[Openmcl-devel] Interesting behavior with CCL and CLOS

[Gary Byers]

See (e.g.) <http://trac.clozure.com/ccl/ticket/877>

On Wed, 1 Aug 2012, clayton stanley wrote:

> During development with CLOS, I often redefine a class within an active lisp
> session. As far as I know, CLOS fully supports this sort of run-time
> redefinition, but I found a particular case where things aren't as I expect
> them to be:
> CL-USER>
> (defclass cup ()
> ? ((filled-with :accessor filled-with :initarg :filled-with :initform
> 'nothing)))
> #<STANDARD-CLASS CUP>
> CL-USER>?
> (filled-with (make-instance 'cup))
> NOTHING
> CL-USER>?
> (defclass cup ()
> ? ((filled-with :accessor filled-with :initarg :filled-with :initform
> 'nothing))
> ? (:default-initargs :filled-with 'coffee))
> #<STANDARD-CLASS CUP>
> CL-USER>?
> (filled-with (make-instance 'cup))
> COFFEE
> CL-USER>?
> (defclass cup ()
> ? ((filled-with :accessor filled-with :initarg :filled-with :initform
> 'nothing)))
> #<STANDARD-CLASS CUP>
> CL-USER>?
> (filled-with (make-instance 'cup))
> COFFEE
> CL-USER>?
> ~ ? ? ? ? ? ? ?
> 
> This was produced using a base Clozure 1.8-r15286 core file, with Swank
> 2012-03-06 loaded on top for the interactive repl.
> 
> I start by defining a class, make an instance, and the initform is used. All
> is well. I then redefine the class and provide a :default-initarg for that
> slot. I make another instance, and that initarg is used, so all is well
> again. But if I then redefine the class and remove the :default-initarg
> section, when I make the last instance, the :default-initiarg for the slot
> is still used. I would have expected the initform to be used in this case,
> since the :default-initarg for that slot should no longer be present.
> 
> I have run into trouble a few times with this behavior, and I've started to
> favor using :initform modifications (over :default-initarg) so that
> dynamic/run-time class redefinitions behave as I would expect them to
> behave. I'm probably just not fully understanding the CLOS specification, so
> my model of how things should behave is off. If that's the case, I would
> appreciate getting an explanation of why :default-initargs seem to be
> maintained across run-time class redefinitions.?
> 
> Thanks,
> -Clayton
> 
>