You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 3 Next »

Tapestry Inversion of Control Container

Why do I need to define an interface for my services? Why can't I just use the class itself?

First of all: you can do exactly this, but you lose some of the functionality that Tapestry's IoC container provides.

The reason for the split is so that Tapestry can provide functionality for your service around the core service implementation. It does this by creating proxies: Java classes
that implement the service interface. The methods of the proxy will ultimately invoke the methods of your service implementation.

One of the primary purposes for proxies is to encapsulate the service's lifecycle: most services are singletons that are created just in time. Just in time means only as soon
as you invoke a method. What's going on is that the lifecycle proxy (the object that gets injected into pages, components or other service implementations) checks on each method invocation
to see if the actual service exists yet. If not, it instantiates and configures it (using proper locking to ensure thread safety), then delegates the method invocation to the service.

If you bind a service class (not a service interface and class), then the service is fully instantiated the first time it is injected, rather than at that first method invocation. Further, you
can't use decorations or method advice on such a service.

The final reason for the service interface / implementation split is to nudge you towards always coding to an interface, which has manifest benefits for code structure, robustness, and testability.

My service starts a thread; how do I know when the application is shutting down, to stop that thread?

This same concern applies to any long-lived resource (a thread, a database connection, a JMS queue connection) that a service may
hold onto. Your code needs to know when the application has been undeployed and shutdown. This is actually quite easy, by using
a service builder method. In your module class:

  public MyService buildMyService(ServiceResources resources, RegistryShutdownHub shutdownHub)
  {
    final MyServiceImpl service = resources.autobuild(MyServiceImpl.class);

    shutdownHub.addRegistryShutdownListener(new RegistryShutdownListener()
    {
      public void registryDidShutdown()
      {
        service.shutdown();
      }
    });

    return service;
  }

This code uses the ServiceResources object to build an instance of MyServiceImpl, with all dependencies injected.
It also injects Tapestry's RegistryShutdownHub service and adds a listener. The example assumes that the service implementation
(but not the service interface) includes a shutdown() method.

A valid alternative to this would be to have MyServiceImpl implement RegistryShutdownListener:

  public MyService buildMyService(ServiceResources resources, RegistryShutdownHub shutdownHub)
  {
    final MyServiceImpl service = resources.autobuild(MyServiceImpl.class);

    shutdownHub.addRegistryShutdownListener(service);

    return service;
  }

It is not recommended that MyServiceImpl take RegistryShutdownHub as a constructor parameter and register itself as a listener Doing so is an example of unsafe publishing.

  • No labels