Hi Chris,
thank you very much for your feedback.
Regarding the referenced bug, I understand your point of view (the
specification dictates a general attitude on the matter), but the
current behaviour is very counter-intuitive and brings to wrong
results as soon as I simply change the owner of the one-to-one
relationship. That is, a change in the mapping (which should
reasonably be transparent) breaks a query in the object model, in
which the intention to get "all the First instances that do not
have a corresponding Second instances" is clear, rather than "give
me always an empty result".
Anyway, the purpose of this message was not to bring that bug (or
enhancement or what you like) to the devs attention (although
that's good), but rather to request help to work around it, i.e.
to implement the outer join you also suggest.
When processing the metamodel in a completely generic way, if I
encounter the problematic case I need to get the foreign-key
attribute of the owning side of the relationship starting from the
non-owning side attribute (which is "mappedBy"). But I still
couldn't succeed on it, as I wrote in my previous e-mail :-(
I'm now investigating on RelationshipAccessor class, but I've yet
to figure out how to get an instance of it...
Thanks again,
Mauro
Il 08/01/2016 17:44, christopher delahunt ha scritto:
Hello Mauro,
Frameworks don't automatically treat the join as a left out join
because traversing a path in JPA's criteria queries or JPQL is
supposed to be treated as an inner join. Section 4.4.4 of the JPA
2.1 specification requires that
"Path _expression_ navigability is composed using “inner join”
semantics. That is, if the value of a non-terminal field in the
path _expression_ is null, the path is considered to have no value,
and does not participate in the determination of the result."
The fact that this works when the relationship is reversed is a
result of EclipseLink optimizations that take advantage of the
fact a join isn't required and not something intentionally done
for null handling. I believe I saw this behavior difference
documented when the feature was done, I just don't have a link
handy.
If you want consistent, spec compliant behavior in your queries,
you are required to specify the relationship use a left join in
the query. Ie:
"Select first from First first left join first.second second
where second is null"
or its equivalent in criteria expressions.
Best Regards,
Chris
One of the problems with considering this issue is that This is
why
On 08/01/2016 10:48 AM, Mauro
Molinari wrote:
Hello all,
I'm trying to implement some sort of "findByExample" feature in
my JPA application (using EclipseLink 2.6.1).
To do this, I'm inspecting the metamodel of my classes and
dynamically creating predicates to be added to a query built
with the Criteria API.
I encountered bug https://bugs.eclipse.org/bugs/show_bug.cgi?id=485414
which is causing me a lot of headache.
To work around it, I'm trying to implement the following logic:
"if the attribute is mapping a bidirectional relationship which
is not owned by this entity, then do an outer join and apply a
condition on the corresponding relationship attribute of the
target entity".
The simple case would be:
Predicate predicate = cb.equal(from.get(attribute),
exampleValue);
which substantially is: "WHERE MyEntity.attribute =
<exampleValue>"
But if exampleValue is NULL and attribute is a
@OneToOne(mappedBy="inverseAttributeName"), so it's mapped to a
non-owned relationship, I have to do a different thing to work
around the aforementioned bug.
So, I have:
- attribute, which is a SingularAttribute
- attributeType, which is the attribute type (
attribute.getType() )
- if
attributeType.getPersistenceType() ==
PersistenceType.ENTITY (=> a relation attribute)
then I get:
DatabaseMapping mapping = ((SingularAttributeImpl)
attribute).getMapping();
The DatabaseMapping object is promising, because it has the
method getRelationshipPartnerAttributeName() (or even
getRelationshipPartner()) which, if I understand it correctly,
should give me the attribute name of the related entity, if this
is a bidirectional mapping.
However what I see while debugging is that both methods always
return null (also the protected field "mapping.mappedBy" is
null!!), even if attribute is a SingularAttributeImpl of an
attribute which is mapped like this:
@OneToOne(cascade = CascadeType.ALL, fetch =
FetchType.LAZY, mappedBy = "inverseAttributeName", optional =
true, orphanRemoval = true)
and the inverse mapping is present (so the relationship actually
*is* bidirectional):
@Id @OneToOne(fetch = FetchType.EAGER, optional = false)
Something similar seems to be the fields
mapping.targetToSourceKeyFields and
mapping.sourceToTargetKeyFields, but those are maps containing
DatabaseFields, not attributes, so I doubt I can use them to
build my query (unless, of course, I am lucky and the attribute
name matches the database column name).
On the other hand, I can't find anything else useful in
SingularAttributeImpl (not to say the plain JPA interface...) to
get the desired result.
I'm almost at it, but I already spent a lot of time on this
final detail without success :-(
Any help would be *really* appreciated.
Thanks in advance,
Mauro
_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/eclipselink-users
_______________________________________________
eclipselink-users mailing list
eclipselink-users@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/eclipselink-users
|