Home / Browse / ArmedBear / J / Mail / Archive

ArmedBear / J

Email Archive: armedbear-j-devel (read-only)

[j-devel] patch: anonymous generic functions

From: Alex Mizrahi <killerstorm@ne...> - 2007-08-01 17:21

helo

currently it's not possible to do (make-instance 'standard-generic-function) 
in ABCL -- it barfs when tries to setf fdefinition when name is an unbound 
slot.

included patch fixes this problem skipping setting fdefinition when name is 
unbound.

now it's possible to do stuff like that:

CL-USER> (defparameter *gf* (make-instance 'standard-generic-function))
*GF*
CL-USER> (mop:set-funcallable-instance-function *gf* (lambda (xx yy) (+ xx 
yy)))
#<FUNCTION (LAMBDA (XX YY)) {1997447}>

CL-USER> (funcall *gf* 3 4)
7

it appears that less trivial stuff like creating subclasses of generic 
function doesn't work, hope it will be fixed somewhen later :)

here's an (obvious) patch:

RCS file: /cvsroot/armedbear-j/j/j/src/org/armedbear/lisp/clos.lisp,v
retrieving revision 1.212
diff -u -r1.212 clos.lisp
--- clos.lisp   23 Feb 2007 10:46:47 -0000      1.212
+++ clos.lisp   1 Aug 2007 17:20:08 -0000
@@ -848,8 +848,10 @@
                     (initial-discriminating-function ,gf args))
                  nil))
   ;; FIXME Do we need to warn on redefinition somewhere else?
-  (let ((*warn-on-redefinition* nil))
-    (setf (fdefinition (%generic-function-name gf)) gf))
+  (let ((*warn-on-redefinition* nil)
+        (name (%generic-function-name gf)))
+    (unless (eq name +slot-unbound+)
+       (setf (fdefinition name) gf)))
   (values))

 (defun make-instance-standard-generic-function (generic-function-class

Thread View

Thread	Author	Date
[j-devel] patch: anonymous generic functions	Alex Mizrahi <killerstorm@ne...>

Re: [j-devel] patch: anonymous generic functions

From: Peter Graves <peter@ar...> - 2007-08-06 07:52

On Wed, 1 Aug 2007 at 20:21:25 +0300, Alex Mizrahi wrote:
> helo
>
> currently it's not possible to do (make-instance 'standard-generic-function) 
> in ABCL -- it barfs when tries to setf fdefinition when name is an unbound 
> slot.
>
> included patch fixes this problem skipping setting fdefinition when name is 
> unbound.
>
> now it's possible to do stuff like that:
>
> CL-USER> (defparameter *gf* (make-instance 'standard-generic-function))
> *GF*
> CL-USER> (mop:set-funcallable-instance-function *gf* (lambda (xx yy) (+ xx 
> yy)))
> #<FUNCTION (LAMBDA (XX YY)) {1997447}>
>
> CL-USER> (funcall *gf* 3 4)
> 7

Well, this may seem to work, for some value of "work", but it bypasses
the normal protocol for setting the discriminating function of a
generic function, and what you end up with is not really a generic
function at all: it has no methods, and no specializers, and it returns
an answer, when you'd expect a "no applicable method" error.

The real problem here is that ABCL's MOP is nowhere near compliant, and
it would take quite a bit of work to fix it properly. I think patching
clos.lisp as you suggest would just make things even more confusing.

-Peter

Re: [j-devel] patch: anonymous generic functions

From: Alex Mizrahi <killerstorm@ne...> - 2007-08-13 16:43

(message (Hello 'Peter)
(you :wrote  :on '(Mon, 6 Aug 2007 00:52:10 -0700))
(

 PG> Well, this may seem to work, for some value of "work", but it bypasses
 PG> the normal protocol for setting the discriminating function of a
 PG> generic function,

well, just (make-instance 'standard-generic-function) returns more-or-less 
normal generic function -- i think after setting all stuff it will be quite 
functional. finalize-generic-function is still called, so it will have it's 
discriminating function as normal.

 PG>  and what you end up with is not really a generic function at all: it
 PG> has no methods, and no specializers, and it returns an answer, when
 PG> you'd expect a "no applicable method" error.

if somebody sets discriminating function with 
set-funcallable-instance-function it's completely up to him to decide if 
that's what he wants, no?
it should be possible to implement custom discriminating function this way 
or whatever.

in my case i just wanted to make a funcallable object -- i didn't want it to 
be a generic function..

 PG> it would take quite a bit of work to fix it properly. I think patching
 PG> clos.lisp as you suggest would just make things even more confusing.

i'm afraid patching clos.lisp as i suggest is inevitable -- no matter how 
CLOS will change, it still wouldn't be possible to make function fbound to 
some name when it has no name at all.
even if you'll completely rewrite FINALIZE-GENERIC-FUNCTION, i think you'll 
still have a form like (setf (fdefinition name) gf) in it. although if 
you'll get rid of FINALIZE-GENERIC-FUNCTION..

)
(With-best-regards '(Alex Mizrahi) :aka 'killer_storm)
"choose no life")