Re: Re: defmethod vs &aux?

At 12:42 PM 3/11/03 +0100, Frode Vatvedt Fjeld wrote:
>Kenny Tilton <[email protected]> writes:
>
> > (defmethod c-link-ex ((used c-user-notifying)
> >         &aux (user (car *c-calculators*)))
> >    ...etc....
> >
> > produces this in CormanLisp:
> >
> > ;;; An error occurred in function COMPILE-FORM:
> > ;;; Error: &KEY following &AUX in lambda list: (USED &AUX (USER (CAR
> > *C-CALCULATORS*)) &KEY &ALLOW-OTHER-KEYS)
> >
> > actually, some of my macros have the same bug (if this is a bug), not
> > handling &aux. I'll rewrite to work around. fyi.
>
>I don't know Corman Lisp so well, but I know Closette a bit, and I
>suspect the problem is the kludge-arglist function from Closette that
>doesn't take &aux into account.

Yeah, that's almost certainly the problem.

;;; N.B. The function kludge-arglist is used to pave over the differences
;;; between argument keyword compatibility for regular functions versus
;;; generic functions.

(defun kludge-arglist (lambda-list)
   (if (and (member '&key lambda-list)
            (not (member '&allow-other-keys lambda-list)))
       (append lambda-list '(&allow-other-keys))
       (if (and (not (member '&rest lambda-list))
                (not (member '&key lambda-list)))
           (append lambda-list '(&key &allow-other-keys))
           lambda-list)))

the Hyperspec for DEFMETHOD says

" The lambda list of the generic function is congruent with the
lambda list of the method being defined; if the defmethod form
mentions keyword arguments, the lambda list of the generic function
will mention  ..... key (but no keyword arguments). If function-name
names an ordinary function, a macro, or a special operator,
an error is signaled."

(I didn't put ..... in there; that's what's there.  Presumably
its a typo and is supposed to read '&key &allow-other-keys'.

So the first clause of KLUDGE-ARGLIST makes sense.  But I can't
find justification for the 2nd IF clause.  Section 7.6.4 doesn't
seem to justify it.

I guess it's just a simplification:  Always allow any sort of
keywords to appear in any call to a generic function, and not
worry about exactly when they are allowed and when they are not.

I leave as an exercise for some user of &AUX variables in DEFMETHODS
to rewrite KLUDGE-ARGLIST to do the obvious fix.

Note that the last sentence of my quote from the HS is violated
in Corman Lisp:  no error is signalled when a regular function
is defined, and then a method by that name is defined.

(I'm not sure why the HS insists on an error being signalled;
a warning would seem more appropriate.)