ONJava.com    
 Published on ONJava.com (http://www.onjava.com/)
 See this if you're having trouble printing code examples


Aspect-Oriented Programming and JBoss

by Bill Burke and Adrian Brock
05/28/2003

Overview

Aspect-oriented programming (AOP) is an exciting new paradigm that should have the same effect on software development that object-oriented programming (OOP) had 15-20 years ago. AOP and OOP are not competing technologies, but actually complement each other quite nicely. OOP is great for modeling common behavior on a hierarchy of objects. Its weakness is in applying common behavior that spans multiple non-related object models; this is where AOP comes in. AOP allows you to define cross-cutting concerns that can be applied across separate, and very different, object models. It allows you to layer — rather than embed — functionality so that code is more readable and easier to maintain. We like to think of OOP as top-down software development, while AOP is left-right; they're completely orthogonal technologies that complement each other quite nicely.

Where the tools of OOP are inheritance, encapsulation, and polymorphism, the components of AOP are advices/interceptors, introductions, metadata, and pointcuts. Let's take a look at these definitions.

Advices/Interceptors

An advice is logic that is triggered by a certain event. It is behavior that can be inserted between a caller and a callee, a method invoker and the actual method. Advices are really the key to AOP. These constructs allow you to define cross-cutting behavior. Advices allow you to transparently apply things like logging and metrics to an existing object model.

In JBoss AOP, we implement advices using interceptors. You can define interceptors that intercept method invocations, constructor invocations, and field access. Later, we'll examine how to apply these interceptions to an existing object model.

Introductions

Introductions are a way to add methods or fields to an existing class. They allow you to even change the interfaces an existing class currently implements and to introduce a mix-in class that implements these new interfaces. Introductions allow you to bring multiple inheritance to plain Java classes. A great use case for an introduction is when you have an aspect for which you want to have a runtime interface. You want to apply your aspect across different object hierarchies, but you still want application developers to be able to invoke aspect-specific APIs.

Apple apple        = new Apple();
LoggingAPI logging = (LoggingAPI)apple;
Apple.setLoggingLevel(VERBOSE);

Introductions can be a means to attach a new API to an existing object model.

Metadata

Metadata is additional information that can be attached to a class, either statically or at runtime. It's even more powerful when you can dynamically attach metadata to a given instance of an object. Metadata is great when you are writing truly generic aspects that can be applied to any object, but the logic needs to know class-specific information. A good analogy of metadata in use is the EJB specification. In EJB XML deployment descriptors, you define transaction attributes on a per-method basis. The application server knows when and where to begin, suspend, or commit a transaction because you have defined methods as Required, RequiresNew, Supports, etc., within metadata bindings between your EJB class and the transaction manager, in the bean's XML configuration files.

C# has metadata built right into the language. XDoclet is another fine example of metadata in action. If you have ever used XDoclet to generate EJB files and deployment descriptors, you know the power of metadata. The Java Community Process (JCP) agrees, as metadata is being added to the Java language in JDK 1.5 (see JSR175). Until JSR 175 becomes a reality, though, a good AOP framework should provide a mechanism to declare class-level metadata that is available at runtime.

Pointcuts

If interceptors, introductions, and metadata are the features of AOP, pointcuts are the glue. Pointcuts tell the AOP framework which interceptors to bind to which classes, what metadata to apply to which classes, or what classes to which an introduction will be introduced. Pointcuts define how various AOP features are applied to the classes in your application.

AOP in Action

Example 1. Using Interceptors

JBoss 4.0 comes with an AOP framework. This framework is tightly integrated with the JBoss application server, but you can also run it standalone on your own applications as well. You can never truly understand a concept until you have seen it in action, so let's use examples from JBoss AOP to illustrate how all of this stuff works together. In the rest of this article, we will build a simple tracing framework using AOP.

Defining an Interceptor

The first thing that must be done to implement our little tracing framework is to define the interceptor that will do the actual work. All interceptors in JBoss AOP must implement the org.jboss.aop.Interceptor interface.

public interface Interceptor
{
   public String getName();
   public InvocationResponse invoke(Invocation invocation) throws Throwable;
}

All fields, constructors, and methods that are intercepted in JBoss AOP get turned into a generic invoke call. Method parameters are stuffed into an Invocation object, and the return value of a method, field access, or constructor is stuffed into an InvocationResponse object. The Invocation object also drives the interceptor chain. In the interest of clarification, let's see how all of these objects fit together in an example.

import org.jboss.aop.*;
import java.lang.reflect.*;

public class TracingInterceptor implements Interceptor
{
   public String getName() { return TracingInterceptor; }
   public InvocationResponse invoke(Invocation invocation) 
       throws Throwable
   {
      String message = null;

      if (invocation.getType() == InvocationType.METHOD)
      {
         Method method = MethodInvocation.getMethod(invocation);
         message       = method:  + method.getName();
      }
      else if (invocation.getType() == InvocationType.CONSTRUCTOR)
      {
         Constructor c = ConstructorInvocation.getConstructor(invocation);
         message       = constructor:  + c.toString();
      }
      else
      {
         // Do nothing for fields.  Just too verbose.
         return invocation.invokeNext();
      }

      System.out.println(Entering  + message);

      // Continue on.  Invoke the real method or constructor.
      InvocationResponse rsp = invocation.invokeNext();
      System.out.println(Leaving  + message);
      return rsp;
   }
}

The above interceptor will intercept all calls to a field, constructor, or method. If the invocation type is a method or constructor, a trace message with the signature of the method or constructor will be output to the console.

Attaching an Interceptor

Okay, so we've defined the interceptor. But how do we attach the interceptor to an actual class? To do this, we need to define a pointcut. For JBoss AOP, pointcuts are defined within an XML file. Let's see what this looks like.

<?xml version="1.0" encoding="UTF-8">
<aop>
   <interceptor-pointcut class="POJO">
      <interceptors>
         <interceptor class="TracingInterceptor" />
      </interceptors>
   </interceptor-pointcut>
</aop>

The pointcut above attaches the TracingInterceptor to a class named POJO. This seems a bit cumbersome; do we have to create a pointcut for each class we want to trace? Luckily, the class attribute of interceptor-pointcut can take any regular expression. So if you wanted to trace every class loaded by the JVM, the class expression would change to .*. If you only wanted to trace a particular package, then the expression would be com.acme.mypackge.*.

When running JBoss AOP standalone, any XML file that fits the META-INF/jboss-aop.xml pattern will be loaded by the JBoss AOP runtime. If that relative path is contained in any JAR or directory in your CLASSPATH, that particular XML file will be loaded by the JBoss AOP runtime at startup.

Running the Example

We'll use the pointcut defined above to run the example. The POJO class looks like this:

public class POJO
{
   public POJO() {}
   public void helloWorld() { System.out.println(Hello World!); }
   public static void main(String[] args)
   {
      POJO pojo = new POJO();
      pojo.helloWorld();
   }
}

The TracingInterceptor will intercept calls to main(), POJO(), and helloWorld(). The output should look like this:

Entering method: main
Entering constructor: public POJO()
Leaving constructor: public POJO()
Entering method: helloWorld
Hello World!
Leaving method: helloWorld
Leaving method: main

You can download JBoss AOP and sample code here. To compile and execute:

$ cd oreilly-aop/example1
$ export CLASSPATH=.;jboss-common.jar;jboss-aop.jar;javassist.jar
$ javac *.java
$ java -Djava.system.class.loader=org.jboss.aop.standalone.SystemClassLoader POJO

JBoss AOP does bytecode manipulation to attach interceptors. Because there is no compilation step, the AOP runtime must have total control of the ClassLoader. So you must override the system classloader with a JBoss-specific one if you are running outside of the JBoss application server.

Example 2. Using Metadata

The TracingInterceptor does not trace field access because it is a bit too verbose. It is a common practice for developers to implement get() and set() methods to encapsulate field access. It would be nice if the TracingInterceptor could filter out and not trace these methods. This example shows you how to use JBoss AOP metadata to implement this filtering on a per-method basis. Usually, metadata is used for more complex things like defining transaction attributes, per-method security roles, or persistence mappings, but this example should be good enough to illustrate how metadata could be used in an AOP-enabled application.

Defining Class Metadata

To add this filtering functionality, we will provide a flag that you can use to turn tracing off. We will go back to our AOP XML file to define the tags that will remove tracing for get() and set() methods. Actually, it's kind of pointless to be tracing the main() function, so let's filter that out, too.

<?xml version="1.0" encoding="UTF-8">
<aop>
   <class-metadata group="tracing" class="POJO">
      <method name="(get.*)|(set.*)">
        <filter>true</filter>
      </method>
      <method name="main">
        <filter>true</filter>
      </method>
   </class-metadata>
</aop>

The above XML defines a grouping of attributes called tracing. The filter attribute will be attached to every method starting with get or set. The regular expression format uses JDK-1.4-defined expressions. This metadata can be accessed within the TracingInterceptor through the Invocation object.

Accessing Metadata

For this metadata to be useful, it must be accessible at runtime. Class metadata is accessible through the Invocation object. To make use of it in our example, the TracingInterceptor must be modified a tiny bit.

public class TracingInterceptor implements Interceptor
{
   public String getName() { return TracingInterceptor; }
   public InvocationResponse invoke(Invocation invocation) 
       throws Throwable
   {
      String filter = (String)invocation.getMetaData(tracing, filter);
      if (filter != null && filter.equals(true))
	  	return invocation.invokeNext();

      String message = null;

      if (invocation.getType() == InvocationType.METHOD)
      {
         Method method = MethodInvocation.getMethod(invocation);
         message       = method:  + method.getName();
      }
      else if (invocation.getType() == InvocationType.CONSTRUCTOR)
      {
         Constructor c = ConstructorInvocation.getConstructor(invocation);
         message       = constructor:  + c.toString();
      }
      else
      {
         // Do nothing for fields.  Just too verbose.
         return invocation.invokeNext();
      }

      System.out.println(Entering  + message);

      // Continue on.  Invoke the real method or constructor.
      InvocationResponse rsp = invocation.invokeNext();
      System.out.println(Leaving  + message);
      return rsp;
   }
}

Running Example 2

The POJO class has been expanded a bit to add get() and set() methods.

public class POJO
{
   public POJO() {}
   public void helloWorld() { System.out.println(Hello World!); }

   private int counter = 0;

   public int getCounter() { return counter; }
   public void setCounter(int val) { counter = val; }
   public static void main(String[] args)
   {
      POJO pojo = new POJO();
      pojo.helloWorld();
      pojo.setCounter(32);
      System.out.println(counter is:  + pojo.getCounter());
   }
}

The TracingInterceptor will intercept calls to main(), POJO(), and helloWorld(). The output should look like this:

Entering constructor: public POJO()
Leaving constructor: public POJO()
Entering method: helloWorld
Hello World!
Leaving method: helloWorld

You can download JBoss AOP and sample code here. To compile and execute:

$ cd oreilly-aop/example2
$ export CLASSPATH=.;jboss-common.jar;jboss-aop.jar;javassist.jar
$ javac *.java
$ java -Djava.system.class.loader=org.jboss.aop.standalone.SystemClassLoader POJO

Example 3. Using Introductions

It would be cool if we could turn tracing off and on for specific instances. JBoss AOP has an API to attach metadata to an object instance, but let's pretend an actual tracing API is a better solution. In this example, we'll change the definition of the POJO class itself through the use of an introduction. We will force the POJO class to implement a tracing interface and provide a mix-in class that handles the new tracing API. This will be the tracing interface:

public interface Tracing
{
   public void enableTracing();
   public void disableTracing();
}

Defining a Mix-in Class

The Tracing interface will be implemented in a mix-in class. An instance of this mix-in class will be attached to the POJO class when a POJO is instantiated. Here's the implementation:

import org.jboss.aop.Advised;

public class TracingMixin implements Tracing
{
   Advised advised;

   Public TracingMixin(Object obj)
   {
      this.advised = (Advised)obj;
   }

   public void enableTracing()
   {
      advised._getInstanceAdvisor().getMetaData().addMetaData(
	  	"tracing", "filter", true);
   }

   public void disableTracing()
   {
      advised._getInstanceAdvisor().getMetaData().addMetaData(
	  	"tracing", "filter", false);
   }
}

The enableTracing() method attaches the filter attribute to the object instance. The disableTracing() method does the same, but turns the filter attribute to false. These two methods are examples of how metadata can be applied to more than the class level. That metadata can applied at the instance level, as well.

Attaching an Introduction

Okay, so we've defined the tracing interface and implemented the mix-in class. The next step is to apply the introduction to the POJO class. As for interceptors, we must define another pointcut in XML. Let's see what this looks like.

<?xml version="1.0" encoding="UTF-8">
<aop>
   <introduction-pointcut class="POJO">
      <mixin>
         <interfaces>Tracing</interfaces>
         <class>TracingMixin</class>
         <construction>new TracingMixin(this)</construction>
      </mixin>
   </introduction-pointcut>
</aop>

The above pointcut will force the POJO class to implement the Tracing interface. Now, when an instance of POJO is instantiated, an instance of TracingMixin will also be instantiated. The way TracingMixin is instantiated is defined in the <construction> tag. You can put any one-line Java code you want in the <construction> tag.

Running Example 3

The POJO class has been expanded a bit to show how the Tracing API can now be accessed. The TracingInterceptor hasn't changed from Example 2.

public class POJO
{
   public POJO() {}
   public void helloWorld() { System.out.println(Hello World!); }

   public static void main(String[] args)
   {
      POJO pojo     = new POJO();
      Tracing trace = (Tracing)this;
      pojo.helloWorld();

      System.out.println("Turn off tracing.");

      trace.disableTracing();
      pojo.helloWorld();

      System.out.println("Turn on tracing.");

      trace.enableTracing();
      pojo.helloWorld();
   }
}

Notice that we can typecast POJO to the Tracing interface. The output should look like this:

Entering constructor: POJO()
Leaving constructor: POJO()
Entering method: helloWorld
Hello World!
Leaving method: helloWorld
Turn off tracing.
Entering method: disableTracing
Leaving method: disableTracing
Hello World!
Turn on tracing.
Entering method: helloWorld
Hello World!
Leaving method: helloWorld

Notice that the interceptor-pointcut that added the TracingInterceptor also applies to the methods introduced by the Tracing introduction.

To compile and run this example:

$ cd oreilly-aop/example3
$ export CLASSPATH=.;jboss-common.jar;jboss-aop.jar;javassist.jar
$ javac *.java
$ java -Djava.system.class.loader=org.jboss.aop.standalone.SystemClassLoader POJO

Conclusions

Aspect-oriented programming is a powerful new tool for software development. With JBoss 4.0, you can implement your own interceptors, metadata, and introductions, to make your software development process more dynamic and fluid. Come visit us at www.jboss.org for more detailed documentation. There are a few surprises waiting for you, as we've implemented a suite of services on top of our new framework. Have a look and happy coding.

Bill Burke is a Fellow at the JBoss division of REd Hat Inc. A long time JBoss contributor and architect, his current project is RESTEasy, RESTful Web Services for Java.

Adrian Brock is Director of Support for JBossGroup, LLC.


Return to ONJava.com.

Copyright © 2009 O'Reilly Media, Inc.