final fields should be handled by FieldSerializer if setFieldsAsAccessible is set to true about kryo HOT 7 CLOSED

From [email protected] on April 05, 2010 14:15:39

The value of a final field can be inlined in method bodies. Changing the value of a
final field is possible through reflection, but should never be done because the
changes may not be seen by all code accessing the field.

Status: WontFix

from kryo.

From martin.grotzke on April 05, 2010 14:19:54

Ok, that's true. But in my case the final field is only set when an object is deserialized in another jvm, therefore there has not
been any code that has seen this value.

from kryo.

From martin.grotzke on April 05, 2010 14:21:33

In either case, it would be good to have the possibility to enforce this behavior. Do you see any problems with this?

from kryo.

From [email protected] on April 05, 2010 14:39:42

It appears JLS 17.5.3 details when it is OK to modify final fields and the associated
problems:
http://java.sun.com/docs/books/jls/third_edition/html/memory.html#17.5.3
Based on this, I now think Kryo should serialize final fields. I don't think a
setting is necessary.

Status: Accepted

from kryo.

From martin.grotzke on April 05, 2010 14:49:23

Great!

Though, I just noticed that inner classes are not deserialized correctly with this setting. E.g. for a class

public static class Container {
    private final Body _body;
    public Container() {
        _body = new Body();
    }
    class Body {
    }
}

the enclosing Container instance is not set on the Body instance after deserialization.
The test and serialization/deserialization looks like this:

public void testInnerClass() throws Exception {
    final Container container = new Container();
    final Container deserialized = deserialize( serialize( container ), Container.class );
    assertDeepEquals( deserialized, container );
}

protected byte[] serialize( final Object o ) {
    return new ObjectBuffer(_kryo).writeObject( o );

}

protected <T> T deserialize( final byte[] in, final Class<T> clazz ) {
    return new ObjectBuffer( _kryo ).readObject( in, clazz );
}

from kryo.

From [email protected] on April 06, 2010 01:12:00

Final field support is checked in, r96.

Being unable to instantiate a non-static member class is separate from final field
support. To do this, Kryo would need to pass the reference to the enclosing instance
to Constructor#newInstance(). There is currently no mechanism to obtain the enclosing
instance, and it doesn't appear easy to add.

Status: Fixed

from kryo.

From [email protected] on April 06, 2010 19:47:17

Issue 6 has been merged into this issue.

from kryo.