Are f-expressions needed in normal-order evaluation models or only in applicative-order evaluation models? From SICP, it seems that applicative-order (evaluate the arguments and then apply) is what necessitates the need for special forms.