Dynamic AOP in Java

1
Dynamic AOP in Java
2
Overview

• Dynamic AOP
• Domain Analysis
• Proxy based
• Bytecode manipulation
• Interpreter based
• Final Remarks

3
Dynamic AOP

• Adding/removing aspects during runtime
• Can help avoid
• Re-compiling
• Re-deployment
• Re-start of an application
• More flexible than static weaving

4
Domain Analysis

• 3 main approaches
• Proxy based
• Nanning, Spring, Jeet, dynaop
• Bytecode manipulation
• AspectWerkz, JBossAOP, JAC, Wool
• Bytecode manipulation tools
• Javassist, BCEL, ASM
• Interpreter based
• Plugins/Extensions Prose, Axon
• JVM-level AOP JRockit

5
Proxy-based (1/4)

• Proxy
• A proxy is a surrogate or placeholder for a
  target object to control access to it. (Proxy
  pattern)

6
Proxy based (2/4)

• Creating proxies dynamically
• JDK java.lang.reflect.Proxy

7
Proxy-based (3/4)
8
Proxy-based (4/4)

• What if we want to advice classes that dont have
  interfaces?
• CGLIB
• Code Generation Library / API
• Transforms classes into classes
• ASM
• Similar to JDK dynamic proxies
• MethodInterceptor InvocationHandler
• Enhancer Proxy
• Inheritance instead of aggregation

9
Bytecode Manipulation (1/3)

• Bytecode manipulation tools
• BCEL, Javassist, ASM
• Most popular of the 3 approaches
• AspectWerkz, JBossAOP, JAC, Wool
• Some drawbacks
• Performance CPU-intensive
• Scalability 64kb boundary for class files

10
Bytecode manipulation (2/3)

• HotSwap
• Changing method/constructor bodies of classes at
  runtime and inject them into a running JVM
• Fixed schema
• Preparation step is needed
• Part of JVMTI since Java 5
• java.lang.instrument

11
Bytecode manipulation (3/3)

• Comparison BCEL, Javassist, ASM
• ASM much faster
• gt5 times faster than Javassist/BCEL
• Javassist easiest to use
• source code level instrumentation
• lower learning curve than bytecode
  instrumentation

12
Interpreter-based

• Based on intercepting execution events in the JVM
  
• Adapting the JVM
• Plugins/Extensions
• Prose, Axon
• New JVM
• JRockit

13
Interpreter based Prose (1/2)
14
Interpreter-based Prose (2/2)

• Limitations
• No introductions (cannot change the original
  code)
• Slow because of Debugger overhead
• Small public

15
Interpreter-based

• AOP support built in the JVM
• JRockit
• Prototype / not standardized
• Advantages vs bytecode instrumentation
• No startup-cost due to bytecode instrumentation
• No double bookkeeping building a class DB
• Support for intercepting reflective calls

16
Final Remarks

• Some insight given in different techniques that
  can be used to handle dynamic AOP in Java
• Proxy, bytecode manipulation / HotSwap,
  interpreter-based
• Work this out
• Probably using bytecode manipulation
• Javassist , HotSwap
• Also combinations are possible