of the default applications, because those are
things many people choose to substitute for good and valid reasons.
Now, I know that Ubuntu isn't Debian and Unity isn't GNOME Shell and all
that, and I'm certainly aware that Ubuntu doesn't get this totally right
either, but I do think there's a strong case for erring on the side of
using Recommends in metapackages where there is contention. It just
makes users' lives easier in so many situations.
> I actually somehow doubt, that there are a lot of squeeze users, which
> have the whole gnome meta-package installed but decided to remove NM.
Unfortunately popcon doesn't have this kind of correlation data.
network-manager is installed on more systems than gnome, but I'm not
sure how much information we can derive from that since it'll have some
users from other environments too. It does seem to be something that's
come up a fair bit anecdotally, though. For me, if I were the
metapackage maintainer, the mere presence of a giant argument about one
of the components of my metapackage would be enough to make me
reconsider, regardless of whether I agreed with the reasoning.
If it were somehow to be established that this is indeed vanishingly
rare, I would probably reluctantly withdraw my objection. We can assert
things at each other until the cows come home, but I don't honestly see
how it can be established clearly. Sometimes it happens that as a
maintainer you paint yourself into a corner one way or another, and
sometimes you just have to live with it because the upgrade experience
matters. It's not unusual for maintainers to create new packages in
order to work around this. (For example, I would have no objection to a
solution whereby the GNOME team created a new metapackage which has
stricter dependencies even on a wider range of things than
network-manager, and made that be part of fresh installations of wheezy;
that would seem quite reasonable and would comfortably sidestep any
upgrade questions.)
> I would really appreciate, if the ctte could leave this case as it is
> now and let us concentrate our efforts on fixing real issues and bugs
> instead of having to spend our time writing several pages long emails
> where we need to defend our work. The lack of trust that was shown
> towards us has definitely saddened me and taken out all the fun and
> enthusiasm I have for Debian.
Well, I understand that the GNOME team are frustrated - it's quite plain
to see - but this goes both ways. I'm saddened by the attempt to find a
solution that at least partially satisfied the letter of the TC's
resolution while going against its spirit, and the apparent lack of
respect for the dispute resolution arrangements that we all surely
implicitly signed up to when we joined Debian. We're trying to make
Debian better too.
I don't think we enjoy having to extend this debate any more than you
do; but really it would have been much more helpful to propose
alternative solutions in advance so that we could discuss them. Did
anyone suggest the option of putting the Depends in gnome during the
previous TC discussion? If so, I certainly didn't see it.
Cheers,
--
Colin Watson [cjwatson@debian.org]
Reply to: