[jboss-user] [JBoss Microcontainer Development] - JBoss Reflect Performance Javassist vs Introspection

[Kabir Khan]

Kabir Khan [http://community.jboss.org/people/kabir.khan%40jboss.com] replied to the discussion

"JBoss Reflect Performance Javassist vs Introspection"

To view the discussion, visit: http://community.jboss.org/message/537476#537476

--------------------------------------------------------------
What we generate is effectively something along the lines of

public class JavassistMethod1 implements JavassistMethod
{
   public Object invoke(Object target, Object[] args) throws Throwable
   {
      if (target == null)
         throw new IllegalArgumentException("Null target");
      if (target instanceof SomeClass == false)
         throw new IllegalArgumentException("Wrong target");
      
      if (args == null || args.length != 10) 
         throw new IllegalArgumentException("Wrong number of parameters");
      
      if (args[0] == null) 
         throw new IllegalArgumentException("Parameter 1 cannot be null for SomeClass.someMethod(long, String)");
      if (args[0] != null && args[0] instanceof Long == false) 
         throw new IllegalArgumentException("Parameter 1 is not an instance of Long for SomeClass.someMethod(long, String)");
      
      if (args[1] != null && args[1] instanceof String == false) 
         throw new IllegalArgumentException("Parameter 2 is not an instance of String for SomeClass.someMethod(long, String)");
      
      return ((SomeClass)target).someMethod(((Long)args[0]).longValue(), (String)args[1]);      
   }
}


Disabling the parameter checking so we end up with this instead reduces the time spent in C from about 4.8s to 3.8s

public class JavassistMethod1 implements JavassistMethod
{
   public Object invoke(Object target, Object[] args) throws Throwable
   {
      return ((SomeClass)target).someMethod(((Long)args[0]).longValue(), (String)args[1]);      
 
   }
} 
 


Profiling this, for each JavassistMethod implementation the time taken is roughly

50% generating the bytecode for the method 
30% Converting the bytecode to a byte array and creating the class
13% creating the constructor
6% setting the interfaces

Maybe a better approach would be
1) to do the parameter checking in Javassist[Constructor/Method/Field]Info itself
2) to generate less classes? Something like:

public class SomeBeanAccessor implements JavassistBeanAccessor
{
    public Object newInstance(int index, Object[] args) throws Throwable
    {
       if (index == 0)
          return new SomeClass((String)args[0]);
       else if (index == 1)
          return new SomeClass((String)args[1]);
 
       return null;
    }
 
    public Object invoke(long hash, Object target, Object[] args) throws Throwable
    {
        if (hash == 121267912)
      return ((SomeClass)target).someMethod(((Long)args[0]).longValue(), (String)args[1]);
    if (hash == 128172981)
 
      return ((SomeClass)target).otherMethod(((Long)args[0]).longValue(), (String)args[1], (String)args[2]);
 
   }
 
 
 
   public Object get(String name, Object target) throws Throwable
 
   {
 
      if (name.equals("intField")
 
          return Integer.valueOf(((SomeClass)target).intField);
 
      if (name.equals("stringField")
 
          return ((SomeClass)target).stringField;
 
   }
 
 
 
   public void set(String name, Object target, Object value) throws Throwable
 
   {
 
      if (name.equals("intField")
 
          ((SomeClass)target).intField = ((Integer)value).intValue();
 
          return;
 
      if (name.equals("stringField")
 
          ((SomeClass)target).stringField = (String)value;
 
   }
 
}
 


First call by Javassist[Constructor/Method/Field]Info to access the member would result in this being created for ALL members, and cached in JavassistTypeInfo. 

--------------------------------------------------------------

Reply to this message by going to Community
[http://community.jboss.org/message/537476#537476]

Start a new discussion in JBoss Microcontainer Development at Community
[http://community.jboss.org/choose-container!input.jspa?contentType=1&containerType=14&container=2115]

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/jboss-user/attachments/20100414/d1637993/attachment.html