Hello Sachith, I'm Joachim and I'm the one who implements other parts of the JAXB2 specification for Castor in parallel to you.
I suggest a slightly different approach for the implementation... The current implementation of Castor generates ClassDescriptor (itself or of sub classes) instances containing the mapping information which are used in the marshall and ummarshall process. I would suggest, that your schema generator will use this ClassDescriptor instances as information source and that the annotation reading part generates ClassDescriptor's. If today's ClassDescriptor's need to be extended to hold the information required by JAXB we should extend them. Why do I think that this should be the way: *) because the reading of annotations is needed for schema generation AND (un-)marshall *) schema generation is required at 'generation time' and at '(un-)marshall time' *) it is Castor style to do so and if it doesn't hurt I would like to stick to Castor style Any comments / additions? Is this ok for you? Joachim
