Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
2.0.0
-
None
-
None
Description
I have a Model with one Entity with an Relation to a Entity of the same Class.
Im using the LifecycleListeners to detect delete Events onto this Entities.
I had an Testcase with two Entities e1 and e2. e1 had a reference to e2. Now i'm removing e2:
em.getTransaction().begin();
TestEntity e1=new TestEnity();
em.persist(e1);
TestEntity e2=new TestEntity();
e1.setRef(e2);
em.persist(e2);
em.remove(e2);
em.getTransaction().commit();
In the Listener ill create a Query to find the Pointing Entity pointing to it (here e1) and remove this entity too.
TestEntity toDelete=helper.getReferer(entity); <- a simple query to get the obj with ref=entity
em.remove(toDelete);
assert(em.isRemoved(lastPoint));
This assert here failes inside the Listener.
After some analysis, i think, i found the Problem:
Inside the BrokerImpl is a Set called "_operating". I think it should avoid endless recursions in cycle-Models or twice calls of delete with the same entity. This set is used inside the persist AND the delete Method. The workflow seems like this:
-> remove(e2);
-> call Listeners
-> make query
-> flush needed, so persist the Entities
-> the e1 is now added here into that _operating-Set (BrokerImpl Line 2522)
-> query complete, the right entity was found
-> remove(e1);
-> the _operating set contains the e1
-> ignore the delete command
It could be a Bugfix to create a second _operating Variable for these. One for persists and one for deletes, so that there is no way ignoring a delete-command