When the test fails sometimes the output is truncated, or even absent, as the test's sub-process does not complete. This may happen if the client-side encounters an issue and leaves the non-daemon server-side thread blocked in accept. The test should close the server socket when the client-side completes. This will enable the "real" error to be seen. [ thanks to Chris Yin, for finding this! ]
http://cr.openjdk.java.net/~chegar/8211920.00/ While here, and since the test has been seen to fail occasionally, in the more distant past, I took the liberty of changing the binding of the server socket. It now follows the more stable pattern of using a specific interface. -Chris. P.S. the test remains on the ProblemList since it still fails, but now we can see why. 8211921 tracks the underlying issue in java.lang.invoke
