sql - 通常会出现错误，但在 EXISTS 子查询中运行

Question

以下有问题吗？

DELETE a
FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData a
WHERE EXISTS
        (
        SELECT * 
        FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData b
        WHERE 
            b.[Past28Days] = 1 AND
            a.[Index] = b.[Index]
        HAVING SUM(b.Amount) = 0
        )

我对使用上述脚本感到有点不安的原因是，如果我运行以下脚本，则会出现错误:

SELECT * 
FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData b
WHERE 
    b.[Past28Days] = 1
HAVING SUM(b.Amount) = 0

我明白为什么这个脚本会出错 => 选择没有对任何内容进行分组，因此处理器不喜欢 HAVING 子句中的聚合。

但是作为子查询，这个错误不会发生 - 为什么？这是一个有问题的方法吗？

编辑

最终使用了以下内容:

DELETE a
FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData a
WHERE a.[Index] IN
    (
    SELECT [Index] 
    FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData
    WHERE [Past28Days] = 1
    GROUP BY [Index] 
    HAVING SUM(Amount) = 0
    )

但正如答案中所建议的，通过简单地将 GROUP BY 添加到子查询中，以下内容更具可读性:

DELETE a
FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData a
WHERE EXISTS
    (
    SELECT * 
    FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData b
    WHERE 
            b.[Past28Days] = 1 AND
            a.[Index] = b.[Index]
    GROUP BY b.[Index]
    HAVING SUM(b.Amount) = 0
    )

Answer 1

省略 group by 并仍然执行聚合是合法的，因此 having 仍然是限制结果的一种方式:

select sum(x)
from
(
  select 1 x union all select 2
) a
having sum(x) = 3

Exists() 有效，因为选择列表中的所有内容都被忽略。 Exists() 仅查找行，并在找到行后立即终止。您可以添加 group by b.Index 以使稍后查看代码的任何人都清楚其意图，或者使用内部联接和派生表重写它。

DELETE a
FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData a
INNER JOIN
        (
        SELECT b.[Index]
        FROM WHAnalysis.dbo.tb_r12027dxi_CalculatedData b
        WHERE 
            b.[Past28Days] = 1
        GROUP BY b.[Index]
        HAVING SUM(b.Amount) = 0
        ) b1
   ON a.[Index] = b1.[Index]