[Aptitude-devel] Bug#668515: aptitude resolver is nuts

[Daniel Hartwig]

On 16 April 2012 21:41, Michal Suchanek <michal.suchanek at ruk.cuni.cz> wrote:
>> If you skip through the available solutions is this eventually in the
>> list?  Hinting the resolver may get you there faster.
>
> No, this can be resolved only manually afaict.
>
> I have no idea how to hint the resolver.

This is when you accept or reject certain suggestions, or provide your
own suggestion to install or remove a particular package.

>> > The following actions will resolve these dependencies:
>> >
>> >      Remove the following packages:
>> [wants to remove most (all?) foreign-arch (i386) packages and some
>> entangled native-arch]
>>
>> The current resolver does not handle multi-arch at all.  I'd say that
>> is the cause of the problem given that it is trying to remove most of
>> the foreign-arch packages.
>
> I had issues like this even when multiarch is not involved.

Yes, though reports coming in suggest that this is much more common with
m-a.

>
> What does it mean that resolver does not handle multiarch packages?
>
> Is it hardcoded to remove all foreign arch packages or what?
>
> The i386 packages are just packages.

It is not that aptitude treats any packages different, the resolver
just gets stuck by the added complexity of m-a situations.  This
usually leads to removing all packages for one arch or the other and
no better option even when much better options do exist.

Other APT frontends seem to work better.  They mainly rely on
libapt to handle conflicts whereas aptitude tries to do this itself.