[MLton] property-list vs. hashtable

[Matthew Fluet]

> It would be nice to do an experiment to use hash tables instead of
> property lists.  Unfortunately, I don't see how to do it without
> rewriting all the uses of property lists.  I don't think your proposal
> works.
>
> > functor PropertyList(H: HASH_TABLE): PROPERTY_LIST =
> >   struct
> >      datatype t = T of Word.word * unit ref
> >
> >      local
> >        fun pseudo-rand () = ...
> >      in
> >        fun new () = T (pseudo-rand(), ref ())
> >      end
> >
> >      fun newProperty () =
> >        let
> >          val ht = HashTable.new {hash = fn T (w, _) => w}
> >          ...
> >        end
> >   end
>
> For the hash table to work, you need to hash each element of X,
> i.e. each variable.  What you've written here only has a hash for each
> property.  I don't see how it can be made to work with the current
> interface.  We need something more like

I don't understand your objection.  The hash for each element of X is it
Word.word in the "property-list" associated with X.  Every property get's
its own hash-table, but they all share the same hashing function.