I have done some research on this (google, 'in action', and forums) but haven't found a definative best practice.

I was looking for feedback on how to work with a collection of objects which do not yet have their key(id) defined. It is suggested that for one-many relationships the unordered type SET is the way to go. This works majestically in all instances but one I have found.

What happens if you want to add newly created objects to the collection before they have an ID assigned. Since the hashcode is generated by default from the ID field. All items overwrite each other and you end up having only the last one in the collection.

I know in theory we should be saving the items first to get the ID, and then adding them to the collection. But in some cases this is not really practical. Especially if you want to reuse the Hibernate Pojo's wrapped in a form bean for instance.

I saw a paragraph in 'in action' that mentions that 'bag' can be used, but it is not suggested. This is an issue because then the same element could be added to the collection twice. I guess you could modify the HashCode to use other fields, but I've read that is not recommended as well.

So in the end I would assume the only true way to work around this is to implement a bean identical to your Pojo with the exception of having the SET as ArrayList (or similar), as well as methods to convert from/to the POJO. I guess BeanUtil could be used to map across the values as well but I am not sure how this would handle the List --> Set mapping. PROBLEM: the big problem I have with this is that you have to then pass this extra object around through your business logic (along with the Hibernate Pojo hiearchy).

Sorry for the wordiness, but I am just not sure how best to do this. Looking for some enlightenment!

Thanks

Actually, I think if you do some searching on equals() and hashcode() you will see that hashing on id is not recommended. (especially in the case that you mentioned)

Now that I gave it another look and after reading: http://www.hibernate.org/109.html

What about adding another attribute called transientId and modifying the hashcode to use it and the primary key (id) together. The transientId would have no mapping and would exist only in the POJO - specifically to act as a distiguisher for adding these unsaved keyless objects to the set.

I think that would work.. ??

Any one see anything wrong with that?