The package private field "threadLocals" of java.lang.Thread
appears to be getting smashed to null by the classic VM.

This field is used by java.lang.ThreadLocal to implement
thread local variables.

-----
class Oops {
    static public void main(String[] args) {
        new ThreadLocal().get();
    }
}
-----
% /usr/local/java/jdk1.3/solaris/bin/java -classic Oops
Exception in thread "main" java.lang.NullPointerException
        at java.lang.ThreadLocal.get(ThreadLocal.java:83)
        at Oops.main(Oops.java:3)
-----

This will effect many things including some String constructors:

class Oops {
    static public void main(String[] args) {
        try {
            new String(new byte[5], 0, 5, "UTF8");
        } catch(java.io.UnsupportedEncodingException e) {
            System.err.println(e);
        }
    }
}
-----
% /usr/local/java/jdk1.3/solaris/bin/java -classic Oops
Exception in thread "main" java.lang.NullPointerException
        at java.lang.ThreadLocal.get(ThreadLocal.java:83)
        at java.lang.String.getBTCConverter(String.java:124)
        at java.lang.String.<init>(String.java:403)
        at Oops.main(Oops.java:4)
-----

Does not fail under HotSpot. Only tested under Solaris.

% /usr/local/java/jdk1.3/solaris/bin/java -hotspot Oops
%
% /usr/local/java/jdk1.3/solaris/bin/java -classic -version
java version "1.3.0"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.3.0-H)
Classic VM (build 1.3.0-H, green threads, nojit)
-----

VM is deemed at fault since this field is initialized to a map:

Thread.java:159: Map threadLocals = Collections.EMPTY_MAP;

And only set in two places (both to maps):

ThreadLocal.java:86: map = ourThread.threadLocals = new HashMap(INITIAL_CAPACITY);
ThreadLocal.java:106: map = ourThread.threadLocals = new HashMap(INITIAL_CAPACITY);

Also Thread has definitely been initialized, in fact it is
the current Thread.

robert.field@Eng 1999-09-28