用值替换 NULL

Question

这是我的挑战： 我有一个日志表，每次更改记录时都会添加一条新记录，但会为每条记录中的每个未更改值放置一个 NULL 值。换句话说，仅设置了更改的值，每行中其余未更改的字段仅具有 NULL 值。 现在我想用它上面的不是 NULL 值的值替换每个 NULL 值，如下所示：

源表：Task_log

ID  Owner       Status      Flag
1   Bob         Registrar   T
2   Sue         NULL        NULL
3   NULL        NULL        F
4   Frank       Admission   T
5   NULL        NULL        F
6   NULL        NULL        T

所需的输出表：Task_log

ID  Owner       Status      Flag
1   Bob         Registrar   T
2   Sue         Registrar   T
3   Sue         Registrar   F
4   Frank       Admission   T
5   Frank       Admission   F
6   Frank       Admission   T

如何编写一个查询来生成所需的输出表？

Answer 1

SQLServer 2012新增的一个窗口函数是FIRST_VALUE，名字很直接，可以通过OVER子句进行分区，使用前需要将数据块中的每一列进行划分，一个块用于找到值时开始列。

With Block As (
  Select ID
       , Owner
       , OBlockID = SUM(Case When Owner Is Null Then 0 Else 1 End) 
                    OVER (ORDER BY ID)
       , Status
       , SBlockID = SUM(Case When Status Is Null Then 0 Else 1 End) 
                    OVER (ORDER BY ID)
       , Flag
       , FBlockID = SUM(Case When Flag Is Null Then 0 Else 1 End) 
                    OVER (ORDER BY ID)
  From   Task_log
)
Select ID
     , Owner = FIRST_VALUE(Owner) OVER (PARTITION BY OBlockID ORDER BY ID)
     , Status = FIRST_VALUE(Status) OVER (PARTITION BY SBlockID ORDER BY ID)
     , Flag = FIRST_VALUE(Flag) OVER (PARTITION BY FBlockID ORDER BY ID)
FROM   Block

SQLFiddle演示

UPDATE 查询很容易派生

Answer 2

正如我在评论中提到的，我会尝试修复创建记录的过程，而不是修复垃圾数据。如果这不是一个选项，下面的代码应该会让你指向正确的方向。

UPDATE   t1
set      t1.owner = COALESCE(t1.owner, t2.owner),
         t1.Status = COALESCE(t1.status, t2.status),
         t1.Flag = COALESCE(t1.flag, t2.flag)
FROM     Task_log as t1
INNER JOIN    Task_log as t2
         ON t1.id = (t1.id + 1)
where    t1.owner is null 
         OR t1.status is null 
         OR t1.flag is null

Answer 3

我能想到几种方法。

您可以将COALESCE 与数组聚合函数结合使用。不幸的是，它看起来不像 SQL Server 本身支持 array_agg （虽然有些 nice people have developed some workarounds）。

您还可以对每一列使用子选择。

SELECT id,
       (SELECT TOP 1 FROM (SELECT owner FROM ... WHERE id = outer_id AND owner IS NOT NULL order by ID desc )) AS owner,
       -- other columns

你也可以用窗口函数做一些事情。

Answer 4

一个普通的解决方案是：

select id , owner , coalesce(owner, ( select owner from t t2 where id = (select max(id) from t t3 where id < t1.id and owner is not null)) ) as new_owner , flag , coalesce(flag, ( select flag from t t2 where id = (select max(id) from t t3 where id < t1.id and flag is not null)) ) as new_flag from t t1

相当低效，但应该适用于大多数 DBMS