On 08/03/07, Nathaniel Gray <n8gray@gmail.com> wrote:
> # let f ~x = x;;
> val f : x:'a -> 'a = <fun>
> # f ~x:1;;
> - : int = 1
> # f 1;;
> - : x:(int -> 'a) -> 'a = <fun>

The "1" is assumed to be an argument to "f ~x:a" (where "a" is yet to
be given). Therefore "f ~x:a" must be of type "int -> ...". Now "f
~x:a" is of the same type as (and here actually reduces to)  "a".
Hence this type for the x-labeled argument.

Perhaps looking at this might help, too:
# f 1 1 1;;
- : x:(int -> int -> int -> 'a) -> 'a = <fun>

Now you ask, why are things as they are? Why can't OCaml guess that
the "1" was meant for the label "x"?

Probably it could, but with such a typing rule we would have to write
"(fun ~x:a -> f ~x:a 1)" for what is currently "f 1" (partial
application would be essentially lost here). In contrast, becoming
able to write "f x" instead of "f ~x:1" seems not enough of a gain to
compensate this.

Best,

Roland

-- 
http://www.lix.polytechnique.fr/~zumkeller/