We Keep Coding

Java Spring JPA: How to find the sum of only non null values in a column?

I currently have a query like this:
SELECT DISTINCT t.column1, SUM(t2.column2 IS NOT NULL)
FROM table t
LEFT OUTER JOIN table t2 on table.id = t2.id
GROUP BY column1, column2;
I am trying to implement the query using Spring JPA CriteriaBuilder. I see the CriteriaBuilder.sum() method, but I don't see a way to apply the IS NOT NULL part to the selection. Column2's data type is string.
Sample of my code
criteriaBuilder.multiselect(root.get("column1"), cb.sum(root.get("column2")));

Only in MySQL would such a query run, due MySQL’s relaxed syntax rules.
In particular, in mysql sum(column2 is not null) is a count, not a sum. The expression column2 is not null is boolean and in mysql false is 0 and true is 1, so summing this expression is a mysql hack to count the number of times column2 is not null.
To convert it to standard sql:
select
t.column1,
count(t2.column2)
from table t
left join table t2 on t.id = t2.id
group by t.column1
This works because count() (and all aggregate functions) ignore nulls.
This version also corrects the errant column in the group by clause - in any other database, your query would have produced a “grouping by aggregate expression” error.
This query will produce the same result in MySQL as your current query.

I was able to find a solution to my problem. Thanks to #bohemian for helping me write a correct sum expression.
final CriteriaBuilder cb = em.getCriteriaBuilder();
final CriteriaQuery<Model1> cq = cb.createQuery(Model1.class);
final Root<Model1> root = cq.from(Model1.class);
final Join<Model1, Model1> selfJoin =
root.join("tableJoinColumn", JoinType.LEFT);
selfJoin.on(...);
cq.multiselect(root.get("column1"), cb.sum(cb.selectCase()
.when(cb.isNull(selfJoin.get("column2")), 0).otherwise(1).as(Long.class)));
...
The self join required me to create an additional property on my model.
Model1.java
/**
* Property for LEFT INNER JOIN.
*/
#OneToOne(fetch = FetchType.LAZY)
#JoinColumn(name="id")
private Model1 tableJoinColumn;
How to use JPA CriteriaBuilder selectCase() so that it can have Predicate as result?
Self join in criteria query

Specfiy columns which should be considered for distinct calculation

I'm using javax.persistence.criteria.CriteriaBuilder and javax.persistence.criteria.CriteriaQuery to select some entities.
I now want to select only the entities that are unique which should be specified by a certain column.
There is the method javax.persistence.criteria.CriteriaQuery#distinct which only returns unique entities.
I would rather need something like
CriteriaQuery<T> distinct(String... columnNames)
Do you know how I can bake such a distinct in my JPA CriteriaQuery?
It seems to be possible with Hibernate.

The following statement has no sense:
I now want to select only the entities that are unique which should be
specified by a certain column.
The result sets are filtered by 'distinct' if they are 'exactly the same'.
The entities are not the same if only some fields are the same.
You can make distinct clause on resultset in the following manner:
CriteriaBuilder builder = em.getCriteriaBuilder();
CriteriaQuery query = builder.createQuery();
Root<Friend> c = query.from(Friend.class);
query.multiselect(c.get(Friend_.firstName),c.get(Friend_.lastName)).distinct(true);
then you will get unique combination of firstName and lastName from Friend entities.
So for instance... 'Give me all unique combinations from my Friends where the firstName and lastName of the friend is unique.' But it doesn't mean give me unique friends.

distinct takes a boolean as parameter. You can use multiselect to select more than one column like in this example:
CriteriaQuery<Country> q = cb.createQuery();
Root<Country> c = q.from(Country.class);
q.multiselect(c.get("currency"), c.get("countryCode")).distinct(true);

Hibernate returns list of nulls although executed SQL returns values

I'm using hibernate as an ORMapper. I want to execute an actually rather simple hql query:
SELECT a
FROM Foo a
WHERE a.status = :A0status
ORDER BY a.bookingTypeCode ASC,
a.priority ASC
This hql query is then converted into a sql query which looks something like this:
select a.*
from Foo a
where a.status='A'
order by a.bookingtypecode ASC,
a.priority ASC
When I execute the sql on the oracle database using the Oracle SQL Developer I get 17 rows returned. However, when I execute the hql query (using the list method of a Query I get a list of 17 elements that are all null. Although the number of elements is correct, not a single one of the elements is actually loaded.
This is the way I create and execute my query:
// the hql query is stored in the hqlQuery variable;
// the parameter are stored in a Map<String, Object> called params
Query hQuery = hibSession.createQuery(hqlQuery);
for (Entry<String, Object> param : params.entrySet()) {
String key = param.getKey();
Object value = param.getValue();
hQuery.setParameter(key, value);
}
List<?> result = hQuery.list();
Does anyone know what might be the problem here?
Update 1
I've recently upgrade from hibernate 3.2 to 4.3.5. Before the upgrade everything worked fine. After the upgrade I get this error.

I've set the Log level of hibernate to TRACE and found the problem. It was actually a mapping/logic/database error. The primary key consisted of two columns (according to the entity class) and one of these columns was nullable. However a primary key can never be nullable. Therefore hibernate always returned null.

If you have not set a custom (and buggy) ResultTransformer, my second best guess is that your debugger is lying to you. Does you code actually receives a list of null?
Also make sure to test with the code you are showing is. Too many times, people simplify things and the devil is in the details.

This error is happening to me. MySQL query browser works, but in hibernate of 7 columns and only one column always came with all null fields. I checked all the ids and they were not null. The error was in the construction of SQL Native. I had to change the way of writing it. Ai worked.
SELECT c.idContratoEmprestimo as idContratoEmprestimo,
c.dtOperacao as dataOperacao,
p.cpf as cpf,
p.nome as nome,
(Select count(p2.idParcelaEmprestimo) from EMP_PARCELA p2 where p2.valorPago > 0 and p2.dtPagamento is not null
and p2.idContratoEmprestimo = c.idContratoEmprestimo and p2.mesCompetencia <= '2014-08-01') as parcelasPagas, c.numeroParcelas as numeroParcelas,
pe.valorPago as valorParcela
FROM EMP_CONTRATO c inner join TB_PARTICIPANTE_DADOS_PLANO AS pp on pp.idParticipantePlano = c.idParticipantePlano
inner join TB_PARTICIPANTE as p on p.id = pp.idParticipante
inner join TB_PARTICIPANTE_INSTITUIDOR as pi on pi.PARTICIPANTE_ID = p.id
inner join EMP_PARCELA as pe on pe.idContratoEmprestimo = c.idContratoEmprestimo
where c.dtInicioContrato <= '2014-08-01' and pi.INSTITUIDOR_ID = 1
and c.avaliado is true
and pe.mesCompetencia = '2014-08-01'
and c.deferido is true
and c.dtQuitacao is null
and c.dtExclusao is null
and pe.valorPago is not null
group by c.idContratoEmprestimo
order by p.nome

How to convert nested SQL to HQL

I am new to the Hibernate and HQL. I want to write an update query in HQL, whose SQL equivalent is as follows:
update patient set
`last_name` = "new_last",
`first_name` = "new_first"
where id = (select doctor_id from doctor
where clinic_id = 22 and city = 'abc_city');
doctor_id is PK for doctor and is FK and PK in patient. There is one-to-one mapping.
The corresponding Java classes are Patient (with fields lastName, firstName, doctorId) and Doctor (with fields doctorId).
Can anyone please tell what will be the HQL equivalent of the above SQL query?
Thanks a lot.

String update = "update Patient p set p.last_name = :new_last, p.first_name = :new_first where p.id = some (select doctor.id from Doctor doctor where doctor.clinic_id = 22 and city = 'abc_city')";
You can work out how to phrase hql queries if you check the specification. You can find a section about subqueries there.

I don't think you need HQL (I know, you ask that explicitly, but since you say you're new to Hibernate, let me offer a Hibernate-style alternative). I am not a favor of HQL, because you are still dealing with strings, which can become hard to maintain, just like SQL, and you loose type safety.
Instead, use Hibernate criteria queries and methods to query your data. Depending on your class mapping, you could do something like this:
List patients = session.CreateCriteria(typeof(Patient.class))
.createAlias("doctor", "dr")
.add(Restrictions.Eq("dr.clinic_id", 22))
.add(Restrictions.Eq("dr.city", "abc_city"))
.list();
// go through the patients and set the properties something like this:
for(Patient p : patients)
{
p.lastName = "new lastname";
p.firstName = "new firstname";
}
Some people argue that using CreateCriteria is difficult. It takes a little getting used to, true, but it has the advantage of type safety and complexities can easily be hidden behind generic classes. Google for "Hibernate java GetByProperty" and you see what I mean.

update Patient set last_name = :new_last , first_name = :new_first where patient.id = some(select doctor_id from Doctor as doctor where clinic_id = 22 and city = abc_city)

There is a significant difference between executing update with select and actually fetching the records to the client, updating them and posting them back:
UPDATE x SET a=:a WHERE b in (SELECT ...)
works in the database, no data is transferred to the client.
list=CreateCriteria().add(Restriction).list();
brings all the records to be updated to the client, updates them, then posts them back to the database, probably with one UPDATE per record.
Using UPDATE is much, much faster than using criteria (think thousands of times).

Since the question title can be interpreted generally as "How to use nested selects in hibernate", and the HQL syntax restricts nested selects only to be in the select- and the where-clause, I would like to add here the possibility to use native SQL as well. In Oracle - for instance - you may also use a nested select in the from-clause.
Following query with two nested inner selects cannot be expressed by HQL:
select ext, count(ext)
from (
select substr(s, nullif( instr(s,'.', -1) +1, 1) ) as ext
from (
select b.FILE_NAME as s from ATTACHMENT_B b
union select att.FILE_NAME as s from ATTACHEMENT_FOR_MAIL att
)
)
GROUP BY ext
order by ext;
(which counts, BTW, the occurences of each distinct file name extension in two different tables).
You can use such an sql string as native sql like this:
#Autowired
private SessionFactory sessionFactory;
String sql = ...
SQLQuery qry = sessionFactory.getCurrentSession().createSQLQuery(sql);
// provide an appropriate ResultTransformer
return qry.list();

JPQL get most recent rows

Let's say I have the following tables
my_profile_data
-------------
integer: my_profile_data_id
integer: profile_id
integer: profile_data_type_id
date: date_changed
string: value
my_profile
-------------
integer: profile_id
string: name
profile_data_type
-------------
integer: profile_data_type_id
string: name
I want to get the most recent profile information for each profile data type. In plain SQL this would look something like:
select mpd.profile_id, mpd.profile_data_type_id, mpd.value, max(mpd.date_changed)
from my_profile_data mpd, my_profile mp
where mpd.profile_id = mp.profile_id and mp.name='The Profile I Want'
group by mpd.profile_data_type_id
I've tried different variants of the following JPQL query, but can't get it to work.
SELECT mpd FROM MyProfileData mpd LEFT JOIN
(SELECT mpd.profileId profileId, MAX(mpd.dateChanged) FROM MyProfileData mpd
LEFT JOIN mp.profile
WHERE mp.name = :name
GROUP BY mpd.profileDataTypeId) recent
ON (rp.profileid = recent.profileId)
Is this query doable in JPA?
I'm using EclipseLink as my JPA provider.
The innermost exception I get when I try to run this is
Caused by: NoViableAltException(81!=[506:7: (n= joinAssociationPathExpression ( AS )? i= IDENT | t= FETCH n= joinAssociationPathExpression )])
at org.eclipse.persistence.internal.jpa.parsing.jpql.antlr.JPQLParser.join(JPQLParser.java:3669)
... 73 more

Assuming DATE is actually a timestamp that you're not worried about colliding, it seems like your query could be as simple as
select mpd
from MyProfileData mpd
where mpd.profile.name = :name
and mpd.date = (select max(mpd1.date) from MyProfileData mpd1 where mpd1.profile.name = :name)
Are you using a DBMS like an older MySQL that hates subselects?
I am also thinking that perhaps your problem is you haven't mapped the object relationship from MyProfileData to ProfileData, and all you have is the actual integer value of the field. This will make writing JPQL queries quite hard in general.
Edit:
Continuing on the assumption that the dates don't collide for any given profile + profile data type combo, (so date uniquely identifies a row within the subset of a particular profile + profile type combination) you can just grab all the dates:
select mpd from MyProfileData
where mpd.profile.name = :name
and mpd.date in (select max(mpd1.date)
from MyProfileData mpd1
where mpd1.profile = mpd.profile group by mpd.profileDataType)
Your original SQL example isn't actually legal, so it's tough to come up with a way to reproduce what it looks like it's trying to do without having a way to uniquely identify the rows while excluding the value.

I gave up on trying to create this query in JPA and wrote a native query instead