The specification for {Server,ServerSocket}.setReuseAddress
currently states :-

     * The SO_REUSEADDR socket option affects the <code>bind()</code>
     * operation and serves 2 main different purposes:
     * <UL>
     * <LI>Allows a listening server to start and <code>bind</code> its port
     * even if previously established connections exists that use this port
     * as their local port. This typically happens when a ServerSocket is
     * started and listen on that port, accepts an incoming connection and
     * spawn a child process to handle the socket. Then the listening server
     * terminates but the child continues to service the client. At that, to
     * be able to restart the listening ServerSocket, the SO_REUSEADDR option
     * needs to be turned on.</LI>
     * <LI>Allows multiple instances of the same server to be started on
     * the same port, as long as each instance binds a different local
     * IP address.</LI>
     * </UL>


This wording has proved difficult for those developing compatibility
tests - the paragraph "Allows multiple instances ..." is in fact
not correct as the SO_REUSEADDR socket option isn't required to
co-exist sockets bind to different local addresses.

Thus we need to :-

1. Re-word the spec along the lines of the classic Stevens explaination.

2. Remove section of spec dealing with "multiple instances" - this is
esp. true for TCP sockets as there is too much variation in OS behaviour.

3. Properly specify that SO_REUSEADDR should be enable prior to
binding in order to bind to a socket address with a connection that still
in time wait.

4. Specify the default setting.

5. Indicate that the behaviour as a result of enabling/disable after
the socket is bound is not defined. This is OS specific so it cannot
be specified.

###@###.### 2001-10-02