Modules with null ClassLoaders

[flavia.rainone]

I have just found out that ClassPoolRepository is registering null class loaders.

JBossClDelegatingClassPoolFactory.create method has a call to registerBootstrapClassLoaders

public synchronized AbstractClassPool create(ClassLoader cl, ClassPool src, ScopedClassPoolRepository repository)
   {
      if (this.currentClassLoaders.contains(cl))
      {
         return null;
      }
      currentClassLoaders.add(cl);
      try
      {
         ClassPool parent = getCreateParentClassPools(cl, src, repository);
         
         if (cl instanceof RealClassLoader)
         {
            Module module = registry.getModule(cl);
            if (module == null)
            {
               module = getModuleForClassLoader(cl);
            }
            registerModuleCallback.registerModule(module);
 --->           registerBootstrapLoaders(module, repository);
...
}

This is the implementation of registerBootstrapLoaders:

private void registerBootstrapLoaders(Module skip, ScopedClassPoolRepository repository)
   {
      Collection<Module> unregistered = registerModuleCallback.getUnregisteredModules();
      if (unregistered.size() > 0)
      {
         for (Module module : unregistered)
         {
            if (module == skip)
            {
               continue;
            }
            ClassLoader loader = getClassLoaderForModule(module);
            if (this.currentClassLoaders.contains(loader))
            {
               continue;
            }
            ClassPool classPool = repository.registerClassLoader(loader);
            if (classPool == null)
            {
               repository.unregisterClassLoader(loader);
            }
         }
      }
   }

The problem is that many of those modules, if not all, have a null class loader at this point.

I'll try skipping the registration of such modules to see if any test breaks. Does anybody has a better idea on how to fix this?

[flavia.rainone]

[e-mail] wrote:

I'll try skipping the registration of such modules to see if any test breaks. Does anybody has a better idea on how to fix this?

That didn't break of the tests. Thinking again, it makes a lot of sense to skip the registration of such modules. Given that the newly created class pool will be mapped to a null class loader, it is useless. We are just generating overhead that way.

Now I wonder which scenarios registerBootstrapLoaders is really needed for.

[kabirkhan]

[e-mail] wrote:

[e-mail] wrote:

I'll try skipping the registration of such modules to see if any test breaks. Does anybody has a better idea on how to fix this?

That didn't break of the tests. Thinking again, it makes a lot of sense to skip the registration of such modules. Given that the newly created class pool will be mapped to a null class loader, it is useless. We are just generating overhead that way.

Now I wonder which scenarios registerBootstrapLoaders is really needed for.

That was probably just me fumbling in the dark :-) If some of them are not null, the purpose is probably to register those. Or are all of them null? I'm not familiar enough with the classloading/Module stuff to really know why the modules would have null classloaders

[flavia.rainone]

I created a Jira for this issue: JBREFLECT-95

[flavia.rainone]

[e-mail] wrote:

That was probably just me fumbling in the dark :-) If some of them are not null, the purpose is probably to register those. Or are all of them null? I'm not familiar enough with the classloading/Module stuff to really know why the modules would have null classloaders

Thanks for the response!

I'll check that to see if this method is reallly needed.

[flavia.rainone]

[e-mail] wrote:

I'll check that to see if this method is reallly needed.

I'll wait until all tests are passing before I try to remove that method.

[flavia.rainone]

[e-mail] wrote:

That was probably just me fumbling in the dark :-) If some of them are not null, the purpose is probably to register those. Or are all of them null? I'm not familiar enough with the classloading/Module stuff to really know why the modules would have null classloaders

I just removed registerBootstrapLoaders body to test that. And it turns out that there are modules that are not null and that should be registered. I'm getting several failures at the AOP/AS tests with registerBootstrapLoaders disabled. So, I'll keep things the way they are.