From mboxrd@z Thu Jan  1 00:00:00 1970
Received: by margaux.inria.fr, Wed, 5 May 93 17:43:33 +0200
Received: from concorde.inria.fr by margaux.inria.fr, Wed, 5 May 93 11:53:35 +0200
Received: from couchey.inria.fr by concorde.inria.fr; Wed, 5 May 1993 11:53:57 +0200
Received: by couchey.inria.fr (5.57/Ultrix3.0-C)
	id AA04243; Wed, 5 May 93 11:53:31 +0200
Date: Wed, 5 May 93 11:53:31 +0200
From: Damien.Doligez@inria.fr (Damien Doligez)
Message-Id: <9305050953.AA04243@couchey.inria.fr>
To: caml-list@margaux
Subject: Re: new library modules
Sender: weis@margaux

Xavier Leroy writes :

>I guess you mean (in the case of "set"):
[...]
>        type 'a ordering == 'a -> 'a -> int;;
>
>        type ('a, 'b) set_operations =
>          { empty: 'a t;
>            is_empty: 'a t -> bool;
>            mem: 'a -> 'a t -> bool;
>            add: 'a -> 'a t -> 'a t;
>            iter: ('a -> 'b) -> 'a t -> unit;
>            (* and so on *) };;
>
>        value make: 'a ordering -> ('a, 'b) set_operations;;

This is not what I meant.  What I meant was :

  type 'a t;;
  type 'a ordering == 'a -> 'a -> int;;

  value make : 'a ordering -> 'a t
  and   is_empty : 'a t -> bool
  and so on ...

and in the implementation :

  type 'a t = { my_order: 'a ordering; data: (* internal representation *) };;

This has the advantage that you cannot use the wrong ordering when working
with a given set, and that you do not have to use a type annotation when
creating the set (with the "make" function).

Damien Doligez