On Thu, 15 Sep 2005, Michael Sperber wrote:

> Andre van Tonder <xxxxxx@later.het.brown.edu> writes:
>
>>  As an example, the hash-table example could be expressed:
>>
>>   (define-type hash-table
>>     (constructor (k) (lambda (pred hasher size)
>>                        (k pred
>>                           hasher
>>                           (make-vector (nearest-prime size))
>>                           0)))
>>     (fields (pred   immutable)
>>             (hasher immutable)
>>             (data   mutable)
>>             (count  mutable))))
>>
>>   (define-type eq-hash-table
>>     (parent hash-table)
>>     (constructor (k) (lambda (pred hasher size)
>>                        (k pred
>>                           hasher
>>                           size
>>                           0)))
>>     (fields (gc-count mutable)))
>
> I'm confused about the K argument in the EQ-HASH-TABLE example.  Does
> it call the primitive constructor of the underlying record type, or
> the construction procedure of the supertype?  If the former, there's
> one argument missing.  (And I'd ask how I could reuse the construction
> procedure of the supertype.)  If the latter, what's it's return value,
> and how do I transform it into an instance of the subtype?

I meant the latter, but I realize now that it would involve an unnecessary
allocation and copy of a supertype instance as an intermediate step, which is
probably unacceptable.

Here is an alternative suggestion:

     (define-type hash-table
       (constructor (lambda (pred hasher size)
                      (values pred
                              hasher
                              (make-vector (nearest-prime size))
                              0))
       (fields (pred   immutable)
               (hasher immutable)
               (data   mutable)
               (count  mutable))))

    (define-type eq-hash-table
      (parent hash-table)
      (constructor (lambda (pred hasher size)
                     (values pred
                             hasher
                             size
                             0)))
      (fields (gc-count mutable)))

This is a little more concise than my previous suggestion, and presumably
allocating intermediate VALUES is cheaper than allocating intermediate
supertype instances.

In the above, the subtype constructor would call the procedure provided for the
supertype with the first three values, which returns four values used to
populate the first four fields of the subtype instance.  The 5th field is then
populated with the last value 0.

This suggestion cannot absorb INIT!, though, as the previous suggestion, so
INIT! would still need to be separate.

Cheers
Andre