Re: [Caml-list] A (Silly?) Question About Universal Type Quantification

[Martin Jambon <martin.jambon@ens-lyon.org>]

Will M Farr wrote:
> Hello,
> 
> I recently encountered a situation where I had (effectively) the
> following polymorphic type:
> 
> type 'a record = { id : int; data : 'a }

You could do this:

type record = { id : int; data : 'a . 'a }

The only minor problem is that you can't create values of such type :-)


> and the following compare function
> 
> let compare {id = id1} {id = id2} = Pervasives.compare id1 id2
> 
> and wanted to put such records into a set.  However, I could not figure
> out how to make the polymorphic 'a in the type definition "disappear" in
> the module argument to the Set.Make functor.  For example, the obvious
> 
> Set.Make(struct
>   type t = 'a record
>   let compare = compare
> end)
> 
> fails because the 'a in the type definition for t is unbound.  Is there
> no way to do this?  I'm thinking of some sort of "forall" designation,
> which universally quantifies the type parameter, like
> 
> Set.Make(struct
>   type t = forall 'a : 'a record
>   let compare = compare
> end)
> 
> (I'm sure that there is better terminology for this---please pardon my
> ignorance about types and type theory.)
> 
> I ended up solving my problem by placing the record type into a functor,
> whose argument specified the concrete type for data, but I'm curious if
> other solutions exist.

Looks like the right approach.

You could also used a defunctorized version of Set, at the cost of losing the
static guarantee that you won't mix sets using inconsistent comparison functions.


Martin

-- 
http://mjambon.com/