linking revisions of a series

[Andy Doan]

On 03/16/2017 10:04 PM, Sean Farley wrote:
> Andy Doan <andy.doan at linaro.org> writes:
>
>> On 03/16/2017 05:12 PM, Sean Farley wrote:
>>> Andy Doan <andy.doan at linaro.org> writes:
>>>
>>>> On 03/09/2017 07:46 PM, Florian Fainelli wrote:
>>>>> On 03/08/2017 05:38 AM, Andy Doan wrote:
>>>>>> I was wondering if anyone had interest in adding support to allow
>>>>>> patchwork to somehow track revisions of a series? Basically, I'm needing
>>>>>> to track as series from start to finish and a Series object in patchwork
>>>>>> currently has no links to related series. I have the following rough
>>>>>> idea in mind and wanted to get some feedback before getting too deep
>>>>>> into this.
>>>>>>
>>>>>>  if new series created
>>>>>>      get list of all series with matching project and submitter
>>>>>>      if the subject of one is similar to the new series:
>>>>>>           add some marking (foreign key?) to the new series.
>>>>>>
>>>>>> We'd then have a link-list from start to finish. Is this crazy? Has
>>>>>> anyone tried something like this already?
>>>>>
>>>>> I like the idea, and there seems to be at least two ways this could be
>>>>> helpful:
>>>>>
>>>>> - statistics on the patch series (how many, time between them, etc. etc.)
>>>>> - automatic superseding of patches?
>>>>
>>>> I would very much like to enable automatic superseding. I actually have
>>>> some out-of-tree patches for things like that which we use inside
>>>> Linaro, and I think its a nice benefit. As described above, I do think
>>>> there will need to be some amount of fuzzy-matching involved in case the
>>>> author makes a slight change to the series title. However, python has a
>>>> really handy "difflib" library that can show if 2 different titles are
>>>> X% similar.
>>>
>>> I was planning doing exactly this (perhaps even using difflib for the
>>> patch content). Are your patches somewhere for this?
>>
>> I have no patches for this specifically, just a rough idea. I'd be happy
>> to help/review/test anything you are doing in this area.
>
> Ah, cool! My idea was as follows:
>
> 1) create a migration to add patch.next and patch.previous
> 2) runn the difflib on each series for the incoming patches
>
> (2) worries me because of O(n^2). Any thoughts for that?

That's my rough thinking. I was hoping to make singly linked list, but I 
think you need both next and previous. Its not exactly a cheap 
operation, but you can greatly limit the number of comparisons by only 
looking at the series objects that match the author and project.