Java bytecode and classes

Playing with Java Classes
and Bytecode
What is a Class?
How it is loaded and used?
How to write a Java program that writes itself at runtime?
Yoav Abrahami
Wix.com

Java Classloading
• Why do we care?
– Because if we’re gonna write code at runtime, we’d better know
how to load and use it…
– Because we don’t really understand classes
• So… what identifies a class?
– Its name
– Its package
– Its classloader
• It means that
– We can have multiple instances of a class loaded at the same time
– Two instances of the same class from different classloaders are not compatible and not
assignable
– Static variables are static only in the context of a classloader, not globally as we’re
always told

Java Classloading
• So what is this classloader?
– A Java class (subclass of java.lang.ClassLoader), responsible for loading other classes
used by the JVM
– Classloaders are arranged as a tree
• Bootstrap classloader
– Loads the Java system
• jre/lib/resources.jar – series of resource files
• jre/lib/rt.jar – the java.*, javax.*, etc packages
• jre/lib/sunrsasign.jar
• jre/lib/jsse.jar – secure socket extension
• jre/lib/jce.jar – Java cryptography extension
• jre/lib/charsets.jar
• jre/classes
– The important stuff is in rt.jar – the base Java classes
Bootstrap
classloader
Ext
classloader
Application
/ System
classloader

Java Classloading
Commandline Java App Tomcat (6)
Bootstrap
classloader
Ext
classloader
Application
/ System
classloader
Application
/ System
classloader
Common
classloader
WAR 1
classloader
WAR 2
classloader

Java Classloading
Commandline Java App Tomcat (6)
Bootstrap
classloader
Ext
classloader
Application
/ System
classloader
Application
/ System
classloader
Common
classloader
WAR 1
classloader
WAR 2
classloader
Loads the Application Jars
from the classpath
Loads only
bootstrap.jar and
tomcat-juli.jar
Loads Tomcat
commons library jars
Loads jars in the
webapp lib directory

LETS TALK BUSINESS
Ok, we understand classes. Where is the cool stuff?

What we’re gonna talk about
• Aspect Oriented Programming
– Java Proxy
– Spring Aspects
– AspectJ Aspects
• Doing the really cool stuff
– The bootstrap classloader
– The javaagent and class instrumentation
– Writing bytecode at runtime
• All in context of when I’ve had to use them

The Java Proxy
• What does it do?
– Allows implementing one or more interfaces dynamically
• When do we use it?
– Generic implementation of an interface – such as in the case of a client calling a service.
The client uses an interface with a generic implementation that marshals the method
calls to whatever “on-the-wire” format. For instance, creating a SOAP client on the fly
using an interface and WSDL
– Simplistic AOP – to catch method calls, perform some pre/post/around logic and
delegate the call to the real implementation. Can be used for transaction handling,
logging, etc.
• Limitations
– Only supports Java interfaces
– Intercepts only calls to the proxy instance. Direct calls to the
delegate will not be intercepted (for instance, a call from one
delegate method to another)

The Java Proxy
• Example code
public interface SomeInterface {
public void doSomething();
public void doAnotherThing(String name);
}
Object p = Proxy.newProxyInstance(this.getClass().getClassLoader(),
new Class<?>[]{SomeInterface.class}, new InvocationHandler() {
public Object invoke(Object proxy, Method method, Object[] args) throws Throwable {
System.out.println("invoking method:" + method.getName());
if (args != null)
for (Object arg: args)
System.out.println(" arg: " + arg);
return null;
}
});
SomeInterface s = (SomeInterface)p;
s.doSomething();
s.doAnotherThing("hello");

Spring AOP
– Allows intercepting method calls to Spring beans, with or without an interface
– Simpler code compared to the Java Proxy
– Based on the proxy model – Spring BeanFactory returns a proxy to the real bean
– Supports choosing the methods to intercept using AspectJ selectors
– Aspects are written using the AspectJ Java syntax
– AOP on Spring beans
– Transaction handling, logging, security handling – anything AOP is good for
• Limitations
– Only supports Spring beans
– Intercepts only calls to the proxy instance. Direct calls to the
delegate will not be intercepted (for instance, a call from one
delegate method to another)
– Supports only a subset of AspectJ selectors

Spring AOP
• Configuring Spring AOP
• Example Aspect
• Activating the Aspect
– Using Spring component scan, by adding the @Component annotation on the aspect
– Using Spring Beans XML
@Aspect
public class ExampleAspect {
@Around("@annotation(com.experiments.RequireLogin)")
public Object around(ProceedingJoinPoint joinPoint) throws Throwable
{
...
}
}
<bean class="com.experiments.ExampleAspect"/>
<aop:aspectj-autoproxy/>

AspectJ
– Allows intercepting any JoinPoint such as method calls, exceptions, etc.
– Supports writing aspects using Java or AspectJ syntax
– Modifies the actual class bytecode
– Supports choosing the methods to intercept using AspectJ selectors
– Modify existing classes, adding methods, members and super-interfaces to them
– Weaves aspects on load time or compile time
– AOP on any Java class
– Transaction handling, logging, security – anything AOP is good for
– Introduce compiler-like coding rules
• Limitations
– For compile-time weaving – replaces the standard Java Compiler
– For load-time weaving – requires elaborate JVM configuration

AspectJ
• Configuration options
– Compile-time weaving
– Load-time weaving
– Load-time weaving with Spring
• Dependencies
– aspectjrt.jar must be in the classpath
• Configuring AspectJ load-time weaving with Spring
– Simple, isn’t it?
• Well, not so simple…
<aop:aspectj-autoproxy/>
<context:load-time-weaver/>

AspectJ
• Need to include META-INF/aop.xml
– declares the aspects for load-time weaving and the target packages to weave into
<!DOCTYPE aspectj PUBLIC "-//AspectJ//DTD//EN" "http://www.eclipse.org/aspectj/dtd/aspectj.dtd">
<aspectj>
<weaver>

<include within="com.experiments..*"/>
</weaver>
<aspects>

<aspect name="com.experiments.ExampleAspect"/>
</aspects>
</aspectj>

AspectJ
• Per-platform configuration for class instrumentation
– Command line & JUnit applications – using the -javaagent to load instrumentation jars
• -javaagent:spring-instrument-<version>.jar -javaagent:aspectjweaver-<version>.jar
– Maven running JUnit – using the maven-surefire-plugin, passing it the javaagent args
– Note the two javaagent params must be at the same line
– Tomcat 6 –
• spring-instrument-tomcat-<version>.jar has to be copied to the Tomcat lib directory
• The webapp must have a context.xml file in WEB-INFcontent.xml, with the minimum content
– There are alternative locations for the context file. This is the location I find easiest to use.
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-surefire-plugin</artifactId>
<version>2.6</version>
<configuration>
<forkMode>once</forkMode>
<argLine> -javaagent:"${settings.localRepository}/org/springframework/spring-
instrument/${org.springframework.version}/spring-instrument-
${org.springframework.version}.jar"
-javaagent:"${settings.localRepository}/org/aspectj/aspectjweaver/
${org.aspectj.version}/aspectjweaver-${org.aspectj.version}.jar" </argLine>
<useSystemClassLoader>true</useSystemClassLoader>
</configuration>
</plugin>
<Context path="/">
<Loader loaderClass="org.springframework.instrument.classloading.tomcat.
TomcatInstrumentableClassLoader"/>
</Context>

ASPECTJ 101
As long as we’re talking about aspects…

AOP 101 – AspectJ Style
• Aspect – a concern that cuts across multiple classes. Examples are logging,
transaction handling, security, etc.
• Join Point – a point during the execution of a program. Examples are when
– A method is executed
– A method is called
– A constructor is executed
– A constructor is called
– An exception handler is executed
– An advice is executed
– Static initialization is executed
– Initialization and pre-initialization of an object
• Advice – an action taken by the aspect at a particular pointcut. Types of advice:
– Before, after returning, after throwing, after finally, around
• Pointcut – a selector of Join Points using predicates. An advice is associated with a
pointcut expression and runs on any point matched by the pointcut

• More on pointcuts
– Pointcuts are query expressions on the code
– May be joined using the and (‘&&’), or (‘||’) or not (‘!’) operators
• Some pointcut predicates
– execution(…) – when a particular method body executes
– call(…) – when a method is called
– handler(….) – when a particular exception handler executes
– this(…) – when the currently executing object is of a certain type
– target(…) – when the target object is of type
– args(…) – when the method arguments match a certain type
– within(…) – when the executing code belongs to a certain class
– cflow(…) – when the program flow is in a certain method call (the method is a parent in
the stack)
– @annotation(…) – methods annotated with a certain annotation
– @target(…) – when the target executing object has a certain annotation
– @args(…) – when the runtime type of an argument has a certain annotation
– @within(…) – limits to pointcuts within a class that has a certain annotation
– bean(…) – a spring bean name

• More on pointcuts
– Pointcuts are query expressions on the code
– May be joined using the and (‘&&’), or (‘||’) or not (‘!’) operators
• Some pointcut predicates
– execution(…) – when a particular method body executes
– call(…) – when a method is called
– handler(….) – when a particular exception handler executes
– this(…) – when the currently executing object is of a certain type
– target(…) – when the target object is of type
– args(…) – when the method arguments match a certain type
– within(…) – when the executing code belongs to a certain class
– cflow(…) – when the program flow is in a certain method call (the method is a parent in
the stack)
– @annotation(…) – methods annotated with a certain annotation
– @target(…) – when the target executing object has a certain annotation
– @args(…) – when the runtime type of an argument has a certain annotation
– @within(…) – limits to pointcuts within a class that has a certain annotation
– bean(…) – a spring bean name
Not
Supported by
Spring AOP
Only Spring
AOP

• Some examples
– execution (int *())
• JoinPoints that are integer-returning method executions that do not take parameters
– @annotation(com.example.Log) && execution(* *(..))
• JoinPoints that are executions of methods annotated with the @Log annotation, regardless of
the method return type, class, name or parameters
– call(public * *(..))
• Call to any public method
– !this(Point) && call(int *(..))
• Any call to a method returning an integer when the executing object is of any type other
than Point
– cflow(P) && cflow(Q)
• All JoinPoint that are both in the control flow of P and in the control flow of Q

• Coding Aspects using AspectJ Java Syntax
• The AspectJ syntax
@Aspect
public class MyAspect {
@Pointcut("within(com.springsource..*)")
public void inSpring() {}
@Pointcut("@Annotation(java.lang.Deprecated)")
public void inDepracated() {}
@Pointcut("inSpring() && inDepracated()")
public void deprecatedInSpring() {}
@Before("deprecatedInSpring()")
public void pointcutRef() {
...
}
@Before("within(com.springsource..*) && @Annotation(java.lang.Deprecated)")
public void pointcutInplace() {
...
}
}
aspect A {
pointcut fooPC(): execution(void Test.foo());
pointcut gooPC(): execution(void Test.goo());
pointcut printPC(): call(void java.io.PrintStream.println(String));
before(): cflow(fooPC()) && cflow(gooPC()) && printPC() && !within(A) {
System.out.println("should occur");
}
}

• Adding code to existing classes using aspects (Inter-type declarations)
– Adding implemented interfaces
– Changing the superclass of a class
– Adding methods to a class
– Adding members to a class
– Implement Mix-ins in Java
• Using Join Points to raise custom compiler errors or warnings
• Privileged aspects
– Can access private, package and protected members of classes, bypassing the java
member visibility constraints

• Example of an aspect modifying a class
aspect PointAssertions {
private boolean Point.assertX(int x) { return (x <= 100 && x >= 0); }
private boolean Point.assertY(int y) { return (y <= 100 && y >= 0); }
before(Point p, int x): target(p) && args(x) && call(void setX(int)) {
if (!p.assertX(x)) {
System.out.println("Illegal value for x");
return;
}
}
before(Point p, int y): target(p) && args(y) && call(void setY(int)) {
if (!p.assertY(y)) {
System.out.println("Illegal value for y");
return;
}
}
}
public class Point {
int x, y;
public void setX(int x) { this.x = x; }
public void setY(int y) { this.y = y; }
public static void main(String[] args) {
Point p = new Point();
p.setX(3);
p.setY(333);
}
}

The bootstrap classloader
– Allows replacing Java system classes
– Bypasses all Java security policies set by SecurityManager
– Meddling with the core Java APIs
– Terracotta uses it to replace the Java HashMap class with a distributed cache
implementation
• Limitations
– Classes loaded by the bootstrap classloader can’t load classes that are not included in
the bootstrap classpath
• Usage
– With the Java command line options
• –bootclasspath: - list of jars to use instead of the standard list
• –bootclasspath/a: - list of jars to append to the standard list
• –bootclasspath/p: - list of jars to prepend to the standard list

Javaagent and Instrument
– Allows instrumenting / transforming / changing a class as it is loaded
– AspectJ uses it for load-time weaving
– To modify classes as we load them
• Limitations
– Works on the class file bytes – requires intimate knowledge of the class file structure
– Not very useful without a framework like AspectJ, BCEL, cglib, etc.
• Usage
– Using the –javaagent command line option to introduce an Agent jar
– The Agent jar has to have a premain method
• public static void premain(String agentArguments, Instrumentation
instrumentation)
– The Instrumentation class allows to redefine a class,
add & remove transformers, and re-transform a class.
– ClassFileTransformer has one method to transform a class
file bytes (as byte[]).

Writing Bytecode
– Write new classes or modify existing classes at runtime
– Load the new/modified classes
– Instantiate them
– Use them as any other Java class
– Implementing Proxies on any class – Spring AOP and Hibernate do just that
– Hibernate uses bytecode manipulations of classes to introduce mechanisms for lazy-
loading of members
• My own experience
– Performance – when a mapping library (in my case XML to objects) was too slow due to
excessive use of reflection, coding simple mapping classes proved much faster
– Adaptor for Beans – a reporting framework used reflection on beans to read the field
definitions from a JavaObject datasource. When adapting a
Java query interface that returns map-like objects, we had to
generate wrapper classes adapting the map to the beans the
framework expected.

Writing Bytecode
• Limitations
– Hard to use
– Generated bytecode cannot be debugged normally – there’s no source code 
• BCEL – I’ve found using BCEL to be the easiest
• BCEL is included with JRE 1.6
– In package com.sun.org.apache.bcel
– However, the JRE version didn’t work for me
– Eventually, I’ve used the Apache version – BCEL 5.2
• Guidelines
– Keep the generated code small. Use helper methods / superclasses whenever you can.
Remember, you can debug helper classes or superclasses, but you can’t debug
generated bytecode
– Classes can be generated at build-time or runtime.
I’ve found runtime generation simpler to use
– Use class generation-by-example

Writing Bytecode
• Code generation or bytecode generation?
• Code generation
– Performed at build-time - inputs to the generation process must be available at
build time
– Easier to debug
– Tends to clutter the project source code
– Code-generation code tends to be longer than bytecode-generation code
– Hard to get the build right – separation of generated code from actual code
• Bytecode generation
– Performed at runtime – inputs to the generation process can be resolved at runtime
– Requires a higher skill level
– More complex to debug
• At the philosophical level
– If it can be generated, it’s not code!
– We’re just configuring the JVM, not coding

Writing Bytecode
• Class generation-by-example
– Write an example class of what you want to generate
– Compile it
– Run org.apache.bcel.util.BCELifier. It generates a factory class that generates your
example class bytecode
– Modify the generated factory class so that it generates the bytecode for the
classes you want
– Use a bytecode verifier to verify the generated classes
– Decompile generated files and verify method logic. I recommend that you set the
"annotate" option so you can see the bytecode instruction as comments.
– BCEL also provides a tool to generate HTML documentation for your example class
bytecode
• Class loading
– Code a new subclass of ClassLoader
– Use the defineClass (protected) method to define the new
generated class
– Remember to link your classloader to a parent classloader

Writing Bytecode
• Example – mapping objects from type A to type B
Using a superclass helper
public abstract class BaseMapper {
protected Object a;
protected Object b;
public BaseMapper(Object a, Object b) {
this.a = a;
this.b = b;
}
public abstract Object mapToB(Object a);
public abstract Object mapToA(Object b);
}
public class ExampleMappedClass extends BaseMapper{
public ExampleMappedClass(Person person, PersonDTO personDTO) {
super(person, personDTO);
}
@Override
public PersonDTO mapToB(Object a) {
Person person = (Person)a;
PersonDTO personDTO = new PersonDTO();
personDTO.name = person.name;
personDTO.lastName = person.lastName;
personDTO.id = Integer.toString(person.id);
return personDTO;
}
@Override public Person mapToA(Object b) {
PersonDTO personDTO = (PersonDTO)b;
Person person = new Person();
person.name = personDTO.name;
person.lastName = personDTO.lastName;
person.id = Integer.parseInt(personDTO.id);
return person;
}
}
The example class

Writing Bytecode
• Running BCELifier on the class
– BCELifier has a main method, accepting one parameter – the class name
• The created factory class
– Defining the class and the class members
public class ExampleMappedClassCreator implements Constants {
private InstructionFactory _factory;
private ConstantPoolGen _cp;
private ClassGen _cg;
public ExampleMappedClassCreator() {
_cg = new ClassGen("com.experiments.ExampleMappedClass", "com.experiments.BaseMapper",
"ExampleMappedClass.java", ACC_PUBLIC | ACC_SUPER, new String[] { });
_cp = _cg.getConstantPool();
_factory = new InstructionFactory(_cg, _cp);
}
public void create(OutputStream out) throws IOException {
createMethod_0();
createMethod_1();
createMethod_2();
createMethod_3();
createMethod_4();
_cg.getJavaClass().dump(out);
}
...

Writing Bytecode
• Creating the Factory class (continued)
– Creating the constructor
– Which generates bytecode for the constructor
private void createMethod_0() {
InstructionList il = new InstructionList();
MethodGen method = new MethodGen(ACC_PUBLIC, Type.VOID, new Type[] {
new ObjectType("com.experiments.Person"),
new ObjectType("com.experiments.PersonDTO") },
new String[] { "arg0", "arg1" },
"<init>", "com.experiments.ExampleMappedClass", il, _cp);
InstructionHandle ih_0 = il.append(_factory.createLoad(Type.OBJECT, 0));
il.append(_factory.createLoad(Type.OBJECT, 1));
il.append(_factory.createInvoke("com.experiments.BaseMapper", "<init>",
Type.VOID, new Type[] { Type.OBJECT, Type.OBJECT }, Constants.INVOKESPECIAL));
InstructionHandle ih_6 = il.append(_factory.createReturn(Type.VOID));
method.setMaxStack();
method.setMaxLocals();
_cg.addMethod(method.getMethod());
il.dispose();
}
public ExampleMappedClass(Person person, PersonDTO personDTO) {
super(person, personDTO);
}

return personDTO;
}private void createMethod_1() {
MethodGen method = new MethodGen(ACC_1PUBLIC, new ObjectType("com.experiments.PersonDTO"),
new Type[] { Type.OBJECT }, new String[] { "arg0" }, "mapToB", "com.experiments.ExampleMappedClass", il, _cp);
il.append(_factory.createCheckCast(new ObjectType("com.experiments.Person")));
il.append(_factory.createStore(Type.OBJECT, 2));
InstructionHandle ih_5 = il.append(_factory.createNew("com.experiments.PersonDTO"));
il.append(InstructionConstants.DUP);
il.append(_factory.createInvoke("com.experiments.PersonDTO", "<init>", Type.VOID, Type.NO_ARGS,
Constants.INVOKESPECIAL));
il.append(_factory.createFieldAccess("com.experiments.Person", "name", Type.STRING, Constants.GETFIELD));
il.append(_factory.createFieldAccess("com.experiments.PersonDTO", "name", Type.STRING, Constants.PUTFIELD));
il.append(_factory.createFieldAccess("com.experiments.Person", "lastName", Type.STRING, Constants.GETFIELD));
il.append(_factory.createFieldAccess("com.experiments.PersonDTO", "lastName", Type.STRING, Constants.PUTFIELD));
il.append(_factory.createFieldAccess("com.experiments.Person", "id", Type.INT, Constants.GETFIELD));
il.append(_factory.createInvoke("java.lang.Integer", "toString", Type.STRING, new Type[] { Type.INT },
Constants.INVOKESTATIC));
il.append(_factory.createFieldAccess("com.experiments.PersonDTO", "id", Type.STRING, Constants.PUTFIELD));
InstructionHandle ih_41 = il.append(_factory.createReturn(Type.OBJECT));
il.dispose();
}
Writing Bytecode
– Creating the mapToB method

return personDTO;
il.dispose();
}
Writing Bytecode
public
Return type
Input arguments
Class name
Method name
Input argument names

return personDTO;
il.dispose();
}
Writing Bytecode
Store the result as Object 2
Take Object 1 – the a parameter
(Object 0 is this)
Cast it to Person

return personDTO;
il.dispose();
}
Writing Bytecode
Store the result as Object 3
Create new PersonDTO
Invoke the personDTO constructor

return personDTO;
il.dispose();
}
Writing Bytecode
Pop the stack, writing the value to the name property of object 3
Load object 3 to the stack - Person
Read the name field from object 2 to the stack
Load object 2 to the stack - PersonDTO

Writing Bytecode
– Creating the mapToB method – we got a second mapToB method
– The first has signature PersonDTO mapToB(Object)
– The second has signature Object mapToB(Object)
– The second overrides the superclass mapToB method and calls the first
– Narrowing the return type of mapToB was apparently not a good idea, given that
generated classes loaded at runtime do not benefit from the narrowed type constraint
private void createMethod_4() {
MethodGen method = new MethodGen(ACC_PUBLIC | ACC_BRIDGE | ACC_SYNTHETIC, Type.OBJECT,
new Type[] { Type.OBJECT }, new String[] { "arg0" }, "mapToB",
"com.experiments.ExampleMappedClass", il, _cp);
il.append(_factory.createInvoke("com.experiments.ExampleMappedClass", "mapToB",
new ObjectType("com.experiments.PersonDTO"), new Type[] { Type.OBJECT }, Constants.INVOKEVIRTUAL));
il.dispose();
}

class MyClassLoader extends ClassLoader {
MyClassLoader(ClassLoader parent) {
super(parent);
}
public void defineClass(JavaClass javaClass) {
byte[] classBytes = javaClass.getBytes();
defineClass(javaClass.getClassName(), classBytes, 0, classBytes.length);
}
}
public JavaClass create() {
createMethod_0();
createMethod_1();
createMethod_2();
createMethod_3();
createMethod_4();
return _cg.getJavaClass();
}
MyClassLoader myClassLoader = new MyClassLoader(this.getClass().getClassLoader());
ExampleMappedClassCreator creator = new ExampleMappedClassCreator();
JavaClass javaClass = creator.create();
myClassLoader.defineClass(javaClass);
Class<?> clazz = myClassLoader.loadClass(javaClass.getClassName());
Constructor<?> constructor = clazz.getConstructor(Person.class, PersonDTO.class);
BaseMapper mapper = (BaseMapper)constructor.newInstance(new Person(), new PersonDTO());
Writing Bytecode• Using the generated class
– The Class loader
– Creating and loading the new class
– Add a method returning JavaClass to the factory class
– And use it

class MyClassLoader extends ClassLoader {
MyClassLoader(ClassLoader parent) {
super(parent);
}
public void defineClass(JavaClass javaClass) {
byte[] classBytes = javaClass.getBytes();
defineClass(javaClass.getClassName(), classBytes, 0, classBytes.length);
}
}
public JavaClass create() {
createMethod_0();
createMethod_1();
createMethod_2();
createMethod_3();
createMethod_4();
return _cg.getJavaClass();
}
MyClassLoader myClassLoader = new MyClassLoader(this.getClass().getClassLoader());
ExampleMappedClassCreator creator = new ExampleMappedClassCreator();
JavaClass javaClass = creator.create();
myClassLoader.defineClass(javaClass);
Class<?> clazz = myClassLoader.loadClass(javaClass.getClassName());
Constructor<?> constructor = clazz.getConstructor(Person.class, PersonDTO.class);
BaseMapper mapper = (BaseMapper)constructor.newInstance(new Person(), new PersonDTO());
Writing Bytecode• Using the generated class
– The Class loader
– Creating and loading the new class
– Add a method returning JavaClass to the factory class
– And use it
Create the classloader
Create the class factory
Create the class bytecode
Define the new class
Load the new class
Get the new class constructor
Call the constructor

Java bytecode and classes

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Java bytecode and classes (20)

Recently uploaded (20)

Java bytecode and classes