[MLton] flexible types

[Stephen Weeks]

> SML/NJ rejects this program, claiming that u is a definitional spec and
> can't be redefined.

This is a bug in SML/NJ.

> MLton accepts it; are flexible types in signatures
> really transitive in this manner?  (i.e., t is definitely flexible, and
> can be "where type" constrained, but does that mean u can be as well?)
> 
> signature S =
>    sig
>       datatype t = A | B | C
>       type u = t
>    end
> 
> signature R =
>    sig
>       structure S : S
>    end
> 
> signature P =
>    sig
>       structure S : S
>       structure R : R where type S.u = S.u
>    end

Once you have "type u = t", then both t and u denote the same flexible
tycon, so "where S.u = S.u" is exactly the same as "where S.t = S.u",
i.e. they both define the same flexible tycon.