On Fri, Oct 29, 2010 at 11:10 PM, Stefan Monnier <monnier@iro.umontreal.ca>
 wrote:

> > type _ t =
> >   | IntLit : int -> int t
> >   | BoolLit : bool -> bool t
> >   | Pair : 'a t * 'b t -> ('a * 'b) t
> >   | App : ('a -> 'b) t * 'a t -> 'b t
> >   | Abs : ('a -> 'b) -> ('a -> 'b) t
>
> > There's something "Haskellish" about this syntax, in the sense that type
> > constructors are portrayed as being like functions.
>
> Indeed IIRC OCaml does not accept "App" as an expression (you have to
> provide arguments to the construct).  Maybe this is a good opportunity
> to lift this restriction.


It was actually the case in Caml Light : each datatype constructor
implicitly declared a constructor function with the same name. I don't
exactly know why this feature was dropped in Objective Caml, but I think I
remember (from a previous discussion) that people weren't sure it was worth
the additional complexity.

Note that, as in Jacques's examples, the constructor function was not
curryfied. (type t = A of bool * int) would generate a function (A : bool *
int -> t). It doesn't help the already tricky confusion between (A of bool *
int) and (A of (bool * int))...
By the way, it is unclear if
  | App : ('a -> 'b) t -> 'a t -> 'b t
would be accepted in Jacques proposal. If not, I think going back to a "of
..." syntax would be wiser.