It appears that the JDK is generating incorrect
repository ids in a very small number of cases. We
have only found this because we started generating
repository ids ourselves this release. The possible
bug in the JDK (and the RI ORB) means that we cannot
currently pass the CTS (TCK_J2EE), hence the urgency
to determine whether this is indeed a bug or not.

See comments for location of a test case.

Using this testcase, we found the SVUID computations are:

1.5 com.sun.corba.se.impl.io.ObjectStreamClass is:
        -3686524756480333392
1.4.2_04 com.sun.corba.se.internal.io.ObjectStreamClass is:
        3013006116053782630
1.5 java.io.ObjectStreamClass is
        3013006116053782630
1.4.2_04 java.io.ObjectStreamClass is
        3013006116053782630

These results were obtained for J2SE 5.0 b51. It's possible
that this got caught in a later 5.0 build.