casting null support

Ok, now this is a real need for sure, and one I've known we would hear about eventually.

Your reflection workaround is clever, and probably the only way at the moment. It is obviously less than desirable.

I think this bug may lead the way toward a practical casting syntax/API we can use for Java integration. I'm marking this a blocker for 1.4.

Another option would be to simplify the reflection approach which would then help other cases. Live Connect 3 has something like this:
http://www.mozilla.org/js/liveconnect/lc3_method_overloading.html (see the bottom of the page)

perhaps this could look like:
ObjClass.java_method(:doSomething, Component, :int).invoke(obj, nil, 0)

If this is worth it then there could be some special method on the obj itself so this would be possible:
obj.java_method(:doSomething, Component, :int).invoke(nil, 0)

Scott: That's definitely planned, probably in the form of making the proxy classes (the Class object you get from e.g. java.lang.System-style accesses) provide a complete set of reflective capabilities. For the moment, most reflective logic you would need is available via the Java::JavaClass object you get from calling .java_class, which allows working around most things (in a very painful way).

I started to tackle this but it gets pretty involved; we don't have any logic at present for handling a normal JavaClass/JavaObject/proxy wrapping a null value, so we may want to introduce a new custom type that the method selection can recognize. I'm attaching what I have, which does allow typed null references for simple cases but blows up a bunch of tests and specs.

Not getting fixed for 1.4. java_method and java_send can be used to work around this problem for 1.4 too.

Working off the patch that Charles submitted, I just tracked it down and ran the tests to see if I could find the cause of the problem.

The problem was that the change in src/org/jruby/RubyBasicObject.java would cause more involved tests to fail: getMetaClass().dataGetStruct() was returning null.

I've added a check which should preserve the behaviour of Charles's patch while not breaking any tests.

Now, as for whether getMetaClass().dataGetStruct() should ever return null: this is probably better left to a grander effort tackling more general aims, if you guys would like to pursue it.

I think this is resolved.

In JRuby 1.4 we added the ability call a specific overload of a method with any arguments. So using the bug's example:

# given these methods in class Foo
doSomething(Frame f, int i)
doSomething(Component c, int i)

# could be called using specific signatures using java_send
obj.java_send(:doSomething, [Frame, Java::int], nil, 0)
obj.java_send(:doSomething, [Component, Java::int], nil, 0)

# retrieved as unique-overload methods
obj.java_method(:doSomething, [Frame, Java::int]).call(nil, 0)
obj.java_method(:doSomething, [Component, Java::int]).call(nil, 0)

# or aliased to specific names
class Foo
  java_alias :doSomethingFrame, :doSomething, [Frame, Java::int]
  java_alias :doSomethingComponent, :doSomething, [Component, Java::int]
end