Matching via individual languages might create pretty big query though... We could match against labels_all, but that would not work for descriptions.

Certainly matching an _all field is the only thing reasonably performant here. We could create a descriptions_all if that's needed.

Change 514415 had a related patch set uploaded (by Smalyshev; owner: Smalyshev):
[mediawiki/extensions/WikibaseCirrusSearch@master] Implement haslabel:*

https://gerrit.wikimedia.org/r/514415

Change 514415 merged by jenkins-bot:
[mediawiki/extensions/WikibaseCirrusSearch@master] Implement haslabel:*

https://gerrit.wikimedia.org/r/514415

Interestingly enough, it seems to work on Wikidata but not Commons. I wonder why?

I have no clue why it's not working yet but I think youwe should have indexed the captions to description fields not label fields. Label fields are optimized for exact matches which is not suited for captions.

Right now from what I understand we're indexing them as label fields since they are recorded as labels. Should this change? Data model lists MediaInfo as having both labels and descriptions, though I am not sure what this means.

@Cparle could you shed some light on this?

Yeah, I think they're being indexed as label fields, and also written into opening_text. See https://commons.wikimedia.org/wiki/File:Mute_swan_(Cygnus_olor)_in_Bowness-on-Windermere,_England.jpg?action=cirrusDump

Created T226722 to fix this.

Change 519514 had a related patch set uploaded (by Smalyshev; owner: Smalyshev):
[mediawiki/extensions/WikibaseCirrusSearch@master] Fix haslabel:* to use labels_all.plain field

https://gerrit.wikimedia.org/r/519514

Change 519514 merged by jenkins-bot:
[mediawiki/extensions/WikibaseCirrusSearch@master] Fix haslabel:* to use labels_all.plain field

https://gerrit.wikimedia.org/r/519514

Works on test.wikidata, needs to be verified on main production after T227136: Reindexing search index wikidatawiki for eqiad fails is fixed.