[Debian-olpc-devel] Bug#769510: sugar-moon-activity: not installable

[David Kalnischkies]

On Fri, Nov 14, 2014 at 11:14:02AM +0100, Cyril Brulebois wrote:
> I can reproduce this in testing:
> | # apt-get install sugar-moon-activity
> …
> | The following packages have unmet dependencies:
> |  sugar-moon-activity : Depends: python-sugar-0.86 but it is not installable or
> |                                 python-sugar-0.88 but it is not going to be installed
> |                        Depends: python-sugar-toolkit-0.86 but it is not installable or
> |                                 python-sugar-toolkit-0.88 but it is not going to be installed
> | E: Unable to correct problems, you have held broken packages.
> 
> but specifying python-sugar-0.88 and python-sugar-toolkit-0.88 on the
> apt-get install command line makes apt find a solution:
> | # apt-get install python-sugar-0.88 python-sugar-toolkit-0.88 sugar-moon-activity
> …
> | 0 upgraded, 410 newly installed, 0 to remove and 0 not upgraded.

With some additional debug options you can see the reason:
$ apt-get install sugar-moon-activity -so Debug::pkgProblemResolver=1 \
  -o Debug::pkgDepCache::Marker=1 -o Debug::pkgDepCache::AutoInstall=1
# ^ the first is responsible for the Mark* lines, the later for the rest,
# which means either could be used to see it, I just like verbosity
[… apt start bla bla …]
  MarkInstall sugar-moon-activity [ amd64 ] < none -> 11-1.1 > ( x11 ) FU=1
  Installing python-sugar-0.88 as Depends of sugar-moon-activity
    MarkInstall python-sugar-0.88 [ amd64 ] < none -> 0.88.0-4 > ( python ) FU=0
[… other dependencies …]
  Installing sugar-session-0.98 as Recommends of sugar-moon-activity
    MarkInstall sugar-session-0.98 [ amd64 ] < none -> 0.98.8-1 > ( x11 ) FU=0
    Installing python-sugar-0.98 as Depends of sugar-session-0.98
      MarkInstall python-sugar-0.98 [ amd64 ] < none -> 0.98.0-1 > ( python ) FU=0
       Removing: python-sugar-0.88
        MarkDelete python-sugar-0.88 [ amd64 ] < none -> 0.88.0-4 > ( python ) FU=0
[… which happens to the other -0.88 packages as well …]
[… now to what was already visible with the resolver debug …]
Broken python-sugar-0.98:amd64 Conflicts on python-sugar [ amd64 ] < none > ( none )
  Conflicts//Breaks against version 0.96.0-1 for python-sugar-0.96 but that is not InstVer, ignoring
  Considering python-sugar-0.88:amd64 2 as a solution to python-sugar-0.98:amd64 11
  Added python-sugar-0.88:amd64 to the remove list
  Conflicts//Breaks against version 0.84.2-4 for python-sugar-0.84 but that is not InstVer, ignoring
  Fixing python-sugar-0.98:amd64 via keep of python-sugar-0.88:amd64
  MarkKeep python-sugar-0.88 [ amd64 ] < none -> 0.88.0-4 > ( python ) FU=0
# ^ keep means keep at current version – so no version in that case
[… apt going boom …]

The last part is apts strange way of saying: I have multiple providers
of python-sugar in the solution, but thanks to conflicts, I am not
allowed to keep them all and have to choose one.

What it does now is deciding to keep the "wrong" provider as
sugar-moon-activity depends explicitly on -0.88, but just recommends the
-0.98 stuff – but I have some sympathy here for it as a lot of stuff
(score 11) depends on the -0.98 toolbag compared to -0.88 (score 2) so
the decision it made is a good one as it prevents breaking lots of stuff
(which here means not installing recommends even if we really should),
which just ultimately prevents the installation as a whole…

So, in closing, I would say this is a bug in packaging as it can't be
that a simple "install" commands makes the system "unusual" (as the
policy puts not installing recommends) and not so much in apt for not
solving it even if a (bad) solution exists.


Best regards

David Kalnischkies
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.alioth.debian.org/pipermail/debian-olpc-devel/attachments/20141114/fa32e6b7/attachment.sig>