复制分层数据时保留父子关系

Question

我们有一个表，表示与实体（称为项目）关联的值树，其中 ParentID 列指行父级的 id 列。id 列是一个自动递增的 IDENTITY 列和主键。根节点的 ParentID 为 0。

我们希望能够克隆给定项目的数据，并让生成的 ParentID 引用复制值的适当新 ID，以满足示例下面描述的限制。

例如，复制下表中 ProjectID 611 的数据：

    id      ProjectID    Value         ParentID
--------------------------------------------------
     1      611           Animal        0
     2      611           Frog          1
    13      611           Cow           1
    14      611           Jersey Cow    13
    25      611           Plant         0
    29      611           Tree          25
    31      611           Oak           29

应该导致：

    id      ProjectID    Value         ParentID
--------------------------------------------------
     1      611           Animal        0
     2      611           Frog          1
    13      611           Cow           1
    14      611           Jersey Cow    13
    25      611           Plant         0
    29      611           Tree          25
    31      611           Oak           29
    32      612           Animal        0
    33      612           Frog          32
    34      612           Cow           32
    35      612           Jersey Cow    34
    36      612           Plant         0
    37      612           Tree          36
    38      612           Oak           37

限制：

• 解决方案必须适用于SQL Server 2005。也就是说，我们不能使用 MERGE（唉）。
• 我们不愿意对 ids 或它们与 ParentID 的比较做出假设；例如，该解决方案原则上应适用于 uniqueid 的 ids/ParentID。
• 我们不想向表中添加额外的列。（我当前的解决方案添加了一个“OldId”列，复制过程在复制行时设置该列。因此，我当前正在使用 INSERT-SELECT 和 UPDATE-FROM 的组合，将 OldId 列加入 ParentID 列以获取新 id .) 我们不想仅仅为了支持此复制操作而在所有分层表中添加 OldId 列。
• 解决方案必须具有合理的性能；我最初的解决方案是一组复杂的递归函数调用和循环，一次处理一个项目。我很快就放弃了那条路！

Answer 1

CTE 可以很好地与 SQL Server 2005 配合使用MERGE，但在 SQL Server 2005 中会出现问题。对于之前的误导性评论，我们深表歉意。

下面展示了如何克隆项目（具有多棵树）并修复起源以将新森林与旧森林分开。请注意，它不依赖于 Id 的任何特定排列，例如它们不需要是密集的、单调递增的……。

-- Sample data.
declare @Projects as Table
  ( Id Int Identity, ProjectId Int, Value VarChar(16), ParentId Int Null );
insert into @Projects ( ProjectId, Value, ParentId ) values
  ( 611, 'Animal', 0 ),
  ( 611, 'Frog', 1 ),
  ( 611, 'Cow', 1 ),
  ( 611, 'Jersey Cow', 3 ),
  ( 611, 'Plant', 0 ),
  ( 611, 'Tree', 5 ),
  ( 611, 'Oak', 6 );
-- Display the raw data.
select * from @Projects;

-- Display the forest.
with IndentedProjects ( Id, ProjectId, Value, ParentId, Level, Path ) as
  ( -- Start with the top level rows.
  select Id, ProjectId, Value, ParentId, 0, Convert( VarChar(1024), Right( '000' + Convert( VarChar(4), Id ), 4 ) )
    from @Projects
    where ParentId = 0
  union all
  -- Add the children one level at a time.
  select P.Id, P.ProjectId, P.Value, P.ParentId, IP.Level + 1, Convert( VarChar(1024), IP.Path + '<' + Right( '000' + Convert( VarChar(4), P.Id ), 4 ) )
    from IndentedProjects as IP inner join
      @Projects as P on P.ParentId = IP.Id
  )
  select Space( Level * 2 ) + Value as [IndentedValue], Id, ProjectId, Value, ParentId, Level, Path
    from IndentedProjects
    order by Path;

-- Clone the project.
declare @OldProjectId as Int = 611;
declare @NewProjectId as Int = 42;
declare @Fixups as Table ( OldId Int, [NewId] Int );
begin transaction -- With suitable isolation since the hierarchy will be invalid until we apply the fixups!
insert into @Projects
  output Inserted.ParentId, Inserted.Id
    into @Fixups
  select @NewProjectId, Value, Id -- Note that we save the old Id in the new ParentId.
    from @Projects as P
    where ProjectId = @OldProjectId;
-- Apply the fixups.
update PNew
  set ParentId = IsNull( FNew.[NewId], 0 )
  -- Output the fixups just to show what is going on.
  output Deleted.Id, Deleted.ParentId as [ParentIdBeforeFixup], Inserted.ParentId as [ParentIdAfterFixup]
  from @Fixups as F inner join
    @Projects as PNew on PNew.Id = F.[NewId] inner join -- Rows we need to fix.
    @Fixups as FOld on FOld.OldId = PNew.ParentId inner join
    @Projects as POld on POld.Id = FOld.OldId left outer join
    @Fixups as FNew on FNew.OldId = POld.ParentId;
commit transaction;

-- Display the forest.
with IndentedProjects ( Id, ProjectId, Value, ParentId, Level, Path ) as
  ( -- Start with the top level rows.
  select Id, ProjectId, Value, ParentId, 0, Convert( VarChar(1024), Right( '000' + Convert( VarChar(4), Id ), 4 ) )
    from @Projects
    where ParentId =0
  union all
  -- Add the children one level at a time.
  select P.Id, P.ProjectId, P.Value, P.ParentId, IP.Level + 1, Convert( VarChar(1024), IP.Path + '<' + Right( '000' + Convert( VarChar(4), P.Id ), 4 ) )
    from IndentedProjects as IP inner join
      @Projects as P on P.ParentId = IP.Id
  )
  select Space( Level * 2 ) + Value as [IndentedValue], Id, ProjectId, Value, ParentId, Level, Path
    from IndentedProjects
    order by Path;