ANOTHER Java NIO Socket Bug??

Question

I think I'm being really dense here but for the life of me, I can't get this code to work properly. I am trying to build a NIO socket server using JDK 1.4.1-b21. I know about how the selector OP_WRITE functionality has changed and that isn't my problem. My problem is that when I run this code:
if (key.isAcceptable()) {
     System.out.println("accept at " + System.currentTimeMillis());
     socket = server.accept();
     socket.configureBlocking(false);
     socket.register(selector, SelectionKey.OP_READ | SelectionKey.OP_WRITE);
if (key.isWritable()) {
     SocketChannel client = (SocketChannel)key.channel();
     System.out.println("write at " + System.currentTimeMillis());
if (key.isReadable()) {
     SocketChannel client = (SocketChannel)key.channel();
     readBuffer.clear();
     client.read(readBuffer);
     System.out.println("read at " + System.currentTimeMillis());
}the isWritable if statement will always return (which is fine) and the isReadable if statement will NEVER return (which is most certainly NOT FINE!!). The readBuffer code is there just to clear out the read buffer so isReadable is only called once per data sent.
This SEEMS to be a bug in how the selector works? I would expect to see isReadable return true whenever data is sent, but that is not the case. Now here is the real kicker ;) Go ahead and change this line:
socket.register(selector, SelectionKey.OP_READ | SelectionKey.OP_WRITE);to this:socket.register(selector, SelectionKey.OP_READ);And now it appears that isReadable is running as expected?! To let people run this code on their own, I have uploaded a copy of the entire java file here:
http://www.electrotank.com/lab/personal/mike/NioTest.java
Please forgive the code, it's just the smallest test harness I could make and much of it is lifted from other posts on this forum. You can test this by using Telnet and connecting to 127.0.0.1:8080. You can test the isReadable piece by just typing in the Telnet window.
Someone else has listed something as a bug in the Bug Parade, but the test case is flawed:
http://developer.java.sun.com/developer/bugParade/bugs/4755720.html
If this does prove to be a bug, has someone listed this already? Is there a nice clean workaround? I'm getting really desperate here. This bug makes the NIO socket stuff pretty unusable. Thanks in advance for the help!!
Mike Grundvig
[email protected]
Electrotank, Inc.

Answer 1

Yeah, isReadable crashed for me too.
My solution was to not call it. I set up two selectors for the two operations I wanted notifying (accept and read) and used them independently in different threads. The accept thread passes them over to the read thread when they're accepted.
This way I don't need to call isReadable since it is bound to be readable otherwise it wouldn't have returned, as read is the only operation I'm being notified about.
--sam