树结构广泛用在各类分级管理设计中。但他的展现方式也是很让人头疼的事情。比如展开一个靠id和parentid建立关系的分级树，SQL2005已经可以用CTE来递归查询。我们看如下测试代码：
set nocount on
print '--SQL2005 CTE 树结构测试
'
declare @t table ( id varchar ( 10) , pid varchar ( 10), name varchar ( 10))
insert into @t values ( 'a' ,null, '000' )
insert into @t values ( 'b' , 'a' , '111' )
insert into @t values ( 'c' , 'b' , '222' )
insert into @t values ( 'd' , 'b' , '333' )
insert into @t values ( 'f' , 'c' , '444' )
insert into @t values ( 'e' , 'c' , '555' )
; with t ( id, name , pid, path )
as (
select a. id, a. name , a. pid, cast ( a. id as varchar ( 20)) as path
from @t as a
where pid is null
union all
select a. id, a. name , a. pid, cast ( path + '>' + a. id as varchar ( 20))
from @t a
join t as b
on a. pid = b. id
)
select * from t
/*
--SQL2005 CTE 树结构测试
id name pid path
---------- ---------- ---------- --------------------
a 000 NULL a
b 111 a a>b
c 222 b a>b>c
d 333 b a>b>d
f 444 c a>b>c>f
e 555 c a>b>c>e
*/
set nocount off
当你还不知道这个写法的时候可能会有点激动，但是别忘了他是递归，性能并不比以前的循环+表变量 好。我用他展开一个26000条数据，最深5级的权限组分级表时，10分钟后仍然没有查询结束。而我尝试用CLR写扩展函数来实现同样的效果，展开同样的结构，只需要不到2秒 .
下面我们看这个clr的写法，主要思路是提前在内存中build一个链表，可以返回任意节点的路径。当然和之前的几个clr的处理并发方式一样，需要维护一个并发key。
using System;
using System.Data;
using System.Data.SqlClient;
using System.Data.SqlTypes;
using Microsoft.SqlServer.Server;
public partial class UserDefinedFunctions
{
private class TreeNode
{
public TreeNode(string id)
{
this ._id = id;
}
private string _id;
public TreeNode parentNode=null ;
public override string ToString()
{
System.Text.StringBuilder path = new System.Text.StringBuilder ();
path.Append(this ._id);
TreeNode parent = this .parentNode;
while (parent != null )
{
path.Insert(0, ">" );
path.Insert(0, parent._id);
parent = parent.parentNode;
}
return path.ToString();
}
}
private class TreeNodeCollection
{
private System.Collections.Generic.Dictionary <string , TreeNode > _nodeList = new System.Collections.Generic.Dictionary <string , TreeNode >();
public TreeNode AddNode(string id,TreeNode node)
{
_nodeList.Add(id, node);
return node;
}
public TreeNode GetNode(string id)
{
if (_nodeList.ContainsKey(id))
return _nodeList[id];
else
return null ;
}
}
static System.Collections.Generic.Dictionary <string , TreeNodeCollection > _tempNodeList = new System.Collections.Generic.Dictionary <string , TreeNodeCollection >();
[Microsoft.SqlServer.Server.SqlFunction ]
public static SqlBoolean TreeBuilder(SqlString key, SqlString id,SqlString pid)
{
if (key.IsNull || id.IsNull) return true ;
TreeNodeCollection nodeList = _tempNodeList[key.Value];
TreeNode node = nodeList.GetNode(id.Value);
if (node == null )
{
node = new TreeNode (id.Value);
nodeList.AddNode(id.Value,node);
}
if (pid.IsNull) return true ;
TreeNode pnode = nodeList.GetNode(pid.Value);
if (pnode == null )
{
pnode = new TreeNode (pid.Value);
nodeList.AddNode(pid.Value, pnode);
}
node.parentNode = pnode;
return true ;
}
[Microsoft.SqlServer.Server.SqlFunction ]
public static SqlString GetTreePath(SqlString key, SqlString id)
{
if (key.IsNull || id.IsNull) return "Null" ;
TreeNodeCollection nodeList = _tempNodeList[key.Value];
TreeNode node = nodeList.GetNode(id.Value);
if (node == null ) return "No Node" ;
return node.ToString();
}
[Microsoft.SqlServer.Server.SqlFunction ]
public static SqlBoolean InitKey(SqlString key)
{
try
{
_tempNodeList.Add(key.Value, new TreeNodeCollection ());
return true ;
}
catch
{
return false ;
}
}
[Microsoft.SqlServer.Server.SqlFunction ]
public static SqlBoolean DisposeKey(SqlString key)
{
try
{
_tempNodeList.Remove(key.Value);
return true ;
}
catch
{
return false ;
}
}
};
把上面的代码编译为TestTree.dll并复制到服务器的目录中。然后用如下sql语句发布
/*
drop function dbo.xfn_TreeBuilder
drop function dbo.xfn_GetTreePath
drop function dbo.xfn_initTreeKey
drop function dbo.xfn_disposeTreeKey
drop ASSEMBLY TestTreeForSQLCLR
*/
CREATE ASSEMBLY TestTreeForSQLCLR FROM 'E:/sqlclrdata/TestTree.dll' WITH PERMISSION_SET = UnSAFE;
--
go
CREATE FUNCTION dbo. xfn_TreeBuilder
(
@key nvarchar ( 255),
@id nvarchar ( 255),
@pid nvarchar ( 255)
)
RETURNS bit
AS EXTERNAL NAME TestTreeForSQLCLR. [UserDefinedFunctions]. TreeBuilder
go
CREATE FUNCTION dbo. xfn_GetTreePath
(
@key nvarchar ( 255),
@id nvarchar ( 255)
)
RETURNS nvarchar ( 4000)
AS EXTERNAL NAME TestTreeForSQLCLR. [UserDefinedFunctions]. GetTreePath
go
CREATE FUNCTION dbo. xfn_initTreeKey
(
@key nvarchar ( 255)
)
RETURNS bit
AS EXTERNAL NAME TestTreeForSQLCLR. [UserDefinedFunctions]. InitKey
go
CREATE FUNCTION dbo. xfn_disposeTreeKey
(
@key nvarchar ( 255)
)
RETURNS bit
AS EXTERNAL NAME TestTreeForSQLCLR. [UserDefinedFunctions]. DisposeKey
上面包括4个函数，两个负责维护并发key，一个负责建立链表，一个负责返回任意节点的路径。测试sql语句如下
set nocount on
print '--SQL2005 CLR 树结构测试
'
declare @t table ( id varchar ( 10) , pid varchar ( 10), name varchar ( 10))
insert into @t values ( 'a' ,null, '000' )
insert into @t values ( 'b' , 'a' , '111' )
insert into @t values ( 'c' , 'b' , '222' )
insert into @t values ( 'd' , 'b' , '333' )
insert into @t values ( 'f' , 'c' , '444' )
insert into @t values ( 'e' , 'c' , '555' )
declare @key varchar ( 40) , @b bit
set @key= newid ()
select @b= dbo. xfn_inittreekey( @key)
select @b= dbo. xfn_treebuilder( @key, id, pid) from @t
select *, cast ( dbo. xfn_gettreepath( @key, id) as varchar ( 20)) as path from @t
select @b= dbo. xfn_disposetreekey( @key)
go
/*
--SQL2005 CLR 树结构测试
id pid name path
---------- ---------- ---------- --------------------
a NULL 000 a
b a 111 a>b
c b 222 a>b>c
d b 333 a>b>d
f c 444 a>b>c>f
e c 555 a>b>c>e
*/
set nocount off
这个CLR函数的速度快是因为他牺牲了一部分空间节省了时间，只需要两次全表扫描，时间复杂度应该是O(n)级别的，所以比递归快了很多。目前我还没有发现可以不用提前建立树节点链表集合的方法，所以除了需要调用维护key的两个函数外，还需要调用xfn_treebuilder和 xfn_gettreepath这两个clr函数。