[armedbear-devel] [PATCH] jcall(there are multiple matchingmethods)

[Alessio Stalla]

On Tue, Dec 22, 2009 at 10:03 PM, Tobias C. Rittweiler <tcr at freebits.de> wrote:
> Alessio Stalla writes:
>
>> I assume you mean (funcall (quote j:compareTo) ...), and if we had
>> that, also simply (j:compareTo ...)
>>
>> For that to work, we should alter the semantics of the language quite
>> dramatically, and I'm not even sure the result would be CL anymore.
>> E.g.
>>
>> (let ((fn (symbol-function 'j:someMethod)))
>>   ...much later...
>>   (funcall fn some-obj))
>>
>> should be possible, but in that case the unbound function (or method
>> not found) error can only be signaled at the time of the funcall,
>> since the existence of the method can only be determined given the
>> argument types. I don't think such dramatic changes would buy us
>> anything good.
>
> Why can't it signal an unbound function error?

Because later you might find that the user actually referred to a Java
method, not a Lisp function. You could limit this behavior just to
symbols in the j package, but it would still be an inconsistency with
CL, besides smelling like a hack.

> The existence of methods is checked at call time for generic functions
> as well. If it does not exist, an no-applicable-method error is
> signalled.

Yes, but generic functions are required to be defined before calling
symbol-function to avoid getting an error. Java methods cannot satisfy
this requirement[*]. If symbol-function cannot anymore signal an
undefined function error (at least for some symbols), what about
fboundp? Should it always return T for those symbols? What about
(defun j:something () ...), i.e. redefinition? What about symbol
visibility: are all symbols in j automatically exported so I can write
j:foo for any foo? Etc. etc.
Too many special cases and deviations from CL for my tastes.

A.

[*] We could require Java methods to be declared as well:
(define-jmethod foo "com.xxx.Bar.baz"). It doesn't even require
modifying anything in the language, just have define-jmethod be a
macro expanding to defun. But we would lose all the convenience.