User:Mike Peel/P373

Discussion (part three)

This has become a huge saga, lasting over two years already just with this discussion, and that's not counting the previous discussions that go back to 2013, when the property was created. That clearly demonstrates that this isn't a clear-cut issue, but I think we're iterating towards a sustainable solution.

From my understanding, the key issues and solutions raised in this discussion; their solutions; and my personal experiences from working with this property, are:

• Commons category (P373) duplicates the sitelinks, which we don't do for any other Wikimedia project.
  • It was created as "an intermediate step until we have full Commons support", and it was noted that "it seems that it will take some time to integrate Commons into Wikidata" - we now have that full support (see below)
  • Historically, we only had Commons category (P373) links to Commons, but this has significantly changed in the last few years due to the Wikidata Infobox on Commons, which requires
  • We could do the same for other projects: but sitelinks are automatically updated when pages move, while property text doesn't auto-update
• We have a one-to-one mapping between Wikidata items and Commons pages/categories via the sitelinks, but we need to link from multiple items to the same Commons category.
  • This is solved by essentially having 'topic', 'category' (instance of (P31)=Wikimedia category (Q4167836)) and 'list' (instance of (P31)=Wikimedia list article (Q13406463)) items, and linking between them using topic's main category (P910)/category's main topic (P301)/list related to category (P1753)/category related to list (P1754).
  • We also have more complex links, such as category for ship name (P7782)
  • It's disputed that one-to-one mapping is the best solution here, but it's all we have available to us.
• The sidebar link shows Commons category (P373), and not the sitelinks
  • This has now been fixed by WMDE/WMF via phab:T232927. The sidebar link now uses topic's main category (P910) and category related to list (P1754) to find the link to Commons that should be displayed, rather than just using P373.
• We essentially have Galleries linked to in 'topic' items, and Categories in 'category' items.
  • Not everyone agrees with this, but it works.
• The above still doesn't cover all use cases.
  • Working through enwiki examples: under 10k remaining
  • There will be cases that we can't resolve: that will need articles splitting; links removing; new items creating; new Commons categories creating. The existence of P373 temporarily solves those issues, but we still need to tackle them in the long term.
• We still have Commons gallery (P935).
  • While I've been prioritising the work around categories, but galleries also need to be imported. I think this can be resolved fairly quickly (by importing values to sitelinks, and moving category links to new items) after P373 has been deprecated. Probably resolving this property would have been quicker, but categories felt more important.
• Queries become more complicated
  • This is manageable by using more complex queries - but isn't ideal.
• Many external uses
  • There is no obvious mechanism to alert users of this property across many languages. I've been advertising the updated Lua code that lets wikis access the Commons link through the properties, but there hasn't been much take-up yet.
• We're going to deprecate Commons categories anyway because we now have Structured Data on Commons
  • This could happen over the next few years, but the migration is more complex than expected, so this is taking more time. Meanwhile, Commons categories are the best we have to navigate photos on Commons. Storing the links in a property is good; storing it in sitelinks is better.
• We could just keep this property regardless
  • We need to keep it in sync with the sitelinks
  • This could be done using bot scripts
  • How would we handle user-bot conflicts?

My proposal is still the same as it was two years ago. I propose that we deprecate this property, and remove the values that match the sitelinks (through the various connecting properties). We still have to resolve the more complicated cases, but this change will significantly decrease the duplicated information, and confusion by users about how they should link to Commons.

The steps to remove P373 uses if such a decision is made will be really complex, and this needs to be taken into account with this decision. While deprecating the property provides a transition time (which could last a year or more), this would still cause significant disruption to Wikimedia projects that are using the property. In particular:

• Existing uses would have to be migrated to use the sitelinks (using available Lua code)
• MediaWiki integration should no longer be impacted
• Sister project links may also need to be updated
• And more

If we decide to keep this property, then I think that we would need to:

• Enforce automatic synchronisation between sitelinks and Commons category (P373) values
  • Need to enforce deletions/renaming on Commons with Wikidata
  • This can be complex, but the simplest approach is to sync from sitelinks to P373

What happens if there's no consensus here?

• I don't know.
• Probably see above with keeping this property