[jboss-dev-forums] [Design of Messaging on JBoss (Messaging/JBoss)] - ClassLoader behaviour on remote clustering tests

[flavia.rainone@jboss.com]

We are seeing an unexpected class loader behaviour in the remote clustering tests, which is resulting in a RuntimeException thrown by JBoss AOP, when aop is run in classical instrumentation mode.

The cause of this exception is that JBoss AOP is trying to create a class that has already been created and loaded... JBoss AOP tries to create such a class because the class loader that defined the class is not able of loading the same class, which is totally unexpected.

Look at the pseudocode below:

ClassLoader myLoader = ....;
  | Class class = myLoader.defineClass(....);
  | System.out.println("Same class loader: " + (myLoader == class.getClassLoader()));
  | System.out.println("I wanted to use this loader: " + loader);
  | System.out.println("I got a class with this loader: " + loadedClass.getClassLoader());
  | myLoader.load(class.getName()); 
  | 

The call you can see at the second line is actually made by Javassist, through reflection... the code above is a simplification.

The result of the code above, when run by JBoss AOP during class generation. is a ClassNotFoundException preceeded by the messages:


  | Same class loader: false
  | I wanted to use this loader:  sun.misc.Launcher$AppClassLoader at 1a5ab41
  | I got a class with this loader:  org.jboss.test.messaging.tools.container.ClassLoaderJMXWrapper at 6fa9fc
  | 

Since the call defineClass has been made to sun.misc.Launcher$AppClassLoader at 1a5ab41, we need this class to be able of loading the class later, instead of throwing a ClassNotFoundException.

Does anybody know why we are seeing this behaviour?

View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4142731#4142731

Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4142731