FULL PRODUCT VERSION :
java version "1.5.0_05"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_05-b05)
Java HotSpot(TM) Client VM (build 1.5.0_05-b05, mixed mode, sharing)


FULL OS VERSION :
Linux Fedora Core 3 kernel 2.6.12-1.1372_FC3


A DESCRIPTION OF THE PROBLEM :
Adding a double or long constant to a class and forcing a redefiniton of it through JDI redefineClasses picks up a seeming random value of the constant.
The correct behaviour is seen for 32 bit constants. JRE1.5.0-b64 does not exhibit this bug.

This makes it difficult to rely on class redefinition for fast prototyping.

THE PROBLEM WAS REPRODUCIBLE WITH -Xint FLAG: Yes

THE PROBLEM WAS REPRODUCIBLE WITH -server FLAG: Yes

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Compile the source code under JDK 1.5.0 and execute it using JRE 1.5.0_05

Then change the line System.out.println(a); to System.out.println(a*2L); or (a*2.0); and force a class redefinion through JDI


EXPECTED VERSUS ACTUAL BEHAVIOR :
Expected: 1000 should be printed once per second
Actual:
1000
1000
1000
1000
==JDIHarness: agen.bin.Test will be replaced
==JDIHarness: Suspending JVM
==JDIHarness: Uploading 1 classes
==JDIHarness: Resuming JVM
-6624593188399085616
-6624593188399085616
[the number printed following redefiniton appears random, and subsequent redefinitions change this number and sometimes get it correct]
REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
public class Test {
    public static void main(String argv[]) throws InterruptedException {
while (true) {
doStuff(1000);
Thread.sleep(1000);
}
    }
    private static void doStuff(int a) {
System.out.println(a);
    }
}
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
Revert to JRE1.5.0 or (possibly) perform the redefinition twice.