O/R mapping and datastore identity

Question

Hi all,
we are currently trying to map an existing database schema to JDO. We have
not been able to map "correctly" to the existing schema in two
different cases. Any input that you may have to us on this is highly
appreciated.
Inheritance with indicators
In one part of the model, we use a single character indicator to
distinguish between which concrete class is to be instantiated. We
thus have a table A with columns:
type varchar2(1)
id number(10)
Other tables relate to this table. Some tables can only relate to some
types. To ensure that the relational model
expresses the correct constraints about which objects can relate to
one another, all our foreign keys include both the type and the
id. Still, the id is taken from a sequence and should be considered
datastore identity.
When mapping this to Kodo JDO, we cannot make (type,id) the primary
key and have datastore identity. We have had to compromise and use
just the id as the primary key (which is in fact unique "by
convention"). This in turn means that we cannot use (type,id) as the
reference from other tables (e.g. B), since a lookup from A to B then
includes the A table in the SQL fired just to get the correct type.
Are we missing something here? I would assume that this would be a
normal problem if using datastore identity and flat inheritance
mapping. And in fact, Kodo must know the value of the type for any A
object, since it has the instance and can simply relate the class to
the type through the metadata (however, this only partly solves the
problem). Is there a workaround we can use? Can we extend the primary
key/reference support to support this ourselves?
Sub-objects
In another part of our model, we have a table C where each record can
have a number of records in a table D:
C:
id number(10)
D:
id number(10)
Cid number(10) (references C.id)
In the database, the primary key for D is (id, Cid). id is drawn from
a sequence and is thus unique "by convention". Other tables relate to
D through (id, Cid), thus we will need a unique index on this.
Again, we have had to make id the primary key of D and let all
relations to it only join on the id field when doing the mapping.
We are reluctant to introduce new indexes in the database since this
will be harmful to our import performance.
Is there a workaround for this, or are we simply not doing this
correctly??

Answer 1

Inheritance with indicators
---------------------------If the type column is part of the primary key, I would suggest using application
identity. You can use extensions like sequence-assigned to still have the id
value auto-assigned.
Sub-objects
-----------Again, I would use application identity.