RMI problems with kaffe 0.10.x

[Robert S. Thau]

Godmar Back writes:
 > Also, it is not clear where to look it up in.  Currently, it only
 > looks for a static field with a given name (serialVersionUID or some
 > such.)  Now if that's everything it's supposed to do, why is it a native
 > method?  They could have used the reflection API instead, it seems.  Or 
 > is it for security reasons that they didn't?

It's supposed to use the value of serialVersionUID if the class has a
field of that name, and compute a default value if serialVersionUID is
not explicitly specified.  See the section titled "Stream Unique
Identifiers" in the Java Object Serialization spec (included with the
JDK 1.1 docs), which specifies what gets hashed to produce the default
value. 

(If this is what's going wrong, then specifying an explicit
serialVersionUID for all the serializable classes involved in the RMI
test may work around the problem.  This is a god idea in the long run
anyway, to control interoperability in the face of version skew ---
see the section on versioning in the Object Serialization spec).

rst