Name: rmT116609 Date: 11/12/2001


java version "1.4.0-beta3"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.0-beta3-b84)
Java HotSpot(TM) Client VM (build 1.4.0-beta3-b84, mixed mode)


Javac compiles this example in violation of JLS 9.2. As a result, the VM throws
a VerifyError:

$ cat Blah.java
class Blah {
  public static void main(String[] args) {
    I.Inner.bar(new I(){});
  }
}
interface I {
  class Inner {
    static void bar(I i) {
      try {
        i.finalize();
      } catch (Throwable t) {
      }
    }
  }
}

$ javac Blah.java
$ java Blah
Exception in thread "main" java.lang.VerifyError: (class: I$Inner, method: bar
signature: (LI;)V) Bad access to protected data
        at Blah.main(Blah.java:3)

According to JLS 9.2, interfaces only implicitly declare method signatures
corresponding to the PUBLIC methods of Object, but not the protected methods.
In external email, Gilad Bracha has told me that he took great pains to
emphatically state that interfaces semantically DO NOT extend Object, even
though that how it is implemented by the JVMS. Therefore, i.finalize() should
not compile, because I does not have a member named finalize(). My little trick
of using a nested class (with full access to any protected members of the
enclosing type) should not have succeeded; but because it did, the result is a
VerifyError.

You may also want to link this bug to bug 4479715, where it is argued that the
JLS permits things like "int clone();" in an interface. I'm not sure I like the
implications of that, as it would make an interface non-instantiable, but it
results from a literal reading of the JLS.

These gaps in the JLS, coupled with behavior of javac which contradicts the JLS,
makes it difficult to implement a 3rd-party compiler with the correct behavior.
(Review ID: 135444)
======================================================================