Spinoff from #2610: In the past, when filters were applied, all scorers were forced to score in order. With random access DocIdSets, this is no longer needed. Some Weights (BooleanWeight) unfortunately return different scorers for in-order/out-of-order, leading to incompatible scores between segments.
For now we enforce in-order execution of scorers for FilteredQuery (as we do in 3.x), but once we fix BooleanWeight or have some other good way to produce compatible scores, we can reenable random access. Maybe we should nuke BooleanScorer2... - Robert and Mike have some ideas how to do that :)
Migrated from LUCENE-3532 by Uwe Schindler (@uschindler)