[24] Local instantiation in static context allowed by ECJ

[jarthana]

The following code that contains an instantiation of a local in a static context is allowed by the compiler. This was allowed by earlier JDK24s, but as of ea34, is rejected).

public class X {
	public void main(String[] args) {
		System.out.println(foo() != 0);
	}
	static int foo() {
		class Local {
			int value = 0;
			class Local2 {
				public static int bar() {
					return new Local().value;
				}
			}
		}
		return Local.Local2.bar();
	}
}

@stephan-herrmann I haven't yet looked at the spec, but it appears that we should be rejecting this code?

I also did some investigation and found this code to be involved (AllocationExpression, line 371:

MethodScope enclosingMethodScope = local.scope.enclosingMethodScope();

First of all, I don't think I understand the !enclosingMethodScope.isStatic in the check in line 372. The second part I don't understand is why we are ignoring the scope of the AllocationExpression itself.

[jarthana]

Here's another related testcase where we are rejecting but Javac is accepting:

public class X {
    static int INT_FIELD = new Object() {
        class Local {
            int value = 0;
            class Local2 {
                public int bar() {
                    return new Local().value;
                }
            }
        }
    }.hashCode();
}

[stephan-herrmann]

Thanks, @jarthana I'll go back to studying the papers 😄

[stephan-herrmann]

Regarding the first example I see two violations, actually:

Allocation new Local():

JLS §15.9.2:

If C is an inner local class, then:

• If C occurs in a static context, then i has no immediately enclosing instance. Let S be the nearest static method declaration, static field declaration, or static initializer that encloses the declaration of C. If the nearest static method declaration, static field declaration, or static initializer that encloses the class instance creation expression is not S, then a compile-time error occurs.

• C = Local
• S = X.foo()
• "nearest static ... that encloses the class instance creation expression" = Local2.bar()
• X.foo() != Local2.bar() => ERROR

Field access new Local().value

deleted

I was confused trying to apply a rule for simple expression name, whereas what we have is a field reference.

=> One error is enough.

[stephan-herrmann]

While fixing this I get into a conflict with tests like org.eclipse.jdt.core.tests.compiler.regression.LocalEnumTest.test077():

public class X {
	
	public static void main(String[] args) {
		enum E {
			A {
				void bar() {
					new M();
				}
			};
			abstract void bar();
			
			class M {
				M() {
					System.out.println("SUCCESS");
				}
			}
		}
		E.A.bar();
	}
}

Hitherto a conforming test, we now start to report

	new M();
	^^^^^^^
Cannot instantiate local class 'E.M' in a static context

javac >= 16 complains similarly ("no enclosing instance"). (before 16 enums were not allowed to be local).

I think the error is correct, because the body of A is an implicit anonymous class inside the initializer of the implicit static field A. So we are in a static context indeed, where the member class M cannot be instantiated without qualification.

@jarthana @mpalat please speak up if you disagree 😄

[stephan-herrmann]

Fixed by #3699