[PATCH setup 00/14] Use libsolv, solve all our problems... (WIP)

[Jon Turney]

On 27/09/2017 21:33, Ken Brown wrote:
> On 9/27/2017 3:14 PM, Jon Turney wrote:
>> On 26/09/2017 17:06, Ken Brown wrote:
>>> On 9/26/2017 10:50 AM, Jon Turney wrote:
>>>> On 15/09/2017 17:53, Ken Brown wrote:
>>>>> On 9/15/2017 11:15 AM, Jon Turney wrote:
>> [...]
>>>> If we select 'curr', then the latest version of all installed 
>>>> packages is selected by the picker and will be shown in the pending 
>>>> view, and gets fed into the solver.
>>>>
>>>> I guess this is technically wrong: really we should ask the solver 
>>>> to do SOLVER_UPDATE | SOLVER_SOLVABLE_ALL, which will come up with a 
>>>> solution which updates all installed packages to the latest possible 
>>>> version, subject to any other constraints which exist.
>>>>
>>>> I suspect there's no difference between these two at the moment, 
>>>> though.
>>>>
>>>> Not sure how to do this properly, though.
>>>
>>> One possibility is that we feed all our information to the solver 
>>> before going to the chooser page.  We could then present the solver's 
>>> initial solution in the pending view as the first thing the user sees.
>>
>> Yes, this sounds about right.
>>
>> Thanks to your work we kind of have a bidirectional conversion between 
>> packagedb pick/desired state and a SolverTransactionList now, so this 
>> should be possible.
>>
>> (Although it should be done when the state of the 'Keep' or 'Current' 
>> control changes - note that changing this setting clears any manual 
>> picks currently. [and there's a mechanism to apply the initial state 
>> of that control initially])
> 
> OK, I'll work on this.  I'm traveling at the moment, but I should be 
> able to do it in a couple weeks.

Thanks, but that was more me thinking out loud.  Since it seems pretty 
straightforward, I've written this change.

I've rebased the topic/libsolv branch, with the following additional 
changes:

* Applied your unrelated change to .gitignore to master
* Dropped a couple patches of mine which weren't supposed to escape yet.
* Squashed your "fix typo" patches into the patch they fix (I'm not sure 
how I managed to make that typo in bootstrap.sh! :))
* Rather than add a parameter to SolvableVersion::depends() to make it 
return obsoletes, I've factored out the key-to-deplist code as a helper 
function, and added SolvableVersion::obsoletes(). (There may be other 
deplist attributes in future, so this seems the sane way to do it.)
* Added handling of Source: lines in setup.ini