ML's type system is good...

...and the biggest reason is the presence of polymorphic static types with type inference.

Polymorphic?

"Polymorphic" simply means that there are functions and values which can have multiple types. Or, put differently, there are types which can "stand for" multiple other types. For example, the value nil:

- nil;
val it = [] : 'a list

is polymorphic because the 'a parameter can stand for any type. Therefore this value can have multiple types: int list, string list, etc. Assigning specific types to type variables of a polymorphic type is called instantiating the type.

Trivia Bit 1: There are many kinds of polymorphism; the form ML uses is called "universal" or "Hindley-Milner" polymorphism. Trivia Bit 2: Yes, there are "monomorphic" type systems. Pascal is a good example.

Inference review and examples

At a very high level, here is the general algorithm for determining types:

1. Label every parameter or value binding with a fresh type variable.
2. Progressively refine polymorphic types by observing how their corresponding values are used, until no more information can be inferred.

fun myIf(x, y, z) = if x then y else z;




fun squid(a, b, c) =
    let val d = [2.0,3.0];
        val e:(int * (int * string)) = (3, c);
    in
        if a > hd(d) then [hd(b), #2(c)] else tl(b)
    end;