Github user dbtsai commented on the issue: https://github.com/apache/spark/pull/23171 @rxin `switch` in Java is still significantly faster than hash set even without boxing / unboxing problems when the number of elements are small. We were thinking about to have two implementations in `InSet`, and pick up `switch` if the number of elements are small, or otherwise pick up hash set one. But this is the same complexity as having two implements in `In` as this PR. @cloud-fan do you suggest to create an `OptimizeIn` which has `switch` and hash set implementations based on the length of the elements and remove `InSet`? Basically, what we were thinking above.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org