From ###@###.### Thu Jan 18 11:49:01 1996
To: Arthur.Vanhoff@Eng
From: Larry Podmolik <###@###.###>
Subject: Bug in javac/appletviewer

Arthur,

[I sent a message similar to this yesterday, but I don't think you
received it. We just switched to a T1 line and our mail has been
flaky as a result...]

I think I've uncovered a strange bug in javac. It lets pass some
questionable code that seems very questionable to me, and that produces
different results in the appletviewer and in Netscape.

Here is the program:

        1 import java.applet.Applet;
        2 class Base {
        3 Base(int x) { }
        4 }
        5 class Derived extends Base {
        6 public Derived() {
        7 super(_x = 5);
        8 }
        9 int _x;
        10 }
        11 public class TestApplet extends Applet {
        12 public void init() {
        13 new Derived();
        14 }
        15 }

This compiles without complaint using javac (JDK b2, Solaris 2.4) and
runs fine in the applet viewer. But shouldn't this code be illegal?
On line 7, I'm initializing the Derived member _x before executing
the Base constructor. In C++ terms, the Derived object doesn't have
its "shape" yet so it seems this should be disallowed.

When this applet is run in Netscape 2.0b5, I get a java.lang.VerifyError
when I try to run this applet -- the first time. If I re-load the
applet things appear to work fine. Strange, huh?

P.S. This was the scenario that led to the VerifyError I wrote to you
about. The person you referred me to never responded to my email, so
I continued to look into the problem.

-- Larry