FULL PRODUCT VERSION :
java version "1.7.0-ea"
Java(TM) SE Runtime Environment (build 1.7.0-ea-b14)
Java HotSpot(TM) Client VM (build 1.7.0-ea-b14, mixed mode, sharing)

java version "1.6.0_01"
Java(TM) SE Runtime Environment (build 1.6.0_01-b06)
Java HotSpot(TM) Client VM (build 1.6.0_01-b06, mixed mode, sharing)

java version "1.5.0_12"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0_12-b04)
Java HotSpot(TM) Client VM (build 1.5.0_12-b04, mixed mode, sharing)

ADDITIONAL OS VERSION INFORMATION :
Microsoft Windows XP [Version 5.1.2600]
Linux myhost 2.6.20-16-generic #2 SMP Thu Jun 7 20:19:32 UTC 2007 i686 GNU/Linux

EXTRA RELEVANT SYSTEM CONFIGURATION :
Windows XP: Single core, single CPU
Linux 2.6.20: Dual core, single CPU

A DESCRIPTION OF THE PROBLEM :
The contract of either Thread#interrupt(), Thread#isInterrupted() or Thread#isAlive() seems to be broken.

My intuitive assumption as that it is the contract of Thread#interrupt() that is broken, seeing as how a related bug has been reported before with bug id 4082705. The contract was changed in from JDK5 to JDK6, so that there are no longer any guarantee what happens if the thread is not alive.

The contracts of interrupt() and isInterrupted() seems to guarantee that isInterrupted() must return true if interrupt() was called on the same thread when that thread was alive. The contract of isAlive() seems to guarantee that a thread must be alive the entire time between two calls to isAlive() that both return true.

Thus, I would expect isInterrupted() to always return true after I call interrupt() between two calls to isAlive() that both returns true. That is not the case on either version that I tested.

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Compile and run the source code.

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
# THE FOLLOWING OUTPUT SHOULD BE OBSERVED:

Testing JDK7 bug on version 1.7.0-ea
0
1000
2000
3000
(i*1000 printed ad infinitum...)
ACTUAL -
# OUTPUT ON LINUX

Testing JDK7 bug on version 1.6.0_01
0
All interrupt of thread 649 are belong to Sun!

Testing JDK7 bug on version 1.5.0_12
0
All interrupt of thread 650 are belong to Sun!



# OUTPUT ON WINDOWS XP

Testing JDK7 bug on version 1.7.0-ea
...
All interrupt of thread 323713 are belong to Sun!

Testing JDK7 bug on version 1.6.0_01
...
All interrupt of thread 293597 are belong to Sun!

Testing JDK7 bug on version 1.5.0_12
...
All interrupt of thread 156847 are belong to Sun!


REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
public class InterruptBeforeRunTest {
    public static void main(String[] args) {
        System.err.println("Testing JDK7 bug on version " + System.getProperty("java.version"));

        int i = 0;
        while (true) {
            if (i % 1000 == 0) {
                System.err.println(i);
            }
            Thread t = new Thread();
            t.start();
            boolean aliveBefore = t.isAlive();
            t.interrupt();
            boolean aliveAfter = t.isAlive();
            if (aliveBefore && aliveAfter && !t.isInterrupted()) {
                System.err.println("All interrupt of thread " + i + " are belong to Sun!");
                break;
            }
            i++;
        }
    }
}
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
No workaround found. I thought of having a scheme like
class MyThread extends Thread {
  private volatile running = false;
  public void run() {
    running = true;
    // do your stuff
    running = false;
  }
  public boolean isRunning() {
    return running;
  }
}
and changing the isAlive() calls to isRunning() calls. This was tested on JDK6 on the LINUX machine, but it turns out that even then I observe that interrupts are missing. (The code for how to provoke this must be tuned per machine, and can be provided upon request)