This bug is filed for tracking purposed.

For TCK 1.0.1 B04, it has new serious problem: it hangs randomly and results
are not stable at all (for both local and remote mode) on Windows (NT/98).

The reasons are:
- I get the following exception for the fail tests:
   java.net.SocketException: Connection aborted by peer: socket write error
  It happens randomly.
   
or,

- Hang is due to the following error:
  Can not launch jnlp client java.io.IOException: CreateProcess: ...
  (The reason for this is because previous test case has not exit javaws
   cleanly. javaws process is still running.)

Please check asap or file a bug.

Thanks.

--Nathan

============================================================================
Hello All,

These TCK failures observed in Win98/NT are actually due to a very old
issue. The root of the problem lies in a link between a JaWS client an TCK
harness. The TCK provides a special launcher, called JNLPLauncher which is
responsible for running the JaWS client and obtaining the exit code from
JaWS client. Initially, it was decided that JNLPLauncher is implemented
by JaWS developers since it requires interaction with a private part of
JaWS client code, so developers know better how to deal with it. The
problem is described in two JaWS bugs (4396040, 4390297) that was filed
against JaWS 1.0 but it seems they are still not completely fixed.

The details of a problem are following:

The TCK needs a feedback from a JNLP client, and it was decided that the
client developers are responsible for providing an entry point and
feedbacks mechanism called JNLPLauncher. The JNLPLauncher mechanism
consists of two parts:

1. Classes:
com/sun/tck/jnlp/lib/JNLPProcess.java
com/sun/tck/jnlp/lib/CommandJNLPLauncher.java
They reside in a JNLP TCK workspace. The JNLPProcess opens tcp socket,
starts the javaws and waits for feedback to come into the socket.

2. The JaWS client part. When the javaws started, it connect to the First
part and send debug information over the tcp connection. Something like
"Java started", "JNLP launched" and so on. But since the JaWS does
not correctly flush and close the TCP socket before exit, some (sometimes
all) such messages are lost, and the First part hangs or reports that the
JaWS does not finish correctly. This effect arises on Windows platform
only and is a cause for  lot of JaWS failures on Windows platforms,
especially for the latest JaWS builds.

This issue has been raised previously both in e-mails and in our working
discussions with Rene. However, it was decided this is not a stopper and
it's not worth to spent time on fixing it since previously this has been
appeared on Win98 only. Now, it seems the problem arises again and in
winNT too, so we should resolve it somehow.

From our side, Artem has fixed everything he could in those code which
lies in a TCK workspace. But we believe a fix is needed from JaWS client
side too. So, we can suggest one of the following:

- an engineer in JaWS team is fixing JaWS part and TCK part
(CommandJNLPLauncher class) and makes sure everything is working fine on
windows

- Artem takes care about it and makes all necessary updates to JaWS client
part.

Which one would be the most acceptable for all?

Thanks,
Andrey.



nathan.wang@Eng 2001-04-19