Hashtables don't accept empty list as key.

[2bt]

To my understanding this programme conforms with r6rs and it runs without problems in racket. IronScheme throws an exception.

(import (rnrs))
(hashtable-ref (make-eq-hashtable) '() 'default)

[leppie]

Hmmm, this is interesting. An empty list is a null and it is exactly that in IronScheme/CLR. I will see what I can do :)

Thanks

[leppie]

Unfortunately, this is a limitation with the .NET implementation of hashtables...

I could work around it, by creating some substitute for null but it seems hacky, and then you have differences in behavior between .NET and IronScheme which I try my best to avoid.

I will think about it a bit more.

[2bt]

In the meantime, I've been doing something like this:

 (define-record-type nil-record (sealed #t) (opaque #t))
 (define nil (make-nil-record))

 (define hashtable-ref*
   (lambda (h k d)
     (hashtable-ref h (if (null? k) nil k) d)))

 (define hashtable-set!*
   (lambda (h k v)
     (hashtable-set! h (if (null? k) nil k) v)))

 (define hashtable-delete!*
   (lambda (h k)
     (hashtable-delete! h (if (null? k) nil k))))

 (define hashtable-contains?*
   (lambda (h k)
     (hashtable-contains? h (if (null? k) nil k))))

 (define hashtable-entries*
   (lambda (h)
     (let-values
       (((kv vv) (hashtable-entries h)))
       (let ((n (vector-length kv)))
             (do ([i 0 (+ i 1)])
                 ((= i n))
               (when (eq? (vector-ref kv i) nil) (vector-set! kv i '()))))
       (values kv vv)))))

You're right. It feels very hacky and I'm not happy with it. But I need this behaviour of a project. Do you maybe have a better idea? Great work, btw.

[leppie]

That's about right.

You could probably use vector-map instead of the do in the last proc.

If you really want to you can use macros to clean it up, but IMO it is too much effort in this case.

Eg:

(define-syntax-rule (null/subst k) (if (null? k) nil k))

[christoff-buerger]

I will think about it a bit more.

Will this be fixed or does...

I could work around it, by creating some substitute for null but it seems hacky, and then you have differences in behavior between .NET and IronScheme which I try my best to avoid.

...outweigh R6RS conformance?

At the moment I use boilerplate code like above to create a special nil-record used in case of empty lists as keys. Its bothersome and fragile however, to maintain several product versions for different Scheme implementations.

[leppie]

I was thinking of building it into IronScheme libraries, but need to investigate. I think some hashtable procedure(s) are still done in C# which makes it tricky.

[leppie]

@christoff-buerger Should be fixed in last commit, let me know if you pick up issues.

[christoff-buerger]

It works! Thanks a lot!