On Sat, 9 Sep 2006, Jacques Garrigue wrote:
> > Jacques Garrigue suggested using Obj.repr. He also wrote ``Type
> > theoretician answer: What you would need to do that transparently
> > inside the type system is generic functions with dynamics.'' and
> > mentioned GADT.
>
> Small comment: By transparently I meant "without any type
> annotation". Then I gave a solution using normal datatypes for
> annotations, and polymorphic methods, and just mentioned that GADTs
> were not useful in this case. Note that my solution cannot directly
> use polymorphic variants, it uses non-regular types, and
> polymorphic variants have to be regular.

Is that to make type inference tractable? If so, any ideas on what
amount of annotation would be needed to get around this?

> This said, I love polymorphic variants anyway...

I like them a lot, and I often wish there was a dual notion of polymorphic
records in Caml as well. But now I wonder about the non-regular types issue
you bring up; I guess it means that you need plain old sum types too
if you wish to use them.

-- Brian