persistence@glassfish.java.net

RE: Fix for issue 1021

From: Gordon Yorke <gordon.yorke_at_oracle.com>
Date: Tue, 26 Sep 2006 10:35:52 -0400

Hello Markus,
    I am a little confused here. There was already a discussion about this behaviour and I believe the conclusion was to allow users to reference non-entity inheritance heirarchy members in their object model. TopLink would persist the state from the parent Entity object. (discussion thread had subject : "How to prevent stack overflow" and was in reference to gf bug 939).
    Contrary to the comments in the bug there is nothing the specification that prevents this behaviour and the specification even covers this case in section 2.1.9.3
    If we are not validating EntityManager and Query apis then that needs to be added but this bug is not a validation issue.
--Gordon
  -----Original Message-----
  From: Markus.Fuchs_at_Sun.COM [mailto:Markus.Fuchs_at_Sun.COM]On Behalf Of Markus Fuchs
  Sent: Monday, September 25, 2006 8:08 PM
  To: persistence_at_glassfish.dev.java.net
  Subject: Fix for issue 1021


  Hi TopLink team,

  I'm trying to fix issue 1021: VALIDATION: Disallow persisting non-entity subclasses

  Can I assume that there's a one-to-one relationship btw. entity classes
  and descriptors? I.e. is the follow statement true:

  Class is annotated as @Entity <=> There's a ClassDescriptor for that class.

  Based on this assumption, the attached change throws a RuntimeException in
  AbstractSession.getDescriptor(Class theClass), if the descriptor for a class
  is not found. Adding it there will also change TopLink behavior for the
  non-JPA case. The exception message must still be localized, of course.
  Is there any reason, why TopLink returns the superclass descriptor, if no
  descriptor for the a class can be found?

  Another place to add this validation would be in EntityManagerImpl,
  similar to how em.contains() is called for some JPA operations. But
  that would not cover the cases were the operation is cascaded to an
  instance, as the internal operation is called directly and not through
  the EM interface.

  I tested the validation by trying to persist a non-entity class or a
  mapped superclass instance.

  What do you think?

  Thanks!

  -- markus.