Why is the same database entry represented by multiple JPA bean instances? - jpa

Question

Today I stumbled over some unexpected behaviour of EclipseLink. (I don't know if this is bound to EclipseLink or if this is the same for all JPA providers.)
I assumed that retrievals of a managed JPA bean always return references to the same object instance when issued inside the same transaction (using the same EntityManager).
If that is right, I don't know why I receive an error when I execute the following test case:
#Test
public void test_1() {
EntityManager em = newEntityManager();
em.getTransaction().begin();
// Given:
Product prod = newProduct();
// When:
em.persist(prod);
em.flush();
Product actual =
em.createQuery("SELECT x from Product x where x.id = "
+ prod.getId(), Product.class).getSingleResult();
// Then:
assertThat(actual).isSameAs(prod); // <-- FAILS
em.getTransaction().commit();
}
The statement marked with "FAILS" throws the following AssertionError:
java.lang.AssertionError:
Expecting:
<demo.Product#35dece42>
and actual:
<demo.Product#385dfb63>
to refer to the same object
Interestingly the following slightly modified test succeeds:
#Test
public void test_2() {
EntityManager em = newEntityManager();
em.getTransaction().begin();
// Given:
Product prod = newProduct();
// When:
em.persist(prod);
em.flush();
Product actual = em.find(Product.class, prod.getId());
// Then:
assertThat(actual).isSameAs(prod); // <-- SUCCEEDS
em.getTransaction().commit();
}
Obviously there is a difference between finding and querying objects.
Is that the expected behaviour? And why?
--Edit--
I think I found the source of the problem: Product has an ID of type ProductId.
Here is the relevant code:
#Entity
#Table(name = "PRODUCT")
public class Product implements Serializable {
private static final long serialVersionUID = 1L;
#Id
#Column(name = "ID", nullable = false)
#Converter(name = "productIdConverter", converterClass = ProductIdConverter.class)
#Convert("productIdConverter")
private ProductId id;
#Column(name = "NAME", nullable = false)
private String name;
[...]
}
The #Convert and #Converter annotations are EclipseLink-specific.
Unlike JPA 2.1 Converters you may place them on ID fields.
But it seems that in certain circumstances EclipseLink has problems to find a managed bean in its session cache if that bean uses a custom type for its ID field.
I guess I have to file a bug for that.

Answer 1

I found the cause of the problem and a solution.
We are using a custom ID class (ProductId) for Product, together with a custom (EclipseLink-specific) Converter-Class ProductIdConverter which has a bad implementation of the convertObjectValueToDataValue(...) method.
Here is the relevant code:
/**
* Convert the object's representation of the value to the databases' data representation.
*/
#Override
public final Object convertObjectValueToDataValue(Object objectValue, Session session) {
if (objectValue == null) {
return null;
}
Long longValue = ((ProductId) objectValue).getLong();
return longValue;
}
Please note that the method returns Long instances (or null).
But since we are using Oracle as our database backend and have declared the product's ID column as NUMBER, the JDBC Driver maps the column value as BigDecimal. This means, we have to make sure, that our convertObjectValueToDataValue(...) also returns BigDecimal instances.
So the correct implementation is:
/**
* Convert the object's representation of the value to the databases' data representation.
*/
#Override
public final Object convertObjectValueToDataValue(Object objectValue, Session session) {
if (objectValue == null) {
return null;
}
Long longValue = ((ProductId) objectValue).getLong();
return BigDecimal.valueOf(longValue);
}
Now this method returns only BigDecimal instances.