Last modified: 2013-10-10 12:39:27 UTC
CirrusSearch doesn't index pages without any text. I made a page who's contents was: = '''zing!''' = and CirrusSearch refused to pick it up.
Well, technically this isn't any text that isn't in a heading. I'm not sure exactly what is up but this is quite reproducible. Triaging to high just in case someone actually wants to search for pages without text in them. I dunno why they would but wikis have lots of workflows I'm not aware of.
For posterity: this is reproducible on test2wiki but not on my development machine.
I think I've figured this one out. SearchUpdate::doUpdate tries to fetch the page and if it doesn't get it then it decides that the update is in fact a delete. OTOH I mostly see not fetching things from the database that I expect to be there as a symptom of MySQL slave lag.
SearchUpdate shouldn't care what sort of content we have, as long as it can fetch the page object.
So, it doesn't care, but I happened to be testing with pages that just had a title. I'm pretty sure that is a red herring - all pages seem to do it. I just copied https://test2.wikipedia.org/wiki/Rashidun_Caliphate from enwiki and it did it.
Fix: https://gerrit.wikimedia.org/r/#/c/86704/
Merged.
Verified fixed on test2wiki.