The new query builder we plan to use as a replacement for QueryString seems to expose bad tokenization behaviors directly to the users. It does not seem wise to go ahead on these languages without investigating better analysis options for such languages (see T151743 for more details).
In order to move forward and continue to activate BM25 on more wikis we decided to go ahead on languages that use spaces.
We will use the new feature offered by https://gerrit.wikimedia.org/r/#/c/319253/ which allows us to use a language tag.
This task is for tacking the reindex process:
- activate BM25 index time config for wikis that use spaces
- reindex codfw
- activate query time options and switch traffic to codfw (except completion)
- reindex eqiad
- switch traffic back to eqiad