[plt-scheme] how to abstract over (new ...) expressions?

How would this work:

(define-struct base (x))
(define-struct (A base) ())
(define-struct (B base) (y))

(define A%
   (class object%
     [init-field fd1]
     (init (st #f))
     (super-new)))

(define B%
   (class object%
     [init-field fd1]
     (init (st #f))
     [init-field (fd2 (B-y st))]
     (super-new)))

(define (convert st)
   (new (if (A? st) A% B%) [fd1 (base-x st)] [st st]))

-- Matthias



On May 5, 2006, at 12:04 PM, Richard Cobbe wrote:

> How does one abstract over (new ...) expressions?
>
> Here's the situation:
>
> (define-struct base (x))
> (define-struct (A base) ())
> (define-struct (B base) (y))
>
> (define A%
>   (class object%
>     [init-field fd1]
>     ...))
>
> (define B%
>   (class object%
>     [init-field fd1]
>     [init-field fd2]
>     ...))
>
> I need to convert instances of A and B into instances of A% and B%
> respectively:
>
> (define (convert st)
>   (if (A? st)
>     (new A% [fd1 (f (base-x st))])
>     (new B% [fd1 (f (base-x st))]
>             [fd2 (g (B-y st))])))
>
> where f and g are some translation functions whose details aren't
> relevant.
>
> There's some common code here that I'd like to abstract out (especially
> because in my real application, the classes share 5 fields and the
> associated 5 translation functions; the second class only adds a single
> field).  But I can't find an elegant way to do this.
>
> Doing this with a straightforward function won't work, since there's
> some syntax involved in the name-expression pairs.
>
> Doing this with a macro doesn't seem to work, because I have to make
> expansion-time decisions based on run-time values -- I have to expand 
> in
> a slightly different way depending on whether the value of the argument
> is an A or a B.
>
> The best I can come up with is playing nasty games with apply,
> make-object, and list manipulation, but I don't like that very much 
> (not
> least because I don't want to rely on the order of these arguments in
> the class definitions).
>
> Are there any possibilities I'm overlooking?
>
> Richard
> _________________________________________________
>   For list-related administrative tasks:
>   http://list.cs.brown.edu/mailman/listinfo/plt-scheme