Hi everyone,

As you might know phase 1 of Wikidata (interwiki links) is live at a lot of Wikipedia's and soon to be turned on for all Wikipedia's. Phase 2 is next, that's basically about infobox data. We are going to need a lot of clever bots to fill Wikidata. To make that possible Pywikipedia should (properly) implement Wikidata. That way bot authors don't have to worry or care about the inner workings of the Wikidata api, they just talk to the framework. At the moment trunk has a first implementation that isn't very clean and in the rewrite it's still missing.

Legoktm and I talked about this on irc. We need to have a proper data model in Pywikipedia. Based on https://meta.wikimedia.org/wiki/Wikidata/Notes/Data_model_primer :
* WikibasePage is a subclass of Page and has some basic shared functions for labels, descriptions and aliases
* ItemPage is a subclass of WikibasePage with some item specific functions like claims and sitelinks (example https://www.wikidata.org/wiki/Q256638)
* PropertyPage is a subclass of WikibasePage with some property specific functions for the datatype (example https://www.wikidata.org/wiki/Property:P22)
* QueryPage is a subclass of WikibasePage for the future query type
* Claim is a subclass of object for claims. Simplified: It's a property (P22, father) attached to an item (Q256638, the princes) linking to another item (Q380949, Willem IV)

You can get these pages like a normal page (site object + title), but you probably also want to get them based on a Wikipedia page. For that there is https://www.wikidata.org/wiki/Special:ItemByTitle/enwiki/Princess%20Carolina%20of%20Orange-Nassau . We should have a staticmethod itemByPage(Page) in which Page is https://en.wikipedia.org/wiki/Princess_Carolina_of_Orange-Nassau and it will give you the itemPage object for https://www.wikidata.org/wiki/Q256638. Currently in trunk the DataPage object has a constructor where you can give a page object and you'll get the corrosponding dataPage. I don't think that's the way to do it because it violates the data model and will get us in a lot of trouble later on when other sites (like Commons) might implement the Wikibase extension.

A WikibasePage should work the same as a normal page when it comes to fetching data. It should have the initial version (just a title, no content) and once you use a function that needs data (or you force it), it will fetch all the data from Wikibase and caches it.
* For an item the data looks like https://www.wikidata.org/w/api.php?action=wbgetentities&ids=Q256638&format=json
* For a property the data looks like https://www.wikidata.org/w/api.php?action=wbgetentities&ids=P22&format=json
Parts of the data (description, aliases and labels) should be processed in the get function of WikibasePage, other parts in ItemPage /PropertyPage

Based on the api we should probably have some generators:
* One or more generator that uses wbgetentities to (pre-)fetch objects
* A search generator that uses wbsearchentities

WikibasePage:
* Set/add/delete label (@property?)
* Set/add/delete description (@property?)
* Set/add/delete alias (@property?)

ItemPage
* Set/add/delete sitelink (@property?)

Claim logic

Not sure how we can use wbeditentity and wblinktitles

We took some notes on https://www.mediawiki.org/wiki/Manual:Pywikipediabot/Wikidata/Rewrite_proposal .

What do you think? Is this the right direction? Feedback is appreciated.

Maarten


_______________________________________________
Pywikipedia-l mailing list
Pywikipedia-l@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/pywikipedia-l