On Aug 17, 2017, at 6:32 PM, John Rose <john.r.r...@oracle.com> wrote: > > Better to make the buffering required for > that reversal explicit to the user, IMO. I know it's a case of > brief notation vs. "picky" explicitness; in this case the pickiness > wins because there is an extra O(N) copy hiding in the pretty > brevity.
One more thought: These concerns could be balanced by giving Stream a reverse() operation. Then reverse().reduce() with a suitable lambda would give you foldRight. The reversed stream would be in sequential mode, of course. Unless the stream is already pointed at a reversible data source, the user will know that it is an expensive operation, requiring the original stream to be run forward into a buffer. So: s .foldLeft(op) == s .sequential() .reduce(op) (Isn't the .sequential() guaranteed to be a no-op if the source is a list or array and there is no parallel transform??) And: s .foldRight(op) == s .reverse() .reduce((x,y)->op.apply(y,x)) To me that seems to hit the right balance between convenience and clarity of performance model.
