我试图在select子句中配置一个子查询，以实现如下sql等价项：

SELECT 
  t0.field1 AS a1, 
  t0.field2 AS a2, 
  t0.field3 AS a3, 
  t0.field4 AS a4, 
  t0.field5 AS a5, 
  (SELECT COUNT(t1.id) from table1 t1 WHERE t1.field1t1 = t2.field1t2) AS a6

FROM 
  table0 t0, 
  table2 t2

WHERE 
  (t2.field1t2 = t0.field5)

我找到了一些关于如何做到这一点的参考资料，比如：
select子句中带有jpa条件api的子查询
因此，我在jpa实现中构建了一个tuplequery，以便首先获取一个实体（在上一个sql示例中，所有字段-a1..a5-来自table0），然后获取count字段（a6）：

CriteriaBuilder cb = em.getCriteriaBuilder();

// Main Query
CriteriaQuery cq = cb.createTupleQuery();
Root<EntityTable0> root = cq.from(EntityTable0.class);

// Subquery
Subquery sub = cq.subquery(Long.class);
Root<EntityTable1> subRoot = sub.from(EntityTable1.class);
Path<EntityTable2> pathTable2 = root.get("field4");
Predicate conditionSubquery = cb.equal(subRoot.get("field1t1"), pathTable2.get("field1t2"));

// Select on the subquery  
sub.select(cb.count(subRoot)).where(conditionSubquery);

// Multi-select on main TupleQuery
cq.multiselect(root, sub.getSelection());

当我运行上述jpa代码时，得到的结果是：

SELECT
  t0.field1 AS a1, 
  t0.field2 AS a2, 
  t0.field3 AS a3, 
  t0.field4 AS a4, 
  t0.field5 AS a5, 
  COUNT(t1.id)

FROM 
  table0 t0, 
  table2 t2, 
  table1 t1

因此，基本上不执行子查询，甚至不考虑“condicinsubquery” predicate 。
我做错什么了？我测试了sql，它按预期工作。
我使用的是jpa的eclipselink 2.7实现。