Spring Data JPA. How to get only a list of IDs from findAll() method

Unfortunately Projections does not work with specifications. JpaSpecificationExecutor return only a List typed with the aggregated root managed by the repository ( List<T> findAll(Specification<T> var1); )

An actual workaround is to use Tuple. Example :

    @Override
    public <D> D findOne(Projections<DOMAIN> projections, Specification<DOMAIN> specification, SingleTupleMapper<D> tupleMapper) {
        Tuple tuple = this.getTupleQuery(projections, specification).getSingleResult();
        return tupleMapper.map(tuple);
    }

    @Override
    public <D extends Dto<ID>> List<D> findAll(Projections<DOMAIN> projections, Specification<DOMAIN> specification, TupleMapper<D> tupleMapper) {
        List<Tuple> tupleList = this.getTupleQuery(projections, specification).getResultList();
        return tupleMapper.map(tupleList);
    }

    private TypedQuery<Tuple> getTupleQuery(Projections<DOMAIN> projections, Specification<DOMAIN> specification) {

        CriteriaBuilder cb = entityManager.getCriteriaBuilder();
        CriteriaQuery<Tuple> query = cb.createTupleQuery();

        Root<DOMAIN> root = query.from((Class<DOMAIN>) domainClass);

        query.multiselect(projections.project(root));
        query.where(specification.toPredicate(root, query, cb));

        return entityManager.createQuery(query);
    }

where Projections is a functional interface for root projection.

@FunctionalInterface
public interface Projections<D> {

    List<Selection<?>> project(Root<D> root);

}

SingleTupleMapper and TupleMapper are used to map the TupleQuery result to the Object you want to return.

@FunctionalInterface
public interface SingleTupleMapper<D> {

    D map(Tuple tuple);
}

@FunctionalInterface
public interface TupleMapper<D> {

    List<D> map(List<Tuple> tuples);

}

Example of use :

        Projections<User> userProjections = (root) -> Arrays.asList(
                root.get(User_.uid).alias(User_.uid.getName()),
                root.get(User_.active).alias(User_.active.getName()),
                root.get(User_.userProvider).alias(User_.userProvider.getName()),
                root.join(User_.profile).get(Profile_.firstName).alias(Profile_.firstName.getName()),
                root.join(User_.profile).get(Profile_.lastName).alias(Profile_.lastName.getName()),
                root.join(User_.profile).get(Profile_.picture).alias(Profile_.picture.getName()),
                root.join(User_.profile).get(Profile_.gender).alias(Profile_.gender.getName())
        );

        Specification<User> userSpecification = UserSpecifications.withUid(userUid);

        SingleTupleMapper<BasicUserDto> singleMapper = tuple -> {

            BasicUserDto basicUserDto = new BasicUserDto();

            basicUserDto.setUid(tuple.get(User_.uid.getName(), String.class));
            basicUserDto.setActive(tuple.get(User_.active.getName(), Boolean.class));
            basicUserDto.setUserProvider(tuple.get(User_.userProvider.getName(), UserProvider.class));
            basicUserDto.setFirstName(tuple.get(Profile_.firstName.getName(), String.class));
            basicUserDto.setLastName(tuple.get(Profile_.lastName.getName(), String.class));
            basicUserDto.setPicture(tuple.get(Profile_.picture.getName(), String.class));
            basicUserDto.setGender(tuple.get(Profile_.gender.getName(), Gender.class));

            return basicUserDto;
        };

        BasicUserDto basicUser = findOne(userProjections, userSpecification, singleMapper);

I hope it helps.

This is now supported by Spring Data using Projections:

interface SparseCustomer {  

  String getId(); 

  String getName();  
}

Than in your Customer repository

List<SparseCustomer> findAll(Specification<Customer> spec);

EDIT:
As noted by Radouane ROUFID Projections with Specifications currently doesn't work beacuse of bug.

But you can use specification-with-projection library which workarounds this Spring Data Jpa deficiency.

I solved the problem.

(As a result we will have a sparse Customer object only with id and name)

Define their own repository:

public interface SparseCustomerRepository {
    List<Customer> findAllWithNameOnly(Specification<Customer> spec);
}

And an implementation (remember about suffix - Impl as default)

@Service
public class SparseCustomerRepositoryImpl implements SparseCustomerRepository {
    private final EntityManager entityManager;

    @Autowired
    public SparseCustomerRepositoryImpl(EntityManager entityManager) {
        this.entityManager = entityManager;
    }

    @Override
    public List<Customer> findAllWithNameOnly(Specification<Customer> spec) {
        CriteriaBuilder criteriaBuilder = entityManager.getCriteriaBuilder();
        CriteriaQuery<Tuple> tupleQuery = criteriaBuilder.createTupleQuery();
        Root<Customer> root = tupleQuery.from(Customer.class);
        tupleQuery.multiselect(getSelection(root, Customer_.id),
                getSelection(root, Customer_.name));
        if (spec != null) {
            tupleQuery.where(spec.toPredicate(root, tupleQuery, criteriaBuilder));
        }

        List<Tuple> CustomerNames = entityManager.createQuery(tupleQuery).getResultList();
        return createEntitiesFromTuples(CustomerNames);
    }

    private Selection<?> getSelection(Root<Customer> root,
            SingularAttribute<Customer, ?> attribute) {
        return root.get(attribute).alias(attribute.getName());
    }

    private List<Customer> createEntitiesFromTuples(List<Tuple> CustomerNames) {
        List<Customer> customers = new ArrayList<>();
        for (Tuple customer : CustomerNames) {
            Customer c = new Customer();
            c.setId(customer.get(Customer_.id.getName(), Long.class));
            c.setName(customer.get(Customer_.name.getName(), String.class));
            c.add(customer);
        }
        return customers;
    }
}

Why not using the @Query annotation?

@Query("select p.id from #{#entityName} p")
List<Long> getAllIds();

The only disadvantage I see is when the attribute id changes, but since this is a very common name and unlikely to change (id = primary key), this should be ok.