The round trip should be at the Ecore to UML superstructure and UML to
Ecore level and mixing model information inside java code is not
recommended.
Not recommended because if you refactor your code then you loose your model
If you annotate you java code classes then your get a dirty java code :-)
The java detection using native JDT which unify Ecore/UML is the only
possible solution in order to cover agile iteration. The omondo
architecture is based on Java and UML Ids both of them need are used at
different stages.