VM 1.2 specs unclear

Question

Suppose you have:
class A
class B extends A
We know the phases:
- loading
- load class/superclasses
- linking (verification, preparation, resolving is optional).
- static initializaton
suppose this:
When we are loading B, class A should be loaded as part of the loading phase, but this means resolving a constant pool entry, which is part of the resolving.
This question is tightly connected with classloaders.
When defineClass() is issued, do the superclasses get loaded or not. Or they are loaded on resolveClass().
Seems like defineClass() in Sun VM and Kaffe VM load, verify and prepare the class. In other words, as if resolveClass() is called.

Answer 1

I agree with you. I read them again and it does seem to be so. Loading does involve some resolution.
resolveClass() is clear: it is linkClass():
- prepare class
- verify class
- optionally resolve class (constant pool).
But defineClass() should NOT class resolveClass() internally. In fact I have a small program I have ran on various VM's and it behaves INCONSISTENTLY on ALL of them: Sun JVM, Kaffe JVM, NewMonics trial VM.
resolveClass() has nothing to do with resolving Constant pool references, if it does so it must be quiet. NewMonics VM resolves them straight when you call resolveClass, which is not correct. Now here is the code:
class A
import java.io.*;
class TL extends ClassLoader
public static void main(String args[])
     TL cl = new TL();
     try
          byte [] data = new byte[1000];
          int dl = 0;
          FileInputStream f = new FileInputStream("A.class");
          dl = f.read(data);
          Class cc = cl.defineClass("A", data, 0, dl);
          cl.resolveClass(cc);
//cc.newInstance();
     catch(Exception e)
          System.out.println(e);
Try this:
Compile TL.java, A.java. Change the 0xB7 (invokespecial bytecode) in A.class to 0xFF in binary editor. Next do: java TL.
Sun VM does NOT give you error, when it should if it properly verifies the file. KAffe VM does not do anything either, because they do not have pass 3 of the verifier implemented. NewMonics VM gives you the error when you do resolveClass(), but it also resolves all references in A.class which is incorrect. Sun VM does not resolve references in A.class which is correct.
Next, try to uncomment the cc.newInstance(). Sun VM will now give you error.
This is incorrect behaviour for the Sun VM, since the resolveClass() does not really verify the class.
All in all, defineClass() should not call resolveClass(). defineClass() finishes what is specified in VMSpec as "Loading". Now you have Class instance. For it to be usable someone must call resolveClass(). But between defineClass() and resolveClass() the class type is inprepared and is not verified.
At least thats how I understand it. If anyone has any other interpretation, please share it.