Re: [OJB-developers] Bug in ODMG transaction

[Thomas M. <tho...@ho...>]

Hi again Rainer,

Bischof, Rainer wrote:

> Hi,
>  
> I just moved to the current release which should fix some bugs with foreign
> keys and the ODMG interface. I discovered two more:
>  
> 1. Object A references Object B of a different class in a 1-1 relationship.
> A's PK consists only of the reference to B (i.e. B's PK, which is
> auto-incremented). Both objects are newly created and properly locked with
> the ODMG transaction. B's PK is incremented at that time but A's FK to B is
> not updated at that time (this is done only in broker.store). Therefore A's
> FK to B is 0 which does not work. OJB places a lock on an object whose PK is
> not yet defined.
> Solution: FKs must be assigned before the lock is performed. I modified the
> TransactionImpl to do this by copying the methods of the broker that perform
> that to the TransactionImpl. That's just a quick and ugly hack but works. 
>  


Thanks for your patch. I will have a look at this issue


>  
> 2. Object A holds a collection of Bs. Auto-Update is set to true in the
> repository. I create A and lock it, than I create some Bs, lock them and add
> them to A's collection. On commit the ObjectEnvelopeTable first stores A
> which also stores it's collections. Afterwards the ObjectEnvelopeTable goes
> through the ObjectEnvelopes of all Bs. All of these Bs have a StateNewDirty
> which always creates INSERT statements. But the Bs have already been stored
> as part of A's collection.
>  
> I currently do not have the time to provide a proper fix for this ( I just
> switched of AutoUpdate and it worked for me).
>  


Auto-Update and -Delete *must not* be set to true if working with the 
ODMG Implementation. There is no fix for this, it just does not work!
This is stated in http://objectbridge.sf.net/tutorial3.html#cascading.

cu,

Thomas


> Regards
> Rainer
>  
>  
>