Hi Alan. The reason we didn't go this route is that we want to be able
to create the RMI Registry on a random port which is not one of the
reserved ports (see TestLibrary.FIXED_PORT_MIN/MAX). To do this, we use
'new ServerSocket(0)', which creates a socket on a random port, then
close the port, check if it's a 'reserved' port, and return the port
number if it's not. Since we have to go through this process anyway,
the current implementation (see webrev05:
http://cr.openjdk.java.net/~dmocek/7142596/webrev.05) already works and
has broader applicability (e.g. this also supports getting a random port
for tests which require a socket rather then hard-coding port numbers
into tests, although this change isn't in this webrev).
Darryl
On 07/14/2012 03:31 AM, Alan Bateman wrote:
On 14/07/2012 00:09, Stuart Marks wrote:
There is at least one test case that wants to create two registries
within the same JVM. The first call to createRegistry(0) will usually
succeed. The second call from the same JVM will throw an
ExportException. So, we catch this and retry using a random port
instead to create the second registry. If *that* fails we give up at
that point instead of retrying repeatedly.
Okay, so it's the export of the object that is the issue because the
objID is based on the argument rather than the actual port. In that
case would it make sense to switch to the factory method that takes
the RMIServerSocketFactory as argument, something like:
static class ServerSocketFactory implements RMIServerSocketFactory {
public ServerSocket createServerSocket(int port) throws
IOException { return new ServerSocket(0); }
}
static class ClientSocketFactory implements RMIClientSocketFactory {
public Socket createSocket(String host, int port) throws
IOException { return new Socket(host, port); }
}
Registry registry = LocateRegistry.createRegistry(0, new
ClientSocketFactory(), new ServerSocketFactory());
-Alan.
