with recent work, we now optimize this into: