Does NHibernate support mapping an abstract class to disparate tables using composite ids?

Question

I have a complex, 3NF database being provided to me for a particular project. I would like to avoid using a class-per-table domain design. Rather, I would like to model my domain objects after how they are used from a conceptual business perspective.

The rub is how to properly persist this information. I know I can go the ADO route, but I'd like to take a stab at using NHibernate, having used it successfully on other projects with more flexible data stores.

So, I need to know if NHibernate will support the following scenario:

I have a conceptual object known as a ProjectStatus, which is comprised of a handful of date stamps for various activities along with some notes about the status. All of the data that comprises the ProjectStatus comes from 2 or more tables. There is no ProjectStatus table.

I know I can do a union-subclass in my NH mapping to get this to work, but...

One of the tables that holds the bulk of the information I need has a composite id (two PK fields that together make up the identity signature). I know NH supports composite ids as well, but how would I go about mapping my the union on the composite key? Do I need to specify a composite key underneath the union-subclass section?

The dba has refused to budge on her near-neurotic 3NF data model, so I'm stuck on that front. If I have to drop to ADO for ease/speed of development, so be it, but I'm hoping NH will rise above...

Answer 1

I gather that you have a base-class and two sub-classes that you want to map to two sub-class tables using the table-per-concrete-class mapping strategy. The two sub-class tables need to declare the same primary keys, and the primary keys need to be mapped in the base-class mapping. You have to declare the id or composite-id in the base-class mapping, so that you can ask NHibernate for an object of the base-class type with the given ID or composite ID. You can't declare them in the subclass-mapping, and you certainly can't declare them to be different for each subclass type (that's what declaring them in the subclass mapping would permit you to do), because from a strongly*-typed object-oriented perspective, that would be leaving the realm of sanity.

*Strong typing, to me, means that variables are statically typed (the type of each variable is known to and enforced by the compiler) and objects are strictly typed (the type of each object is known to and enforced by the runtime).