To me, aggregate is implemented with slightly the wrong abstraction.
I hope someone can correct me if I misunderstand something here.
But aggregate takes two functions
seqop: (B, A) ⇒ B and combop: (B, B) ⇒ B
whereas it should take
unaryop:A=>B and combop: (B, B) ⇒ B
and then should internally compute seqop as
(b:B,a:A)=>combop(b,unaryop(a)).
I believe it makes the assumption that the user’s given combop contains the same code as combop, but doesn’t have anyway to verify that invariant.
Perhaps (I’m guessing) it is (was) defined as it is/was, because unaryop is very often the identity function, and letting the user specify seqop allows a call to the identity function to be optimized away.