On 6/24/07, Till Varoquaux <till.varoquaux@gmail.com> wrote:
> As said previously, in a HM type ineference system type variables are
> generalized only in "let", that is:
>
> let id x = x (*starting with an identity function*)
> let id'= id id (*this is a partial application; it has type '_a ->'_a*)
> let _=
>   print_string (id' "a");
>   print_int (id' 4)
>
> won't work (since id' has type '_a->'_a).
>
> Replacing the definition of id' with the eta expanded version (let id'
> x = (id id) x)  will however work. It is not 100% equivalent since
> every application of the new id' goes twice through id whilst the old
> was a reference to id....
>
> I hope this answers your question.
> Cheers,
> Till

No, it is related to the so called "value restriction" made Caml-ish
way, in a pure functional (no references) HM types can be generalized.
And I restate my question: can someone give an example, where if the
top-level let-definition types were generalized, this would lead to a
crash?