Re: [Caml-list] Confusing behaviour of type inference for polymorphic classes.

[Jeremy Yallop <yallop@gmail.com>]

On 3 December 2013 12:33, Alain Frisch <alain@frisch.fr> wrote:
> On 12/03/2013 11:17 AM, Jeremy Yallop wrote:
>> Using recursive modules certainly makes it possible to write a class
>> with the correct type, but unless I'm mistaken you lose the ability to
>> override the behaviour on recursive calls -- i.e. the call to (new
>> X.m)#t can no longer be replaced with a call through a self variable.
>
> The instance on which the method is called cannot be "self", since it
> corresponds to different type parameters for the class.

That's true for the approach based on recursive modules, but not for
the unfolding approach outlined in my earlier message, which does
support recursive calls through self.

> I don't know what Dmitri is exactly looking for, but maybe a polymorphic method (instead of a parametrized class) would do the job as well.

Agreed.  For straightforward polymorphic recursion either your
recursive module approach or a simple recursive method or function is
sufficient.  Classes are better for open recursion, but only work
either for regular types or for types with finite irregularity (i.e.
where a finite number of unfoldings bring you back to the original
parameter list).