Hello Markus,
Yes, only part of an embedded object can be the PK. I am quite sure that there is nothing in TopLink or the specification that prevents additional mappings in the Embeddable.
In the case of an additional change in the embedded object that did not alter the PK there would be a change record for the aggregate obejct that included a change record for the non-pk mapping. The AggregateObjectMapping would most likely still be flagged as a PK mapping.
--Gordon
-----Original Message-----
From: Markus.Fuchs_at_Sun.COM [mailto:Markus.Fuchs_at_Sun.COM]On Behalf Of Markus Fuchs
Sent: Monday, September 18, 2006 2:22 PM
To: persistence_at_glassfish.dev.java.net
Subject: Re: Fix for issue 702
Hi Gordon,
I've moved the validation into AbstractDirectMapping and OneToOneMapping. Please find some questions below!
Gordon Yorke wrote:
Hello Markus,
The code changes look good. I know you asked about the AggregateObjectMapping before but now that I think about it I don't think we should include it. In TopLink the AggregateObjectMapping can be part of the PK mappings even if only a component of the AggregateObject is a PK mapping. What do you mean by component? Is it that only part of the embedded object is part of the PK?
Your current check would throw an exception even if the values in the Aggregate Object (Embeddable) did not change. Why is that? If there is a change record, there was an update, correct?
The underlying DirectMapping in the Aggregate object should still be validated by the current code if the PK has changed so the PK would continue to be validated even if the code was removed from AggregateObjectMapping.
Makes sense!
Thanks,
-- markus.
--Gordon
-----Original Message-----
From: Markus.Fuchs_at_Sun.COM [mailto:Markus.Fuchs_at_Sun.COM]On Behalf Of
Markus Fuchs
Sent: Friday, September 15, 2006 8:09 PM
To: persistence_at_glassfish.dev.java.net
Subject: Re: Fix for issue 702
Hi Gordon,
The pk update validation is now implemented in
DatabaseMapping.writeFromObjectIntoRowWithChangeRecord(), and is called
from the derived classes AbstractDirectMapping, OneToOneMapping, and
AggregateObjectMapping. The classes ValidationException and
ValidationExceptionResource haven't changed. I also added one more test
for overlapping primary/foreign keys to EntityManagerJUnitTestSuite.
The JUnit tests on Oracle and JavaDB are successful. What do you think?
Thanks!
-- markus.
Gordon Yorke wrote:
Actually I wasn't precise enough. The actual method would be writeFromObjectIntoRowWithChangeRecord(), If not changeRecord.getOwner().isNew() then you must have an update to this attribute and if this.isPrimaryKeyMapping() then you have an error case. There is no need to have the old value to validate that the error occured the fact that the method is called indicates the mapping was updated.
--Gordon
-----Original Message-----
From: Markus.Fuchs_at_Sun.COM [mailto:Markus.Fuchs_at_Sun.COM]On Behalf Of
Markus Fuchs
Sent: Friday, September 15, 2006 2:22 PM
To: persistence_at_glassfish.dev.java.net
Subject: Re: Fix for issue 702
Hi Gordon,
Thinking more about adding the pk update validation in the write phase,
I see two problems:
* How do I know what kind of operation
DirectToFieldMapping.writeFromObjectIntoRow() is called for? I.e. insert
or update (or even delete?)?
* Do I have access to the previous field value at this point, to compare
new and old value?
Thank you!
-- markus.
Gordon Yorke wrote:
Hello Markus,
The code looks good but it will only apply to the DeferredChangeDetectionPolicy. If any other policies are developed or used (say an active change detection policy that calcs changes as they are made) then this validation would not be applicable as well these changes do not really take updatable = false into account.
Perhaps a better spot for this validation would be within the write cycle (DirectToFieldMapping.writeFromObjectIntoRow()). Validation here, during the write cycle, would apply to any change detection policies and could take into account updatable = false? (ie. if the attribute was updated but the user has marked the column as not updatable )?
What do you think?
--Gordon
-----Original Message-----
From: Markus.Fuchs_at_Sun.COM [mailto:Markus.Fuchs_at_Sun.COM]On Behalf Of
Markus Fuchs
Sent: Wednesday, September 13, 2006 9:07 PM
To: Tom Ware
Cc: persistence_at_glassfish.dev.java.net
Subject: Fix for issue 702
Hi Tom,
To address issue "702 - Can't update a primary key of a composite key",
I added validation preventing updates to primary key fields.
I've changed DeferredChangeDetectionPolicy to call
DatabaseMapping.calculateValidatedChange instead of
DatabaseMapping.compareForChange directly.
DatabaseMapping.calculateValidatedChange calls compareForChange to
prepare the ChangeRecord for the field, and than does the validation, if
the ChangeRecord is not null (= the field has been set or updated). Any
updates to primary key fields are disallowed. Setting primary key fields
for inserts is allowed, of course! I added a ValidationException and
changed ValidationExceptionResource for that purpose.
EntityManagerJUnitTestSuite now has two new tests:
a) that the correct exception if thrown in case of a primary key update
b) that updating a primary key field with the same value is allowed
The entity-persistence-tests pass on Oracle. Please do the code-review.
Thank you!
-- markus.