Isn't there at least some technical basis for this (less-than-ideal interactions with type inference IIRC)?