Quoting Pierre Weis (Pierre.Weis@inria.fr):
> > This might be the case for OCaml, but note that SML97 disallows more
> > general type-constraints than the type apparent in the expression without
> > the constraint (cf. rule (9) and comment in the 97 Definition - p22) 
> 
> That's a good point. It's simple to state and understand.
This way of handling type constraints is inherited from Hope.

[...]
> Another problem is the scope of type variables in type
> constraints. What's the meaning of
> 
> let f (x : 'a) (y : 'a) = y;;
> 
> We may need explicit Forall keywords to express type schemes in constraints.
[...]

This has already been solved in the SML standard and even if it is not 
necessarily easy to understand when formalized, this is quite intuitive :

  First un occurrence of 'a in a value declaration [val valbind] is said to
  be unguarded if the occurrence is not part of a smaller value declaration
  within [valbind]. In this case we say that 'a occurs unguarded in the value
  declaration. 

  Then we say that 'a is scoped at a particular occurrence O of [val valbind]
  in a  program if (1) 'a occurs unguarded in this value declaration, and (2)
   'a does not occur unguarded in any larger value declaration containing
  the occurrence O.

			Old Definition of Standard ML p 20

According to this definition, the 'a's denote the same type variable in
your example.

let g (x : 'a) = let f (x:'a) = x in x
let h (x : 'a) = x

is equivalent to 

let g (x : 'a) = let f (x:'a) = x in x
let h (x : 'b) = x

but not to 

let g (x : 'a) = let f (x:'c) = x in x
let h (x : 'b) = x

The only risk of this solution is that you overconstrain an expression 
because one of your type variable got caught in the scope of another unrelated
variable and then you get stuck with your compiler complaining about
a constraint it cannot fulfill. This is a safe risk.

-- 
Pierre Cregut - pierre.cregut@cnet.francetelecom.fr - +33 2 96 05 16 28
FT.CNET - DTL/MSV - 2 avenue Pierre Marzin - 22307 Lannion Cedex - France