On 04/11/2012 08:06 PM, Jim Gish wrote:
Hi,
Please review the following
Description:
The current code uses the fixed port of 22222 and the test fails if
that port is in use. The patch creates the socket on any open port
using new ServerSocket(0) and then determines the port in use with
getLocalPort() on the socket. That port is then used when creating
the client. This ensures the test won't fail with "address already in
use".
Also, all sockets are closed when done. (Would appreciate comments on
whether I have engaged in overkill here, or not.)
Patch:
http://cr.openjdk.java.net/~jgish/7116200/webrev.00/
Tests run: jdk_nio3 -- all pass.
Thanks,
Jim Gish
aaaaah,
there is no guarantee from the memory model that the main thread will
see the field 'so' initialized
even if the thread client has executed the method run().
And there is a kind of a race between the main thread and the thread
Client,
the main thread calls closeSocket() while the method run() may have not
finished.
It's not a clear race because the socket IOs are used as a kind of
synchro but
it's really hairy.
Rémi
