SELECT
p.prod_id,
p.prod_name,
CASE
WHEN sum(r.rating_num)/count(DISTINCT r.rating_id) IS NULL THEN 0
ELSE round(sum(r.rating_num)/count(DISTINCT r.rating_id))
END AS avg_rating
FROM
product p
LEFT OUTER JOIN
rating r
ON p.prod_id=r.prod_id
GROUP BY
p.prod_id,
p.prod_name
HAVING
CASE
WHEN sum(r.rating_num)/count(DISTINCT r.rating_id) IS NULL THEN 0
ELSE round(sum(r.rating_num)/count(DISTINCT r.rating_id))
END>=1
基于MySQL中的下表。
mysql> desc rating;
+-------------+---------------------+------+-----+---------+----------------+
| Field | Type | Null | Key | Default | Extra |
+-------------+---------------------+------+-----+---------+----------------+
| rating_id | bigint(20) unsigned | NO | PRI | NULL | auto_increment |
| prod_id | bigint(20) unsigned | YES | MUL | NULL | |
| rating_num | int(10) unsigned | YES | | NULL | |
| ip_address | varchar(45) | YES | | NULL | |
| row_version | bigint(20) unsigned | NO | | 0 | |
+-------------+---------------------+------+-----+---------+----------------+
5 rows in set (0.08 sec)
此表rating
与另一个表明显多到一的关系product
(prod_id
是外键引用的主键prod_id
的product
表)。
在这个问题中,我们只对该子句中的CASE
结构感兴趣HAVING
。
以下标准查询,
CriteriaBuilder criteriaBuilder = entityManager.getCriteriaBuilder();
CriteriaQuery<Tuple> criteriaQuery = criteriaBuilder.createTupleQuery();
Root<Product> root = criteriaQuery.from(entityManager.getMetamodel().entity(Product.class));
ListJoin<Product, Rating> prodRatingJoin = root.join(Product_.ratingList, JoinType.LEFT);
List<Expression>> expressions = new ArrayList<Expression>>();
expressions.add(root.get(Product_.prodId));
expressions.add(root.get(Product_.prodName));
Expression<Integer> sum = criteriaBuilder.sum(prodRatingJoin.get(Rating_.ratingNum));
Expression<Long> count = criteriaBuilder.countDistinct(prodRatingJoin.get(Rating_.ratingId));
Expression<Number> quotExpression = criteriaBuilder.quot(sum, count);
Expression<Integer> roundExpression = criteriaBuilder.function("round", Integer.class, quotExpression);
Expression<Integer> selectExpression = criteriaBuilder.<Integer>selectCase().when(quotExpression.isNull(), criteriaBuilder.literal(0)).otherwise(roundExpression);
expressions.add(selectExpression);
criteriaQuery.multiselect(expressions.toArray(new Expression[0]));
expressions.remove(expressions.size() - 1);
criteriaQuery.groupBy(expressions.toArray(new Expression[0]));
criteriaQuery.having(criteriaBuilder.greaterThanOrEqualTo(selectExpression, criteriaBuilder.literal(1)));
List<Tuple> list = entityManager.createQuery(criteriaQuery).getResultList();
for (Tuple tuple : list) {
System.out.println(tuple.get(0) + " : " + tuple.get(1) + " : " + tuple.get(2));
}
根据预期生成以下正确的SQL查询。
select
product0_.prod_id as col_0_0_,
product0_.prod_name as col_1_0_,
case
when sum(ratinglist1_.rating_num)/count(distinct ratinglist1_.rating_id) is null then 0
else round(sum(ratinglist1_.rating_num)/count(distinct ratinglist1_.rating_id))
end as col_2_0_
from
projectdb.product product0_
left outer join
projectdb.rating ratinglist1_
on product0_.prod_id=ratinglist1_.prod_id
group by
product0_.prod_id ,
product0_.prod_name
having
case
when sum(ratinglist1_.rating_num)/count(distinct ratinglist1_.rating_id) is null then 0
else round(sum(ratinglist1_.rating_num)/count(distinct ratinglist1_.rating_id))
end>=1
对于技术角度来看,在上述条件查询中查看以下行。
criteriaQuery.having(criteriaBuilder.greaterThanOrEqualTo(selectExpression, criteriaBuilder.literal(1)));
这个问题的类似之处如下。
createQuery.having(criteriaBuilder.greaterThanOrEqualTo(selectExpression, 1));
在这个问题上看到原来的表达方式是完全一样的:
Expression<Integer> selectExpression = criteriaBuilder.<Integer>selectCase()
.when(quotExpression.isNull(), 0)
.<Integer>otherwise(roundExpression);
尝试将此表达式传递给criteriaBuilder.greaterThanOrEqualTo()
如下。
criteriaQuery.having(criteriaBuilder.greaterThanOrEqualTo(selectExpression, 0));
要特别注意greaterThanOrEqualTo()
上面的第二个参数。这是0
。criteriaBuilder.literal(0)
因此,应该是这个问题所提到的例外。
因此,CriteriaBuilder#literal(T value)
在必要时始终坚持使用文字值,同时在CriteriaBuilder#selectCase()
构造中使用表达式。