Name: diC59631 Date: 01/19/99


This problem has been verified on JDK1.1.7beta2 for
Digital Unix 4.0c from DEC, on JDK1.1.5K on SunOS 5.5,
and on Linux_JDK_1.1.5_v7_sbb_unified. It does not seem
to exist on Sun JDK/Symantec JIT on Win32.

If a thread is created and throws an exception in its
constructor, or if the constructor returns but the thread
is never started, the thread is never garbage collected
(even when invoking System.gc() explicitly). The first case is
especially egregious, because there is no valid reference to
do anything to the thread with.

Simple code triggering this problem:

public class test {
  public static void main(String[] args) {
    Broken foo=null;
    while(true){
      try{
foo = new Broken();
foo.start();
      }
      catch (Exception e){
System.err.println("caught exception");
      }
      System.gc();
    }
  }
}

class Broken extends Thread{
  Broken()
    throws Exception {
      sleep(1000);
      throw new Exception("hi");
  }

  public void run(){
  }
}

Attaching a debugger to this code will show an ever
increasing list of zombie threads:
 9. (Broken)0x6e314698 Thread-5 zombie
10. (Broken)0x6e303c68 Thread-6 zombie
11. (Broken)0x6e314658 Thread-7 zombie
12. (Broken)0x6e314620 Thread-8 zombie
13. (Broken)0x6e3146c0 Thread-9 zombie
14. (Broken)0x6e314238 Thread-10 zombie
15. (Broken)0x6e3146f0 Thread-15 zombie
16. (Broken)0x6e313fc8 Thread-16 zombie

The problem will also be encountered if the
exception is not thrown and the start() call is
never made.
(Review ID: 41109)
======================================================================