Re: stack overflow & null pointer exceptions

[Godmar Back <gback@cs.utah.edu>]

> 
> > From: Godmar Back <gback@cs.utah.edu>
> > Date: Wed, 7 Apr 1999 15:26:15 -0600 (MDT) 
> > 
> >  Hi,
> > 
> > it appears that gcj does not yet implement stack overflow errors properly:
> > the following program segfaults.  
> > 
> > Apropos segfaulting: it also appears as though NullPointerExceptions aren't
> > properly caught (?).  See the second test case.
> 
> NullPointerExceptions and ArithmeticExceptions are now caught on x86
> Linux and SPARC Solaris.  I'm still considering what to do about stack
> overflows.
> 

Andrew, did you try the test program I sent?
I updated and tried it on a Linux 2.2.5/glibc2.0.7 RH 5.2 box, and it 
still doesn't quite seem to work.

Here it is again:

/*
 * test that caught null pointers exceptions in finalizers work correctly
 * and that local variables are accessible in null pointer exception handlers.
 */
import java.io.*;

public class NullPointerTest {

    static String s;

    public static void main(String[] args) {
          System.out.println(tryfinally() + s);
    }

    public static String tryfinally() {
        String yuck = null;
        String local_s = null;

        try {
            return "This is ";
        } finally {
            try {
                local_s = "Perfect";
                /* trigger null pointer exception */
                String x = yuck.toLowerCase();
            } catch (Exception _) {
                /* 
                 * when the null pointer exception is caught, we must still
                 * be able to access local_s.
                 * Our return address for the finally clause must also still
                 * be intact.
                 */
                s = local_s;
            }
        }
    }
}


/* Expected Output:
This is Perfect
*/

When run with the .java frontend, I see:

gback@peerless [24](/tmp) > gcj --main=NullPointerTest NullPointerTest.java
gback@peerless [25](/tmp) > ./a.out
Perfectnull
° ... lots of garbage ...

When compiling the .class file only, I see:

gback@peerless [26](/tmp) > javac NullPointerTest.java 
gback@peerless [27](/tmp) > gcj --main=NullPointerTest NullPointerTest.class 
gback@peerless [28](/tmp) > ./a.out
This is Perfect
° ... lots of garbage ...

This may not be a bug in the exception handling code; it looks more like
bugs in the java parser and in the gcj library (?)

	- Godmar