Overview

Jacob can register Java classes for MS application events or callbacks.

Sequence of Events

The normal flow for this is:
  1. Application thread creates an instance of the event handler and registers it with Jacob
  2. The application continues on doing other work.
  3. Some time later, the MS application takes some action and initiates the event callback.
  4. The Java VM receives the event and spins up a new thread to handle it.
  5. The Jacob jni EventProxy in the dll is called by the VM.
  6. The Jacob jni EventProxy creates Variant objects to handle the parameters of the passed in event.
  7. The Jacob jni EventProxy sends the name of the callback and the array of Variant objects to the Java InvocationProxy that was registered to catch events.
  8. The Java InvocationProxy uses reflection to map the event name to a method name with the exact same name.
  9. The Java InvocationProxy sends the message to the registered event handler and returns if the event handler is of type void (standard behavior).
  10. The Java InvocationProxy sends the message to the registered event handler and returns the Variant that resulted from the call back to the Jacob jni EventProxy that then returns it to the windows calling program.

SWING Issues

Swing developers should note that this message comes in on a thread other than the event thread. All objects receiving events that require user intervention or drawing in the UI should use invokeLater() to post requests for actions onto the event queue. Failure to do so will insure random failures in the GUI. Java Web Start (JWS) and other launchers can have additional issues related to the class loader. The Jacob C++ library uses FindClass() to find the Variant class when building the parameter list. FindClass() uses the system class loader which includes only the classes specified at run time or in the CLASSPATH. Most of the application classes in this situation live in an alternate set of class loaders that were created when the launcher located and ran the application classes. This means that the search for Variant will fail usually with the silent and immediate termination of the Java application. The thread classloader probably can�t be used to try and find the class because this new thread does not have a classloader associated with it other than the system class loader. The end result is that the Variant class needs to be located via other means and that the thread classloader should be set to be the context class loader of the event handler class.

1.8 and 1.9 behavior

The Jacob EventProxy class has been modified (off of the 1.8 tree) so that it takes a two step approach towards fixing these problems.
  1. The EventProxy constructor now accepts an extra object, an instance of the Variant class. This gives the EventProxy a way to get to the Variant class and thus to its classloader. All of the callers of the constructor have been modified to pass a Variant object to the EventProxy without programmer intervention.
  2. EventProxy first attempts to locate the Variant class using FindClass()
  3. Failing that, it looks to see if a variant object had been passed in. If so, it calls class() and goes from there.
  4. If all that fails, it logs a message and then fails in the spectacular fashion of the previous versions.

1.10 behavior

The Jacob EventProxy class has been modified so that it takes a different approach towards fixing this problem.
  1. All objects that request event notification are now wrapped in a Java InvocationProxy so that a standard interface is always presented to the JNI EventProxy object.
  2. The EventProxy constructor accepts any Java class. It wraps the class if it is not an InvocationProxy or uses just the passed in object if it is an InvocationProxy. The JNI layer talks to the InvocationProxy instead of talking directly to the event listener as in previous releases.
  3. The Java InvocationProxy has a method on it that will return the Variant class that the EventProxy. The JNI code uses this method to acquire the class so that it can call newInstance().
Developers can receive call back events in JWS other Java launching programs without implementing any additional code. They should be aware that their callback methods may need to set the class loader. If they expect to create any objects.: 
      Public xxx someHandler(Variant[] foo){
            Thread.currentThread().setContextClassLoader(
                  this.getClass().getClassLoader());
            // do something
      }

There may still be a dual event queue issue in JWS applications that needs to be looked at.

1.12 Experimental Behavior

Release 1.12 adds experimental support for event handlers that accept java objects as parameters to closer match the signature of the windows callback. New ActiveXDispatchEvents and ActiveXInvocationProxy operate in tandem in the same way as DispatchEvents and InvocationProxy. DispatchEvents overrides getInvocationProxy() to create a new ActiveXInvocationProxy in place of the normal InvocationProxy. ActiveXInvocationProxy has its own invoke() method that uses reflection to call back using java objects as parameters.

Issues with this approach