It's certainly true that it's possible with JPA, but, at least when using Hibernate as the JPA provider, if you don't explicitly fiddle with the mappings, you get an intermediate table by default; at least, I've always had to add extra mapping information to achieve an intermediate-tableless schema.

Robert

On May 13, 2009, at 5/1312:31 PM , Mike Kienenberger wrote:

It's certainly possible to develop templates that provide entity
interfaces.  I've done so in past projects.

And if you mess around with the generic signatures enough, you can
probably solve those problems as well.

While I won't recommend JPA (since I am now having to use it instead
of Cayenne), it's also not a problem to define 1:n relationships
without intermediate join tables.   We have many one-to-many and
many-to-one relationships defined in JPA, and none of them have
intermediate join tables.  In fact, you can have a one-to-many
relationship defined in a single object, so it's not a JPA issue.   I
still recommend using Cayenne instead of JPA, though.

object:
  - column: ID
  - column: PARENT_ID
  - relationship: parentObject
  - relationship: childObjectList



On Tue, May 12, 2009 at 4:40 AM, Jeremias Maerki <[email protected] > wrote:
Hi there,

after years of not having to deal with databases, I'm back to that topic.
So I'm still figuring out the best way for myself to do ORM. My first
two attempts were to use JCR (Jackrabbit) and JPA (OpenJPA). JCR worked fine although I dropped it because I can easily have >100'000 children
on one node which will result in performance problems or require
work-arounds that I don't particularly like. JPA then just went against my views of how databases should be modeled: an intermediate table for a simple 1:n relation. Not with me. I couldn't get the whole thing to work without the intermediate table. Then I stumbled upon Cayenne (I'm using
the Trunk) and immediately found myself at home.

Now, my problem: I've been trying to make an abstraction of my data
logic using interfaces, i.e. define an interface for each domain object and logic methods grouped in an interface per topic. That works pretty well for the JCR and JPA implementations. I wrote the test code once and
can reuse it for all implementations. With Cayenne, this gets a bit
difficult. I can always implement the domain interfaces in the domain
classes but when it comes to nested collections, this becomes a problem:

public interface Document extends Serializable {

[..]

   Collection<Representation> getRepresentations();

}

public interface Representation extends Serializable {
[..]
}

Generated domain class:

public class DocumentImpl extends _DocumentImpl implements Document {
[..]
}

public abstract class _DocumentImpl extends CayenneDataObject {
[..]
   @SuppressWarnings("unchecked")
   public List<RepresentationImpl> getRepresentations() {
return (List<RepresentationImpl>)readProperty("representations");
   }
}

List<RepresentationImpl> is not compatible with the List<Representation>
in the interface.

So, I'm wondering if there's a better way to do this that avoids this
problem, preferably without removing generics.

I was thinking that adding the interface in the object model could
optionally be specified and the code generator could, for example, do an "implements Document" for _DocumentImpl and return List<Representation>.
Of course, that would require some casting to DataObject in
addToManyTarget() for example. That's probably undesired and would probably
require quite a bit of work. But maybe there's a better way. I could
probably proxy/adapt the whole stuff but that's probably bad
performance- and effort-wise. Maybe I simply have to give up my puristic
data access abstraction. But I wanted to ask around first.

---

While I'm at it: Can I ask why DeleteQuery has been deprecated? I'd
rather use that instead of EJBQLQuery. Not having to concatenate strings to build queries is one of the features that makes Cayenne appeal to me
so much. I've had stuff like that back in my Delphi times and I liked
that very much.


Thanks a lot for any ideas and thoughts,
Jeremias Maerki



Reply via email to