The problem I’ve stabled upon started with a spring data repository like this:
public interface SampleRepository extends CrudRepository<Sample, Integer>{
@Query("select s from Sample s where s.id in :ids")
List<Sample> queryIn(@Param("ids") List<Integer> ids);
}
Actual query was of course more complicated that this. Complex enough to justify not using a query method. The problem emerges when you run this method with an empty collection as argument:
repository.queryIn(Collections.emptyList());
The result is database dependent. There no problem on H2, but on HSQLDB (and also at least on MSSQL) you get:
Caused by: org.hsqldb.HsqlException: unexpected token: )
at org.hsqldb.error.Error.parseError(Unknown Source)
at org.hsqldb.ParserBase.unexpectedToken(Unknown Source)
...
Syntax error in generated SQL query? How come? Lets first look at how in
clause is handled by hibernate.
Starting with a query that is run with a non-empty list parameter
repository.queryIn(Arrays.asList(1, 2, 3));
the SQL generated by hibernate looks something like this
select
sample0_.id as id1_0_,
sample0_.name as name2_0_
from
sample sample0_
where
sample0_.id in (
? , ? , ?
)
Turns out that hibernate can’t pass an array directly to the in
clause. It has to create a sql parameter for
every entry in the collection.
Aha, so when the collection passed to the query is empty, the SQL becomes:
select
sample0_.id as id1_0_,
sample0_.name as name2_0_
from
sample sample0_
where
sample0_.id in ()
Here is where the syntax error comes in. The SQL standard does require at least one value expression between the parenthesis.
So the closing bracket )
, right after (
, causes HSQLDB’s SQL parser to throw a syntax error. Quite rightly though.
Solutions
First of all, let me point out, that a corresponding find..In()
query method
works fine for an empty
parameter:
public interface SampleRepository extends CrudRepository<Sample, Integer>{
List<Sample> findByIdIn(List<Integer> ids);
}
So the following test passes:
@Test
public void findByIdIn() {
List<Sample> result = repository.findByIdIn(Collections.emptyList());
assertThat(result, empty());
}
But what to do, when the query is complex or would yeld a ridiculously long query method name like
findByProduct_Category_EmployeeResponsible_Departament_Location_CityIn(
List<City> cities
)
A custom @Query
won’t work, but there’s another option.
JPA Criteria API. This is what
Spring Data JPA is actually using
under the hood to generate JPA queries from query methods.
The bridge between Criteria API and Spring Data repositories is called
Specifications.
First thing is to make your repository implement JpaSpecificationExecutor<T>
:
public interface SampleRepository extends
CrudRepository<Sample, Integer>,
JpaSpecificationExecutor<Sample> {
// ...
}
Now you can call findAll(Specification<T>)
method on your repository:
repository.findAll(new Specification<Sample>() {
@Override
public Predicate toPredicate(Root<Sample> root,
CriteriaQuery<?> criteriaQuery, CriteriaBuilder criteriaBuilder) {
// ....
}
})
Great! You get a place, where you can dynamically create any WHERE
clause using Criteria API.
The criteria corresponding to SQL’s
WHERE id IN (1,2,3)
is
root.get("id").in(1,2,3);
//or
List<Integer> ids = Arrays.asList(1,2,3);
root.get("id").in(ids);
It’s worth noting here, that most of the Criteria API tutorials use the type-safe generated Metamodel classes. But it might be an overkill to set up persistence provider’s annotation processor just to handle a few queries. Most of your queries will probably be handled fine by Spring Data Query methods. Fortunately there’s an option to use string’s instead of Metamodel properties, like in the example above.
Just using root.get("id").in(ids)
will not save you from the possible SQL syntax error when ids
are empty. But as
the query is created dynamically, you have full control whether to include the in()
statement or not.
To mimic the spring’s standard find...In()
behaviour use this predicate:
@Override
public Predicate toPredicate(Root<Sample> root,
CriteriaQuery<?> criteriaQuery, CriteriaBuilder criteriaBuilder) {
if (ids.isEmpty()) {
return criteriaBuilder.disjunction();
} else {
return root.get("id").in(ids);
}
}
The mystic disjunction()
is a simple clause that is always FALSE
. The exact
javadoc states:
Create a disjunction (with zero disjuncts). A disjunction with zero disjuncts is false.
All right. It’s a working solution, but it can be made less verbose when using Java 8. First step is the replace the anonymous class with a lambda:
findAll((root, criteriaQuery, criteriaBuilder) -> {
if (ids.isEmpty()) {
return criteriaBuilder.conjunction()
} else {
return root.get("id").in(ids);
}
});
In this exact case it may be tempting to even use the conditional (?:
) operator, so that the brackets ({}
)
and return
can be omitted:
findAll((root, criteriaQuery, criteriaBuilder) ->
ids.isEmpty() ? criteriaBuilder.conjunction() : root.get("id").in(ids)
);
The other thing is, that it would be nice to have this method available directly on the repository, not in some service class. Here, next new Java 8 feature come in handy - interface default methods.
import org.springframework.data.jpa.repository.JpaSpecificationExecutor;
import org.springframework.data.repository.CrudRepository;
import java.util.List;
public interface SampleRepository extends CrudRepository<Sample, Integer>,
JpaSpecificationExecutor<Sample> {
// ...
default List<Sample> findIn(List<Integer> ids) {
return findAll((root, criteriaQuery, criteriaBuilder) ->
ids.isEmpty()
? criteriaBuilder.conjunction()
: root.get("id").in(ids)
);
}
}
New you can call this custom query method like any other repository method:
@Test
public void nonEmptySpecIn() {
List<Integer> ids = Arrays.asList(1, 2, 3);
List<Sample> result = repository.findIn(ids);
assertThat(
result.stream()
.map(sample -> sample.id)
.collect(toList()),
equalTo(ids)
);
}
Select all on empty IN
Common case with IN
clauses is when you have a search filter like:
Select categories:
[x] Home & Garden
[ ] Beauty, Health & Food
[ ] Sport & Outdoors
In this case, when the selected categories collection is empty, you want this filter to be ignored.
Having a predicate creation method, this change becomes trivial;
default List<Sample> findIn(List<Integer> ids) {
return findAll((root, criteriaQuery, criteriaBuilder) -> {
if (ids.isEmpty()) {
return null; // or criteriaBuilder.conjunction()
} else {
return root.get("id").in(ids);
}
});
}
As for the null
here. The toPredicate()
documentation says that you are not allowed to return it here.
But it turn out that spring data handles it rather correctly.
I’ve placed a
pull request to update the javadoc.
The complete sample code used in this article in available on github.