On 8/29/2012 2:34 PM, Michael Bouschen wrote:
> Hi Linda,
>
> what I understood from your proposal is implicit joins in an ORDER BY clause are treated as outer joins, correct?
>
Not exactly. The intention of the spec was to constrain the order-by to attributes that
were covered by the SELECT. That is, to not exceed what SQL permits in an ORDER BY.
So, if you retrieve an entity, you can order by the state fields of that entity (but not
by statefields of a related entity not designated by the SELECT.
> But how about implicit joins in other clauses of the query, e.g. the SELECT clause.
> The following query excludes persons not having an address from the result
> select p.address from Person p
> where this one includes them:
> select a from Person p LEFT OUTER JOIN p.address a
>
> I am uncomfortable with defining different semantics of an implicit join depending on whether it is used in an ORDER BY
> clause versus a SELECT or WHERE clause. Is GROUP BY treated the same as ORDER BY, because it should not filter what is
> returned, too? But HAVING is used to filter over groups ... This sounds confusing to me.
>
>> That was the original intent,
>> and why, for example, we disallow relationship joins in the orderby
>> clause.
>
> Did we disallow relationships in an ORDER BY clause? An orderby_item might be a state_field_path_expression which might
> include a relationship navigation, as long as the navigated relationship is included in the SELECT clause. Am I wrong?
>
> Regards Michael
>
>>
>>
>> On 8/29/2012 6:26 AM, Oliver Gierke wrote:
>>> The problem is that adding an order by clause leaks a side effect into what is actually returned from the query,
>>> which shouldn't be the case.
>>>
>>> Yes, it's outlined in the spec but it's also inconsistent in regards of mapping metadata being applied to the JPQL
>>> expression. An @Column(name = "Foo") get's regarded, why shouldn't an @ManyToOne(optional = true) be?
>>>
>>> I think it get's more obvious if you consider a more simple case:
>>>
>>> select p from Person p
>>>
>>> VS.
>>>
>>> select p from Person p order by p.address.city
>>>
>>> The former includes people without addresses, the latter does not? And this is totally inexplainable from a pure
>>> mapping metadata point of view. I wonder why the two lines defining path expressions being forced into inner joins
>>> have made it into the spec in the first place. Were the side effects not considered? Why explicitly not applying the
>>> mapping metadata?
>>>
>>
>>
>> If address is an embeddable, this query is legit. If address is
>> an entity (as in your original example, the order-by is not).
>>
>> Perhaps what needs clarification is the notion of "result".
>>
>> What the spec says (section 4.4.4) is:
>>
>> 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.
>>
>> Perhaps what this should say is:
>>
>> 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 *contents* of
>> the result of the query.
>>
>>
>> With order-by in play, there are two aspects to the result:
>>
>> What is actually retrieved from the database ("contents of the
>> result"), and how it is ordered.
>>
>> In my view, the order-by clause should only affect the ordering of
>> what is returned, not filter it. That was the original intent,
>> and why, for example, we disallow relationship joins in the orderby
>> clause.
>>
>>
>>> Cheers,
>>> Ollie
>>>
>>> PS: I've summarized my findings in this gist (https://gist.github.com/3497047). The comments might be interesting to
>>> consider.
>>>
>>> Am 29.08.2012 um 14:47 schrieb Steve Ebersole<steve.ebersole_at_redhat.com>:
>>>
>>>> Not sure how this "side-effect" is "unpleasant and not easy to grasp".
>>>> It is explicitly called out in the spec.
>>>>
>>>> -1 for changing implicit joins to result in inner or outer joins
>>>> depending on the mapping. In such a case you can no longer see what
>>>> will happen just by looking at the query itself, which in my opinion is
>>>> far more "unpleasant and not easy to grasp".
>>>>
>>>>
>>>> On 08/28/2012 07:44 AM, Oliver Gierke wrote:
>>>>> Hi all,
>>>>>
>>>>> I just came across a JPQL spec scenario that seems to be a bit weird and I wonder whether there's something we
>>>>> should do about. Suppose you have a Person with optional Addresses:
>>>>>
>>>>> @Entity
>>>>> class Person {
>>>>>
>>>>> @OneToOne(nullable = true) Address address;
>>>>> }
>>>>>
>>>>> @Entity
>>>>> class Address {
>>>>> String city;
>>>>> }
>>>>>
>>>>> Now the query scenario here is that we'd like to get all Persons sorted by the Address' city:
>>>>>
>>>>> select p from Person p left outer join p.address order by p.address.city
>>>>>
>>>>> Surprisingly, this query will not return Persons not having an Address associated for the following reason: JPA 2.0
>>>>> spec section 4.4.4. defines path expressions as follows:
>>>>>
>>>>>> 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.
>>>>>
>>>>> That apparently forces persistence providers into adding an additional inner join to the query which rules out the
>>>>> Persons without Addresses in the first place. I think it's rather unfortunate to have this path expression
>>>>> definition applied to order by clauses as users probably don't expect adding a sort definition would strengthen the
>>>>> actual query criteria. So here are my questions:
>>>>>
>>>>> 1. Why was the path expression navigability defined as such in the first place and not as considering the mapping
>>>>> metadata (nullable = true -> outer join, nullable = false -> inner join). Not saying this is utterly wrong, just
>>>>> want to understand the probably available reasons.
>>>>> 2. Should/can this definition be changed to require consideration of the mapping information? The path expression
>>>>> definition is very much written with the purpose of defining selection criterias which is what they are effectively
>>>>> not used for when used in ORDER BY clauses. The current state leaves JPQL in the weird state that adding a sorting
>>>>> criteria affects the returned items not only in order but also in which items are returned at all, a side-effect
>>>>> which is unpleasant and not easy to grasp.
>>>>>
>>>>> Cheers,
>>>>> Ollie
>>>>>
>>>
>
>
> --
> *Michael Bouschen*
> *Prokurist*
>
> akquinet tech_at_spree GmbH
> Bülowstr. 66, D-10783 Berlin
>
> Fon: +49 30 235 520-33
> Fax: +49 30 217 520-12
> Email: michael.bouschen_at_akquinet.de
> Web: www.akquinet.de <http://www.akquinet.de>
>
> akquinet tech_at_spree GmbH, Berlin
> Geschäftsführung: Martin Weber, Dr. Torsten Fink
> Amtsgericht Berlin-Charlottenburg HRB 86780 B
> USt.-Id. Nr.: DE 225 964 680