Re: Bug#645713: fails to upgrade a default GNOME desktop installation from squeeze → sid

[David Kalnischkies]

On Tue, Jan 31, 2012 at 02:55, Andreas Beckmann <debian at abeckmann.de> wrote:
> The actions performed by piuparts are:
>
>  bootstrap-minimal-squeeze
>  apt-get install gnome
>  switch-sources-list-to-wheezy
>  apt-get update
>  apt-get -yf dist-upgrade

Next time, the most important part of the reproducing steps shouldn't
be hidden further down in the mail, as it is unreproducible without:

> Note that recommends are not installed by piuparts.

Recommends have to be disabled before installing gnome to see this.
(Which, as we all know, is not the default for releases, yet alone years…
 so i wonder why piuparts still insist on doing (only) non-default tests,
 but that is a completely different story)


>  E: Could not perform immediate configuration on 'libgstreamer0.10-0'.
>  Please see man 5 apt.conf under APT::Immediate-Configure for details.
>  (2)

apt/experimental is able to calculate an ordering solution, but i haven't
tested so far if it is a correct one. Anyway: Looking at the debugging
output of it raises quiet a few eyebrows (unfortunately i have only two)
about the dependency hell constructed here.

I HIGHLY suggest to have a look at the dependencies in the gnome
universe and remove/refactor a few of them.
Removing 'Breaks: gdm3 (<< 3.0.3)' from libglib2.0-0 will e.g. allow
apt/squeeze to dist-upgrade this situation again.
(The reasoning behind breaking a display manager from a near-
 core-level library is above my mental possibility to understand as
 i am not into gnome packaging at all…
 If this really needs to stay, try something else. This was just my
 first wild-guess and it worked - and i am not good at guessing usually)

So, in effect: I am a bit surprised that it works with recommends,
but i have refrained from looking at why so far, maybe on freaky-Friday…
(suspecting a different choice in alternatives)


> All dependencies of gnome in wheezy have passed the piuparts test,
> recursively, and in the case of alternative dependencies at least one
> alternative has passed before the testing 'gnome' was attempted by
> piuparts.

For a greedy algorithm "one out of many possible alternative" might be
already too few. To have a decent chance it needs to be either the
alternative already installed on the system (but in an older version) or
if not installed the first alternative in the or-group.
It might be that APT finds a way if that way is blocked, but greedy
algorithms are usually quiet bad at reverting decisions…


I have cc'ed the gnome-pkg-team on the bug as i can't really decide
which package should get this one. As it is mentioned in the "old" buglog
apt is in general not the ideal package to assign this to as we can fix
the bug (if it exists) for the future, but even I have no timemachine, so
the update needs to work with whatever we had in "the past".


Best regards

David Kalnischkies, who is desperately looking for a blue police box now