Overloading =========== Simple overloading ------------------ Let's start with an example: let plus = generic (+) | (+.) will define an overloaded function plus. This works by overloading functions for integers and floats, using (+) and (+.): # plus 1 2;; - : int = 3 # plus 2.3 4.5;; - : float = 6.8 Derived overloading ------------------- Using already defined overloaded values, you can define other extensional polymorphic values with the normal let: # let double x = plus x x;; val double : { 'a -> 'a -> 'a < [| int -> int -> int | float -> float -> float |] } => 'a -> 'a = This double is also extensional polymorphic: # double 1;; - : int = 2 # double 1.2;; - : float = 2.4 The type of double tells you that it is basically a polymorphic function of the type 'a -> 'a, but its instantiation is actually constrained, where is explained inside {...} part. It saids that the instantiation is permitted only when 'a -> 'a -> 'a becomes a proper instance of the type of plus, [| int -> int -> int | float -> float -> float |]. This implies that double can be used for int -> int and float -> float. Run time types ============== G'Caml has run time types, values which represent ML types. Definition of the run time types is given in stdlib/rtype.mli. It is almost identical to the representation of G'Caml's types, but the completely internal informations which are usually unvisible to the users are removed. Addition to the original contents of G'Caml's types, data type declaration information is attached to variant types. Using this type declaration values, we can define generic functions which works for various data types, for example. Type declaration information of data type t is obtained by the following typedecl expression: typedecl t *NOTE* Type declaration contains recursive reference to itself, when it is defined recursively. So be careful if you write a function which traverses type declaration values. You will be fallen into an infinite loop, unless you make a note of alrady visited type declarations. *LIMITATION* Only the basic part of types are supported at this moment, no fancy, modern class, object, polymorphic variant types! But they will be supported gradually. Run time type construction syntax ================================= You can build a value of run time type t using the notation [: t :]. For example, [: int -> int :] is the run time type representation of the type int -> int. The scope of type variables inside [: :] notation is independent from the outside context. For instance, in the following expression, fun (x:'a) -> [: int -> 'a :] the first and second occurrences of the type variable 'a have no relationship each other. Even if the first 'a is instantiated to some other type, the second remains as a type variable: fun (x:'a) -> [: int -> 'a :], x + 1 You can use ^x notation inside [: :], to substitute a run time type bound to a variable x. Ex: let x = [: int :] in [: ^x -> ^x :] will return [: int -> int :]. This ^ notation can only take identifiers. For simplicity, you cannot write any other expressions, though they might be useful. For example, the following is NOT permitted: [: ^(List.assoc "type" type_table) -> unit :] You can use this [: :] notation also as patterns. For example: match t with | [: int :] -> ... | [: float :]] -> ... | [: ^x -> ^y :] -> ... As always, you cannot use one pattern variable ^x or one type variable 'a more than once. Generic primitives ================== You can define "generic primitives". Ex: generic val dyn : {'a} => 'a -> dyn = fun ty v -> (ty,v) generic val x : t = e defines a generic primitive x whose type scheme is t, whose semantics is given by the expression e. The expression e will first take run time type arguments, which inform the type instantiations of generalized type variables, specially listed in the constraint { } => of the type scheme t. In the above example, the expression (fun ty v -> (ty,v)) will takes a run time type of the instantiation of the generalized type variable 'a of the value dyn, then create a tuple of the type and the value. Note that the type annotation t is not a type constraint, but the true type scheme of the defined value. Things not yet implemented ========================== * Native code compiler * ...