I am currently attempting to use the Sun reference implemenation of JAXB to process a large, complex schema consisting of three files: a core industry standard file, a company-specific redefinition of certain types within the industry-standard file, and a third company-specific file that defines additional elements. The most obvious unsupported feature is the <redefine> tag. The second is caused by the sheer size of the industry schema (over 1 meg) - the parser runs out of memory trying to process it. To work around these issues, I have written xslt transformation templates to extract only a sub-set of the schema, and merge the industry-standard section with the company-specific redefintions. This works, but I still wind up with over 4,000 java files, which when put in a jar file, exceeds 11Mb!
If anyone knows of a more efficient JAXB-compliant implementation, please let me know.