将varchar值转换为int时转换失败

Question

Microsoft SQL Server 2008(SP1),出现意外的"转换失败"错误.

不太清楚如何描述这个问题,所以下面是一个简单的例子.CTE使用搜索条件提取某些ID的数字部分,以确保数字部分实际存在.然后使用CTE查找最低的未使用序列号(种类):

CREATE TABLE IDs (ID CHAR(3) NOT NULL UNIQUE);

INSERT INTO IDs (ID) VALUES ('A01'), ('A02'), ('A04'), ('ERR');

WITH ValidIDs (ID, seq)
AS 
(
 SELECT ID, CAST(RIGHT(ID, 2) AS INTEGER)
   FROM IDs 
  WHERE ID LIKE 'A[0-9][0-9]'
)
SELECT MIN(V1.seq) + 1 AS next_seq
  FROM ValidIDs AS V1
 WHERE NOT EXISTS (
                   SELECT * 
                     FROM ValidIDs AS V2
                    WHERE V2.seq = V1.seq + 1
                  );

错误是'将varchar值'RR'转换为数据类型int时转换失败.

我无法理解为什么ID = 'ERR'应该考虑转换值,因为谓词ID LIKE 'A[0-9][0-9]'应该从结果集中删除了无效的行.

当基表用等效的CTE代替时,问题就消失了,即

WITH IDs (ID)
AS
(
 SELECT 'A01'
 UNION ALL 
 SELECT 'A02'
 UNION ALL 
 SELECT 'A04'
 UNION ALL 
 SELECT 'ERR' 
),
ValidIDs (ID, seq)
AS 
(
 SELECT ID, CAST(RIGHT(ID, 2) AS INTEGER)
   FROM IDs 
  WHERE ID LIKE 'A[0-9][0-9]'
)
SELECT MIN(V1.seq) + 1 AS next_seq
  FROM ValidIDs AS V1
 WHERE NOT EXISTS (
                   SELECT * 
                     FROM ValidIDs AS V2
                    WHERE V2.seq = V1.seq + 1
                  );

为什么基表会导致此错误？这是一个已知的问题？

---

UPDATE @sgmoore:不,在一个CTE中进行过滤而在另一个CTE中进行过滤仍会导致相同的错误,例如

WITH FilteredIDs (ID)
AS 
(
 SELECT ID
   FROM IDs 
  WHERE ID LIKE 'A[0-9][0-9]'

), 
ValidIDs (ID, seq)
AS 
(
 SELECT ID, CAST(RIGHT(ID, 2) AS INTEGER)
   FROM FilteredIDs 
)
SELECT MIN(V1.seq) + 1 AS next_seq
  FROM ValidIDs AS V1
 WHERE NOT EXISTS (
                   SELECT * 
                     FROM ValidIDs AS V2
                    WHERE V2.seq = V1.seq + 1
                  );

Answer 1

这是一个错误，并且已经被 Erland Sommarskog 报告为SQL Server 不应引发不合逻辑的错误（正如我所说，很难描述这个错误！）。

SQL Server 可编程性团队的回应是：“问题在于，由于在查询执行期间推送谓词/表达式，而不考虑查询的逻辑结果，SQL Server 过于急切地引发错误。”

我现在已经投票赞成修复，请大家也这样做:)