[MLton] (possible) front-end bug

[Matthew Fluet]

> I suspect that SML/NJ is correct; a datatype in a signature would be an
> (implicit) definitional specification, wouldn't it?

But that can't be quite right either.  I do want to be able to say:

signature T =
sig
  structure Y : S where type t = Z.t
                    and type u = Z.u
end

to indicate that any structure matching T will have a substructure Y that
is essentially equal to Z.

I was playing with compiling some of the SML/NJ libraries with MLton and
I'm translating

signature T =
sig
  structure Y : S = Z
end

to the above; the "bug" was a cut-and-paste-and-edit error that ended up
as the code which MLton accepted but SML/NJ did not.