Given,
Abstract ClassA.
- id property
- property-1
- property-2
- many-to-one to ClassX
ClassB extends ClassA
- id property
- property-3
ClassC extends ClassA
- id property
- property-4
ClassX
-id property
-property-5

Above situation is mapped using three hibernate hbm xml files:
- one for abstract ClassA, mapped with attribute abstract="true".
- one for concrete ClassB, mapped with union-subclass element and attribute extends="ClassA".
- one for concrete ClassC, mapped with union-subclass element and attribute extends="ClassA".

ClassA's mapping file contains a mapping for property-1 and 2, mapped to e.g. COLPROP1 and COLPROP2,
and a many-to-one mapping to Class/Table X.
ClassB's mapping file contains a mapping for property-3, mapped to COLPROP3.
ClassC's mapping file contains a mapping for property-4, mapped to COLPROP4.

Hibernate generates the right schema:
- One table for ClassB with four columns: for id, property-1, property-2, property-3 and a column for holding the key to table for ClassX
- One table for ClassC with four columns: for id, for property-1, property-2, property-4 and a column for holding the key to table for ClassX
- One table for ClassX with two columns: for id, for property-5

However I want to let Hibernate generate as many indexes and constraints as possible, so that my schema is as complete as possible.
Therefore I want to add where appropriate unique indexes (over multiple properties/columns) and foreign-key contraints.

Given that I have a union-subclass situation where some of the properties and associations are specified in the abstract base class mapping.

How and where should I specify a unique index e.g. "I1" on properties property-1 (defined in ClassA's hbm file) ánd property-2 (defined in ClassB's hbm file) ?
How and where should I specify the foreign key for the many-to-one association specified in the abstract base class ?

In my opinion the above is only possible if I remove property-1 and the many-to-one association from abstract ClassA's mapping file, and duplicate property-1 and the many-to-one in each of the child mapping files. Only then will I be able to specify the unique index and the foreign-key.

Am I right, or is there another way I can achieve the above without duplicating properties and associations that in fact belong in the abstract parent mapping ?

Any advice would be welcome.
Thanks,
EDH