To: ###@###.###, ###@###.###
CC: ###@###.###
Subject: interfaces extend Object?

javac's implementation of interfaces doesn't quite follow the JLS
description of how interfaces work, and the result isn't quite
seamless. Specifically, javac acts internally as if an interface
(that doesn't extend another interface) inherits from Object. But the
JLS 9.2 says that the interface "implicitly defines" those public
methods from Object, and doesn't extend Object. Here are the
consequences of javac's implementation strategy:

(1) Although the JLS says such methods are "implicitly" defined by the
    interface, they don't appear in the class file. That's a bug;
    they are required to appear in the class file as a consequence of
    9.2 and 13.1. Note that fixing this will affect the APIs visible
    through reflection of most interfaces. 4644624

(2) The qualifying type for Object methods accessed from an interface
    should be the interface according to the JLS, but javac outputs
    Object as the qualifier. I now disagree with the comments on
    bugids 4285499 and 4398789 that the qualifying type should be
    Object due to JLS 13.1; the logic there is incorrect because the
    method is not "declared in Object", but rather is "implicitly"
    declared in the interface; the corresponding method from Object
    is separate and has no relationship to the one in the interface.

(3) javac fails to allow some definitions that it should when those
    definitions are incompatible with protected definitions inherited
    from Object. 4479264

(4) Within a class nested within the interface, the protected members
    of Object are visible but shouldn't be. 4526026

The last two points are strictly compiler issues; I can implement a
constraint in the compiler that protected members are never inherited
into an interface to make those problems go away. However, (1) and
(2) will be serious compatibility issues if javac ever actually
implements what the spec requires. I am uncomfortable leaving things
in their current state, so I'd like some definitive decision as to
whether javac or the specification should be changed.

-Neal