[aspectl-devel] with-special-function-scope

[Pascal Costanza]

Hi Peter,

Sorry for the really long delay, but I am only slowly regaining my 
normal life... ;-)

On 27 Nov 2004, at 13:52, ml13 at onlinehome.de wrote:

> Hi,
>
> I am trying to extend the test example to use a special function that 
> is recursive.
> As an example, I am defining f(x) to compute the sum of the squares
>> from 0 to x.
>
> When I am trying to dynamically redefining the function to simply add 
> 1 to its result, this new definition seems to get called "inside" as 
> well.

Yes, that's right. The methods that you add with dynamic scope are 
active for the respective generic function no matter when and how you 
call it. So even in a recursive call they get called as well when they 
are applicable.

> What do I have to do to get the (assert (eql (f 5) 56))) working??

Special functions use the special method combination by default that 
allows you to :override an existing method. This means that an 
overriden method is canceled out for the current dynamic scope - i.e., 
it is never executed in the current dynamic scoped. You can make use of 
this to do what you want. Here is the code:

(in-package :al-user)

(define-special-function f (x)
   (:definer f*))

(defmethod f* ((scope t) x)
   (if (zerop x)
       0
     (+ (* x x)
        (f (1- x)))))

(assert (eql (f 5) 55))

(with-special-function-scope (f*)
   (defmethod* f* :around ((scope dynamic) x)
     (1+ (with-special-function-scope (f*)
           (defmethod* f* :override :around ((scope dynamic) x)
             (call-next-method))
           (call-next-method))))
   (assert (eql (f 5) 56)))

(assert (eql (f 6) 91))

The trick is a) to use an :around method and b) to :override that 
:around method with a "skipping" method inside itself. a) is important 
because if you would otherwise use a primary method there would be no 
way to override just the newly added one.

(It might be a good idea to draw a diagram for the interaction in order 
to understand the control flow.)

I hope this helps. Maybe you're right that this should be simpler, so I 
will keep this in mind for possible future changes or additions.

> PS: By the way, is there possibly any example existing of the cached 
> fibonacci function using the current version of aspectL?

I don't have any at hand at the moment, sorry. Should be simple though, 
right? ;)


Cheers,
Pascal

--
Tyler: "How's that working out for you?"
Jack: "Great."
Tyler: "Keep it up, then."