Name: ks84122 Date: 05/22/2002


###@###.### 2002-05-22

The following two examples cause jdb to throw NullPointerException
in JDK 1.4/1.4.1 when printing locals:

EXAMPLE 1.

public class badscope {
    public static final void main(String args[])
    {
        // this line should be here
        try {
            System.out.println("hi!");
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            System.out.println("done");
        }
    }
}

Steps to reproduce:
1) javac -g badscope.java
2) jdb badscope
3) inside jdb: > stop at badscope:7
4) inside jdb: run
5) inside jdb: > next
6) inside jdb: > locals

Local variables:
Internal exception:
java.lang.NullPointerException
        at com.sun.tools.example.debug.tty.Commands.printVar(Commands.java:1434)
        at com.sun.tools.example.debug.tty.Commands.commandLocals(Commands.java:1472)
        at com.sun.tools.example.debug.tty.TTY.executeCommand(TTY.java:247)
        at com.sun.tools.example.debug.tty.TTY.<init>(TTY.java:539)
        at com.sun.tools.example.debug.tty.TTY.main(TTY.java:830)

EXAMPLE 2:

public class badscope {
    public static final void main(String args[])
    {
        try {
            int i = -1; // line 5
            System.out.println("hi!");
        } catch (Exception e) { // line 7
            e.printStackTrace();
        } finally {
            System.out.println("done"); // line 10
        }
    }
}


Steps to reproduce:
1) javac -g badscope.java
2) jdb badscope
3) inside jdb: > stop at badscope:7
4) inside jdb: run
5) inside jdb: > next
6) inside jdb: > locals

Method arguments:
args = instance of java.lang.String[0] (id=297)
Local variables:
Internal exception:
java.lang.NullPointerException
        at com.sun.tools.example.debug.tty.Commands.printVar(Commands.java:1434)
        at com.sun.tools.example.debug.tty.Commands.commandLocals(Commands.java:1472)
        at com.sun.tools.example.debug.tty.TTY.executeCommand(TTY.java:247)
        at com.sun.tools.example.debug.tty.TTY.<init>(TTY.java:539)
        at com.sun.tools.example.debug.tty.TTY.main(TTY.java:830)


The original description from Apple follows. They claim this behavior is
due to incorrect variable slot allocation. However, the example 1) above
demonstrates the same problem without overlapping variable slots. So, it
looks like a debugger has problems understanding scopes.

Original Description from Apple:
================================

A properly-formed try-catch-finally block may get an incorrect local variable table when compiled with javac, resulting in debugger or VM crashes.

Consider the following code and its line number and local variable tables from `javac -g`:

public class badscope {
    public static final void main(String args[])
    {
        try {
            int i = -1; // line 5
            System.out.println("hi!");
        } catch (Exception e) { // line 7
            e.printStackTrace();
        } finally {
            System.out.println("done"); // line 10
        }
    }
}

Line numbers for method void main(java.lang.String[])
   line 5: 0
   line 6: 2
   line 7: 10
   line 8: 16
   line 9: 21
   line 10: 27
   line 12: 44

Local variables for method void main(java.lang.String[])
   java.lang.String[] args pc=0, length=45, slot=0
   int i pc=2, length=8, slot=1
   java.lang.Exception e pc=17, length=27, slot=1

int i and Exception e both share slot 1, because the two variables are never alive at the same time. This is not incorrect.

The local variable table info generated by `javac -g` states that int i is valid only inside the try block, and that Exception e is valid inside the catch and finally blocks.

When the code is run and the catch clause isn't executed and the program is stopped at line 10, the jdb `locals` command will fail, possibly crashing the VM. slot 1 has the value 0xffffffff, the last value of i before it went out of scope. GetLocalVariableTable() claims that Exception e is in slot 1 at line 10, even though slot 1 has never contained a valid Exception reference. Bad things happen when the debugger innocently calls GetLocalObject(slot 1) because 0xffffffff isn't an object reference.

In Java 1.3.1, the VM crashes during GetLocalObject(). In Java 1.4.0, jdb throws a NullPointerException but the VM doesn't crash. These occur on both Mac OS X and Solaris (don't know version).



(Review ID: 146869)
======================================================================