ORA-38104: ON 子句中引用的列无法更新

Question

我有一个带有删除标志的简单表（记录应该在此列中更新而不是删除）：

create table PSEUDODELETETABLE
(
  ID        NUMBER(8) not null, -- PKEY
  NAME      VARCHAR2(50) not null,
  ISDELETED NUMBER(1) default 0 not null
)

插入新记录时，我必须检查是否已经有一条记录与主键匹配但 ISDELETED = 1。在这种情况下，我必须将 ISDELETED 更改为 0 并更新其他列。因此我使用以下合并语句：

merge into ET.PSEUDODELETETABLE TARGET
using (select 1 as ID, 'Horst' as NAME from sys.dual) SOURCE
on (TARGET.ISDELETED = 1 and SOURCE.ID = TARGET.ID)
when matched then
  update set ISDELETED = 0, NAME = SOURCE.NAME
when not matched then
  insert values (SOURCE.ID, SOURCE.NAME, 0);

在 Sql-Server 上效果很好，但 Oracle 说：

ORA-38104: Columns referenced in the ON Clause cannot be updated: TARGET.ISDELETED

如果存在 IDELETED = 0 的匹配记录，我希望主键违规作为例外，这就是为什么我不能将“TARGET.ISDELETED = 1”从 on-clause 移动到 update-statement。

Answer 1

我怀疑在这种情况下你最好使用先拍后看算法。

取决于您期望的更常见的情况，或者：

• 更新，如果没有更新行，插入；或
• 插入，如果存在密钥违规，请更新。

Answer 2

与接受的响应相反，实际上有一种方法可以解决这个问题：将有问题的位从 ON 子句中移到 update 语句的 WHERE 子句中：

merge into ET.PSEUDODELETETABLE TARGET
using (select 1 as ID, 'Horst' as NAME from sys.dual) SOURCE
on (SOURCE.ID = TARGET.ID)
when matched then
  update 
      set ISDELETED = 0, 
      NAME = SOURCE.NAME
  where TARGET.ISDELETED = 1 -- Magic!
when not matched then
  insert 
      values (SOURCE.ID, SOURCE.NAME, 0);

Answer 3

“现在可以通过在这些语句上使用 WHERE 子句来进行条件插入和更新。” http://www.oracle-base.com/articles/10g/merge-enhancements-10g.php

Answer 4

我们还需要考虑以下场景，

如果有与IDELETED = 0 匹配的记录，我希望主键违规作为例外，这就是为什么我不能将“TARGET.ISDELETED = 1”从 on-clause 移动到 update-statement。

所以确切的解决方案如下，

begin 
    update ET.PSEUDODELETETABLE set ISDELETED = 0, NAME = 'Horst' 
    where ISDELETED = 1 and ID = 1; 
    if (sql%rowcount = 0) then 
        insert into ET.PSEUDODELETETABLE values (1, 'Horst', 0); 
    end if; 
end;

Answer 5

将列放在某个表达式中并重命名它似乎有效。在下面的例子中，ISDELETED_ 和 ISDELETED 实际上是同一个东西：

merge into (
  select nvl(ISDELETED, ISDELETED) as ISDELETED_, ISDELETED, ID, 
  from ET.PSEUDODELETETABLE
) TARGET
using (select 1 as ID, 'Horst' as NAME from sys.dual) SOURCE
on (TARGET.ISDELETED_ = 1 and SOURCE.ID = TARGET.ID) -- Use the renamed version here
when matched then
  update set ISDELETED = 0, NAME = SOURCE.NAME       -- Use the original version here
when not matched then
  insert values (SOURCE.ID, SOURCE.NAME, 0);

注意：

• 只是重命名不起作用。解析器似乎足够“聪明”，可以检测到它仍然是同一列。但是重命名并把它放在一个“愚蠢的”表达式中比解析器更聪明。
• 这显然是有代价的。索引在重命名的列上可能不容易使用，请检查执行计划。在这个特定的例子中，它可能会起作用
• Oracle 将来可能会“修复”此问题（并使 ORA-38104 检测更加一致），因此此解决方法可能会失效。

这似乎也有效，但绝对不允许任何合理的索引使用（请再次检查您的 Oracle 版本）：

merge into ET.PSEUDODELETETABLE TARGET
using (select 1 as ID, 'Horst' as NAME from sys.dual) SOURCE
on ((select TARGET.ISDELETED from dual) = 1 and SOURCE.ID = TARGET.ID)
when matched then
  update set ISDELETED = 0, NAME = SOURCE.NAME
when not matched then
  insert values (SOURCE.ID, SOURCE.NAME, 0);

即使这样也有效（这引发了对整个 ORA-38104 检查的严重怀疑）！

merge into ET.PSEUDODELETETABLE TARGET
using (select 1 as ID, 'Horst' as NAME from sys.dual) SOURCE
on ((TARGET.ISDELETED, 'dummy') = ((1, 'dummy')) and SOURCE.ID = TARGET.ID)
when matched then
  update set ISDELETED = 0, NAME = SOURCE.NAME
when not matched then
  insert values (SOURCE.ID, SOURCE.NAME, 0);

I have blogged about these workarounds (and execution plans) here.

Answer 6

这不就行了吗？

merge into (select * from ET.PSEUDODELETETABLE where ISDELETED = 1) TARGET
using (select 1 as ID, 'Horst' as NAME from sys.dual) SOURCE
on (SOURCE.ID = TARGET.ID)
when matched then
  update set ISDELETED = 0, NAME = SOURCE.NAME
when not matched then
  insert values (SOURCE.ID, SOURCE.NAME, 0);