Re: [Caml-list] Type inference and marshalling

[Jacques Garrigue <garrigue@math.nagoya-u.ac.jp>]

On 2011/07/06, at 14:11, malc wrote:

> On Wed, 6 Jul 2011, Jacques Garrigue wrote:
> 
>> On 2011/07/05, at 22:59, malc wrote:
>> 
>>> Perhaps someone could explain why following behaves the way it does:
>>> 
>>> ~$ ocaml
>>>       Objective Caml version 3.11.2
>>> 
>>> # let f ic = let i = input_value ic in let j = i + 1 in LargeFile.seek_in ic i;;
>>> Warning Y: unused variable j.
>>> val f : in_channel -> unit = <fun>
>> 
>> The return type of input_value being 'a, which gets generalized by the
>> relaxed value restriction, i gets the polymorphic type "forall 'a. 'a".
>> So you can use it both as an int and an int64.
>> ==> input_value is an unsafe function, you should always write a type
>> 	annotation on its return type.
> 
> Sure i'm well aware of that, but to me "let j = i + 1" means that i has
> type int and after that "LargeFile.seek ic i" makes no sense yet is
> accepted by the type checker.

But this is just the definition of let polymorphism...
If the type of a let-bound value  contains variables, they can be generalized
(with some restriction for soundness).
So i can perfectly have several types.
What makes no sense here is the return type of input_value,
yet this cannot be avoided since there is currently no mechanism
in ocaml to actually check the type of the value received.

I have no simple solution for this with the current standard library.
A potential way to avoid this problem would be to force the user to
provide a monomorphic type:

module type T = sig type t end

let input_value ic (type a) (t : (module T with type t = a)) : a =
  Pervasives.input_value ic

let f ic =
  let i =
    input_value ic (module struct type t = int end : T with type t = int) in
  let _ = i + 1 in seek_in ic i;;

This is verbose, but some syntactic sugar could be easily provided.
In the long term, safe input primitives are the solution.

Jacques Garrigue