[
https://issues.apache.org/jira/browse/OPENJPA-968?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dianne Richards updated OPENJPA-968:
------------------------------------
Attachment: problemPatch.txt
Pinaki - I think I've discovered a problem with one of the changes you made, in
the ValueMetaDataImpl, with the new getCascadeDetach() method. With the way you
did it, the result returns CASCADE_IMMEDIATE for all simple types
(non-Entities). This doesn't make sense and is also not done for the other
cascade types (persist, attach, etc.). I've updated your test case with some
print statements to confirm this. I'm including an updated ValueMetaDataImpl
and TestFlush in this patch. I copied the additional code in getCascadeDetach()
from getCascadeAttach(). But, there are also differences in the other
getCascade.... methods, so I'm not entirely sure I'm doing the right thing. So,
please look at this and comment.
> Change in default detach() behavior for JPA 2.0
> -----------------------------------------------
>
> Key: OPENJPA-968
> URL: https://issues.apache.org/jira/browse/OPENJPA-968
> Project: OpenJPA
> Issue Type: Improvement
> Components: kernel
> Affects Versions: 2.0.0-M1, 2.0.0
> Reporter: Pinaki Poddar
> Assignee: Pinaki Poddar
> Fix For: 2.0.0-M1, 2.0.0
>
> Attachments: problemPatch.txt
>
>
> JPA 2.0 specification has introduced a detach() method in EntityManager.
> OpenJPA already has a detach() method prior to JPA 2.0. There are several
> major differences between the new method introduced in JPA 2.0 spec versus
> existing OpenJPA detach method.
> 1. The method signatures are different
> JPA 2.0: void detach(Object entity);
> OpenJPA 1.x : <T> T detach(T entity);
> 2. This signature difference also points to a basic difference in behavior.
> OpenJPA detach() creates a copy of the given input entity and returns it.
> While JPA 2.0 specifies in-place detach and hence returns a void.
> 3. The other basic difference is OpenJPA detach does not remove the input
> entity from the persistence context. While JPA 2.0 specifies that the input
> entity be removed from the context.
> 4. OpenJPA detach flushes a dirty instance before detaching. This implicit
> flush behavior can be configured but flushing is the default.
> 5. OpenJPA detach() provides several options on which related instances will
> become detached. They are 'loaded', 'fetch-group' and 'all'. With 'loaded'
> being the default.
> Whereas JPA 2.0 introduces a DEATCH cascade type and specified that the
> relationships that are cascaded with DETACH or ALL be traversed during
> detachment.
> It is not clear from the spec, however,
> a) whether an unloaded relation which has DETACH cascade will get loaded
> as a side-effect of detach().
> b) whether a relation that is currently loaded but not cascaded for DETACH
> will be included in the detached graph.
> If answer to (a) is no and answer to (b) is yes, (which the spec does
> seems to suggest) then we can effectively continue with 'loaded' as the
> default behavior.
> To accommodate these differences following actions are proposed:
> A1. Request JPA Spec committee to change the API method in JPA 2.0 to OpenJPA
> detach() method signature. If that request is not met, then change OpenJPA
> API according to JPA 2.0 spec. It will break backward compatibility of
> OpenJPAEntityManager API. There seems to be no way out.
> A2. The default behavior will change in the following way
> a) the detach will be in-place
> b) no implict flush of dirty instances
> c) based on condition, how JPA 2.0 spec clarifies the DETACH
> cascade and loaded fields for inclusion in the detached graph, change
> 'loaded' default to 'cascade' default.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
