On Mon, Apr 28, 2003 at 10:15:39PM -0700, Per Bothner wrote:
> The iterator model is unclear to me.
> First, the name "iterator" may not be the best choice.
> When I use an "iterator" to iterate through a sequence,
> in most languages I change the location of the iterator,
> but the identity of the iterator object does not change.
> In the proposed API, iterators themselvs are immutable,
> and you move to a new position by getting a new iterator.
> This is reasonable for a (mostly-)functional language
> like Scheme, but perhaps "position" would be a better name?

Well, they are still iterators, the difference is that instead of an
iterator object which is mutated as in other languages, we behave
functionally, returning a new state object from iterator-next.

>
> My conclusion is the model is that there are N+2 iterators for
> a sequence of N elements.  I guess this could be made to work,
> but it is different from what I'd expect.  (Input ports follow
> the N+1 model.)  I think there needs to be some motivation for
> this model, and a little exploration of the consequences.

There is a motivation for this model.  It creates symmetry in the case
that a collection supports reversing the iteration direction.  A program
written to walk backwards from within an iterator will have the same
appearance as one which walks forwards:

(if (not (iterator-at-end? i))
    (begin (do-something)
           (loop (iterator-next i))))

and

(if (not (iterator-at-start? i))
    (begin (do-something)
           (loop (iterator-previous i))))

There are a few other reasons for this model, which I will try to
remember when I leave work.

	Scott