Re: [Chicken-users] tinyclos value matching?

[Alex Shinn]

At Thu, 16 Mar 2006 10:20:45 +0100, felix winkelmann wrote:
> 
> On 3/15/06, Topher Cyll <address@hidden> wrote:
> >
> > Is there any way to specify generic methods on values instead of class
> > in Chicken's tinyclos, like CLOS's (eql value) or Swindle's (singleton
> > value) or (name = value) forms?
> 
> TinyCLOS doesn't have anything like this built-in, I'm afraid.
> But I think it might be possible with MOP hacking, since classes
> can be generated easily.

The attached allows you to write code like the following:

(define-generic shortcut-add <value-generic>)

(define-method (shortcut-add (x <number>) (y <number>))
  (print 'x+y)
  (+ x y))

(define-method (shortcut-add (x <number>) (zero 0))
  (print 'x+zero)
  x)

(define-method (shortcut-add (zero 0) (x <number>))
  (print 'zero+x)
  x)

In general, you can specify any non-class value instead of a class.
It matches if it's EQUAL? to the argument, and is always more specific
than being an INSTANCE-OF? a class.

The implementation overrides three methods: COMPUTE-METHOD (to match
values), COMPUTE-METHOD-MORE-SPECIFIC? (to determine value matches are
more specific), and COMPUTE-APPLY-GENERIC (to remove caching).  You
could put value-aware caching back in if you want.

Unfortunately Chicken currently segfaults when you MAKE an inherited
instance of <generic>:

$ csi -q
#;1> (define x (make-class (list <generic>) (list)))
#;2> (make x)
$

so <value-generic> above is just an alias for <generic>, and the
overrides apply to *all* methods.  Once this is fixed, you'll be able
to use the extension only for your chosen generics without any
slowdown to tinyclos in general.

-- 
Alex

value-generic.scm
Description: Binary data