On 24/12/2017 15:00, Ken Brown wrote:
> On 12/13/2017 5:31 PM, Ken Brown wrote:
>> On 12/13/2017 1:05 PM, Achim Gratz wrote:
>>> Ken Brown writes:
>>>> 1. Uninstall A.
>>>> 2. Don't uninstall B.
>>>>
>>>> On the surface, it would seem that libsolv chose 2 by default, because
>>>> it returned an empty transaction list.  This was reflected in the log
>>>> and was also clear when I selected 'Back'.

Yeah, I think what is actually happening here is that the solver returns 
a partial solution, without the problematic transaction.

But yeah, there's no real concept of a default solution, so (lacking a 
UI to choose, which I think is a bit out of scope for the moment), it's 
up to us to define one.

>>> I don't think there is a default in this case.  I also see in zypper
>>> that the order of the proposed solutions (there can be way more than two
>>> if the dependencies are more complicated) is not always the same, so
>>> there is no preference implied by the order as well.
>>>
>>>> Maybe we have to deal with this situation ourselves.  Whenever a
>>>> problem involves a missing dependency, we could choose as default
>>>> solution the one that installs/keeps the dependent package, as is
>>>> currently done.
>>>
>>> That solution unfortunately isn't always the one that causes the least
>>> amount of transactions or even the least amount of breakage.
>>
>> That may be true, but I still think it's a reasonable default.  The 
>> user doesn't have to accept it.  Also, it's consistent with what setup 
>> currently does, so it won't surprise anyone.
>>
>> The attached patch attempts to implement my suggestion.

I came up with a slightly different solution of just picking the first 
solution as a default.

After solving problems we also need to consider the 'install source for 
everything I install' flag, which unfortunately requires quite a bit of 
refactoring.

See attached.