🚀 go-pugleaf

Now that the discussion on class-prototype has died down, I thought I'd throw
this on the fire:

STANDARD-CLASS has the interesting property that it is an instance of itself. 
Suppose I want to make a specilized metaclass that also has this property:

(defclass my-class (standard-class) 
  ((arbitrary-slot :initform (gensym)))  ;to keep things interesting
  )

;;; now make my-class be an instance of itself
(change-class (find-class 'my-class) (find-class 'my-class))

;;; test the result (if any...)
(slot-value (find-class 'my-class) 'arbitrary-slot)

Should this work in a correct implementation of CLOS? Should it matter whether
MY-CLASS has been finalized before the call to CHANGE-CLASS? My intuition is that
the above code should perform as expected--even in the absence of a MOP since the
functions involved (DEFCLASS, CHANGE-CLASS and SLOT-VALUE) are all part of the
'regular' CLOS programmer interface. My experience shows otherwise:

Lisp         Platform            Result
----         --------            ------
Lucid 4.0.2  Sparc2/SunOS 4.1.1  Throws into debugger during CHANGE-CLASS.
                                 "Expanding stack due to overflow..."
                                 FINALIZE-INHERITANCE not implemented.
PCL July 92  Sparc2/SunOS 4.1.1  Works as expected regardless of whether
on Lucid                         MY-CLASS is finalized.
Allegro 4.1  Sparc2/SunOS 4.1.1  Throws into debugger during CHANGE-CLASS.
                                 "No methods applicable for gf: CLOS:CLASS-DIRECT-
                                 SLOTS with args (#<CLOS:FORWARD-REFERENCED-CLASS 
                                 STANDARD-CLASS...> of classes (CLOS:FORWARD-
                                 REFERENCED-CLASS)"
                                 Same error occurs if attempt to finalize MY-CLASS
                                 before CHANGE-CLASS.
MCL 2.0      Quadra 700/Sys 7    Throws into debugger during CHANGE-CLASS, but
                                 will work if MY-CLASS has no local slots.
                                 FINALIZE-INHERITANCE not implemented.

Comments?
..................................................
                                  Len Charest, Jr.
                 JPL Artificial Intelligence Group
                          charest@aig.jpl.nasa.gov