apache
FilterWeight should delegate count()
Description
I was talking with a colleague today about a case where we're getting burned by eager computation of Weight#count() (which we can probably resolve by changing the order of our clauses to move the clause most likely to match nothing to the front of the BooleanQuery).
While talking about options to skip the attempted Weight#count() computation, I suggested throwing together a wrapping query that produces a FilterWeight that overrides the count() method to return -1. That's when I noticed that FilterWeight doesn't delegate the count() method and so it does return -1.
I see from https://github.com/apache/lucene/pull/242 that not delegating count was a conscious decision (and is enforced in TestFilterWeight), maybe because it was safer to assume that any FilterWeight might mess with the wrapped Weight's count. On the other hand, we do delegate Weight#matches by default. I think it's safe to say that if FilterWeight#matches behaves the same as the wrapped Weight, then FilterWeight#count must also behave the same as the wrapped Weight. (So, I feel like we should delegate both or delegate neither. I prefer delegating both.)
As a bonus, if we fix this, then we can remove this whole block from TestFilterWeight, since Weight#bulkScorer() is now final.
This makes sense to me, we should delegate count() in FilterWeight then the delegation happens to wrapped weight.
