Re: [Sml-implementers] flexible records

[Stephen W. <sw...@in...>]

> > In MLton, updates to the spine are seen by all the flexrecords that
> > share the spine.  Thus, there is no need to have a list of
> > back-pointers.
> 
> But it only works that easily if you have a representation of type
> schemes that keeps quantifiers implicit. For various reasons, this is
> not the case in the Alice compiler, for example.

In MLton, during type inference, the only implicitly quantified type
variables in type schemes are those corresponding to unknown fields in
flexrecords.  After type inference, MLton translates to an explicitly
typed IL in which all type variables are explicitly quantified.

> > I personally would like at least strdec, and even better topdec.  I
> > think restricting the scope to be valdec is too painful.
> 
> I agree (from a user's POV, that is ;-) ). Note however that the
> Definition in fact restricts overloading resolution to strdec.
> Unfortunately, choosing strdec also has the disadvantage of being pretty
> ambiguous due to the ambiguous grammar. So valdec still seems to be the
> only choice that is (1) unambiguous and (2) can be made consistent with
> overloading without violating explicit rules (besides being much easier
> to implement).

I guess then that I'd vote for strdec (for both) and fixing the
ambiguity.