[MLton-user] ascription in MLton source

[Matthew Fluet]

On Sat, 23 Feb 2008, Matthew Fluet wrote:
> Second, nearly every declared type is defined as a datatype, which is 
> generative and creates a new type.  If the datatype constructor(s) are not 
> revealed by the signature, then the type is effectively opaque. Consider:
>   structure S1 : sig type t end = struct datatype t = T of int end
>   structure S2 :> sig type t end = struct type t = int end
> Although S1 uses a transparent signature ascription, no client of S1 can 
> exploit the data implementation details -- there is no way to create or 
> destruct values of type S1.t without the S1.T data constructor.

It turns out that there is one (ever so slight) implementation detail that 
a client of S1 can exploit that a client of S2 cannot -- polymorphic 
equality.  That is, the transparent signature ascription on S1 (although 
it hides the T data constructor) does reveal that S1.t is an equality 
type, whereas the opaque signature ascription on S2 does not reveal that 
S2.t is an equality type.

One could take this observation as another argument against equality types 
in SML.