Skip to content

search: make sequential job send unique results only

Warren Gifford requested to merge rvt/seq-dedupe into main

Created by: rvantonder

I'm close to releasing the "feeling lucky" search, which will opportunistically run queries in sequence under different interpretations. While testing, I realized that we don't guard against sending the same result twice, so this PR adds a deduper to the sequential job. While there might be a reason to have a job not dedupe subsequent job results, I couldn't come up with a good one, and this change implements the intuitive (and now default) behavior that I think makes more sense.

All other current behavior that uses sequential job (i.e., recent "run Zoekt before searcher" is unaffected because we are guaranteed Zoekt/Searcher send unique results due to repo partitioning logic)

Test plan

Updated test to cover this behavior.

Merge request reports

Loading