The beta version of Ocaml 3.12 has a couple of new features that relate to a post Stephen wrote a while back on how to ensure that a function definition is polymorphic. In this follow up post I will describe how one of those new mechanisms is essentially what you want for this purpose and the other is perhaps not due to a subtle interaction with how recursive definitions are type-checked.

Polymorphic type annotations

The first new feature is simply the ability to directly annotate definitions with polymorphic types. Consider

let const : 'a 'b. 'a -> 'b -> 'a =
  fun x y -> x

The function const is explicitly declared to have the polymorphic type 'a -> 'b -> 'a.

As Stephen noted, this is not what is meant by free type variables in type annotations. For example, the following example with seemingly analogous type annotations (but a change in behavior)

let wrong_const (x : 'a) (y : 'b) : 'a = y

is accepted, but assigned a less general type than const, namely,

val wrong_const : 'a -> 'a -> 'a

because the type variables in wrong_const indicate merely unspecified types whereas the universally quantified type variables in const indicate arbitrary types. The former are used to ensure that multiple occurrences of the same type variable must refer to the same type within a particular definition. The latter are used to ensure that a type is considered equal only to itself within a particular definition.

Note that a polymorphic type annotation holds inside the body of a recursive definition as well as outside, allowing what is known as polymorphic recursion, where a recursive call is made at some non-trivial instantiation of the polymorphic type.

type 'a perfect_tree = Leaf of 'a | Node of 'a * ('a * 'a) perfect_tree

let rec flatten : 'a. 'a perfect_tree -> 'a list = function
| Leaf x -> [x]
| Node (x, t) ->
  let pairs = flatten t in
  let xs =
   List.fold_right pairs ~init:[]
    ~f:(fun (x1, x2) xs -> x1 :: x2 :: xs)
  in x :: xs

The recursion in the definition of flatten is polymorphic because the recursive call to flatten is only well-typed if we instantiate its declared polymorphic type to ('a * 'a) perfect_tree -> ('a * 'a) list.

Several examples in Chris Okasaki’s classic book Purely Functional Data Structures involve so-called non-regular datatypes like perfect_tree. In order to define any useful functions on such types, one needs polymorphic recursion. Until 3.12 one had to resort to various tricks involving recursively defined modules or records in order to get Ocaml to accept a polymorphically recursive function definition. In 3.12 we can now express such definitions much more directly.

Explicit type parameters

The second new feature in 3.12 relating to polymorphism is explicit type parameters. For non-recursive definitions, this feature may be used to accomplish the same thing as polymorphic type annotations.

let const' (type a) (type b) (x : a) (y : b) : a = x

yields

val const' : 'a -> 'b -> 'a

If we make a mistake in the definition

let wrong_const' (type a) (type b) (x : a) (y : b) : a = y

the type checker lets us know

Error: This expression has type b but an expression was expected of type a

Contrast this with the wrong_const example in the previous section.

However, this simple understanding of explicit type parameters in terms of polymorphism is, sadly, not quite true when one considers recursive functions. For instance, I was surprised to find that the following function type checks.

let rec f (type a) (x : a) : unit = f 42

The assigned type is

val f : int -> unit

What is going on here is quite subtle. The typing rule for fun (type t) -> E considers t abstract while typing E, but elsewhere it is considered to be just another unifiable type variable. The typing rule for let rec f = E deals with two types for the recursively defined value,

  1. the type bound to f while checking the body E (which is constrained according to how f is used in E), and
  2. the type inferred for its body E.

The last step after checking E is to unify these two types together and generalize the one resulting type to obtain the (possibly polymorphic) type of f used henceforth.

In the example above, the type inferred for the variable f is int -> unit. The body of the recursive definition is

fun (type a) (x : a) -> (f 42 : unit)

The type inferred for this expression is '_a -> unit (the type variable is no longer held abstract since it is outside the scope of the explicit type parameter). These two happly unify and the resulting type is int -> unit.

It seems to me that a simple syntactic trick suffices to force the otherwise faulty polymorphic interpretation of explicit type parameters to hold. Simply push the recursion into the scope of the type parameter by transforming

let rec f (type t) = E

into

let f (type t) = let rec f = E in f

In our case, we transform

let rec f (type a) = fun (x:a) -> (f 42 : unit)

into

let f (type a) =
  let rec f = fun (x:a) -> (f 42 : unit) in
  f

which, as hoped, fails with the message

Error: This expression has type int but an expression was expected of type a

Conclusion

The upshot is that polymorphic type annotations are the preferred way to ensure polymorphism outright. They play well with recursive definitions, even going so far as to support polymorphic recursion.

Explicit type parameters, on the other hand, very nearly allow one to state requirements about polymorphic aspects of a function, but this only works straightforwardly for non-recursive definitions, and requires some small rewriting to work for recursive definitions. Even then, polymorphic recursion is beyond the scope of this mechanism.

To be fair, it seems that explicit type parameters were not primarily intended to indicate polymorphism. Their chief purpose, rather, is to give one a way to refer to the type parameters of a function when defining type components of first-class modules. The polymorphism intuitions are strong, however, and it seemed worthwhile to explore their limits.


Do you want to work in a place that understands why functional programming matters? Then join us.