I am unsure about how to create a test case for this one. Reproduction steps:

1. Setup two entities with a bidirectional @OneToMany/@ManyToOne relationship like this:

   @Entity
   class Parent
   {
      @OneToMany(mappedBy="parent", cascade=CascadeType.ALL)
       public List<Child> children;
   }
   
   @Entity
   class Child
   {
      @ManyToOne
      public Parent parent;
   }

2. Create and persist a Parent with no children.
3. In a new session, load the Parent from the database.
4. Add a transient Child to the children collection.
5. Call a read operation like .size() on the children collection.
6. Attempt to persist the Parent
7. Exception will result

The ultimate issue:

1. When user does PersistentBag.add() under these circumstances write operation is queued.
2. PersistentBag.size() triggers PersistentBag.read() which triggers PersistentBag.initialize()
3. org.hibernate.engine.loading.CollectionLoadContext::endLoadingCollection() is eventually called. This is where the actual bug is.
   1. endLoadingCollection() calls org.hibernate.collection.AbstractPersistentCollection::endRead() which flushes all queued operations.
   2. AFTER flushing queued operations, endLoadingCollection() calls org.hibernate.enging.CollectionEntry::postInitialize() which creates what is supposed to be the pre-write snapshot of the bag.
4. Bag snapshot is now incorrect – it contains all entities queued for write. If any of these entities are transient, they will cause an exception in org.hibernate.collection.AbstractPersistentCollection.getOrpahans(). When it iterates over the old snapshot, it makes a call to org.hibernate.engine.ForeignKeys.getEntityIdentifierIfNotUnsaved() that cannot handle transient entities. Normally this wouldn't be an issue, but because snapshot contains queued writes, it can contain transient entities.