Dynamic Pivot

最近碰巧读到了关于sql server pivot 的两篇文章。详见：

1 Cross Tabs and Pivots, Part 1 – Converting Rows to Columns

2 Cross Tabs and Pivots, Part 2 - Dynamic Cross Tabs

这两篇文章对sql server里面行转列的两种技术cross tabs和pivot进行了分析和比较。然后着重介绍了动态cross tab的技术。其实这两种旋转在日常中用到的蛮多的，但是动态的旋转我以前考虑过，一直没有去实现，读了这两篇文章,收获颇丰。这两篇文章没有对动态pivot进行分析。其实动态pivot和动态cross tab原理上是一样的，只需要找到变化的部分，然后把这部分动态转换出来就可以了。本篇文章介绍动态pivot。

首先是填充样例数据表，这两篇文章中有介绍，我就借用下：

 SELECT TOP 1000000
        RowNum       = IDENTITY(INT,1,1),
        SomeInt      = ABS(CHECKSUM(NEWID()))%50000+1,
        SomeLetters2 = CHAR(ABS(CHECKSUM(NEWID()))%26+65)
                     + CHAR(ABS(CHECKSUM(NEWID()))%26+65),
        SomeCSV      = CAST('Part01,Part02,Part03,Part04,Part05,Part06,Part07,Part08,Part09,Part10' AS VARCHAR(80)),
        SomeMoney    = CAST(ABS(CHECKSUM(NEWID()))%10000 /100.0 AS MONEY),
        SomeDate     = CAST(RAND(CHECKSUM(NEWID()))*3653.0+36524.0 AS DATETIME),
        SomeHex12    = RIGHT(NEWID(),12)
   INTO dbo.JBMTest
   FROM Master.dbo.SysColumns t1
  CROSS JOIN Master.dbo.SysColumns t2 
--===== Add the Primary Key
  ALTER TABLE dbo.JBMTest
        ADD PRIMARY KEY CLUSTERED (RowNum)

以上填充了一百万条随机数据到表JBMTest中，这个表中有时间列SomeDate,数据列SomeMoney,还有标志列SomeLetters2。要求是对所有的SomeLetters2，按年月统计somemoney。在结果中所有的时间显示为列。

构造动态sql语句的前提是你首先必须根据你的业务逻辑写出静态sql 语句，然后针对变化的部分，找出替代变化的部分的脚本。首先找出实现结果的静态脚本，如下：

select SomeLetters2,
--dynamic part 1
coalesce([Dec 2005], 0) [Dec 2005],
coalesce([Jan 2006], 0) [Jan 2006],
coalesce([Feb 2006], 0) [Feb 2006],
coalesce([Mar 2006], 0) [Mar 2006],
coalesce([Apr 2006], 0) [Apr 2006],
coalesce([ALL], 0) [ALL]
-- start part
from (
	select   case when grouping(SomeLetters2)  = 1 then 
			'1-Total' else SomeLetters2 end as  SomeLetters2, 
			case when GROUPING( DATEADD(mm, datediff(mm, 0, SomeDate), 0)) = 1 then'ALL' else
					stuff(convert(char(11), DATEADD(mm, datediff(mm, 0, SomeDate), 0), 100), 4, 3, '')  end as dt,
		SUM(somemoney) sm
	from dbo.jbmtest
	where SomeDate >=  '12-5-2005' 
	and SomeDate <= '4-2-2006'  and SomeLetters2 in ('AI', 'AJ', 'AC')
	group by SomeLetters2, DATEADD(mm, datediff(mm, 0, SomeDate), 0) with cube
) as d
pivot(
	sum(sm)
	for dt in(
	--dynamic part 2
	[Dec 2005],[Jan 2006],[Feb 2006],[Mar 2006],[Apr 2006],[ALL]
	)
) as p
order by SomeLetters2

如上所示，动态的部分有2个地方，一个是在select的list里面，这里是选出最终的列；另外是在pivot语句的for语句的in list。其实两个地方无非是列出给定2个日期间的所有月。对于给点的两个日期，找出这两个日期间的以月为单位的时间，只需要使用到dateadd函数和数字表，如下：

declare @startDate datetime = '2005-12-5',
	@endDate datetime = '2006-3-2'
set @endDate = DATEADD(MM, DATEDIFF(mm,  0, @enddate)+1, 0)	
select N, stuff(convert(char(11), DATEADD(mm, N-1, @startDate), 100), 4, 3, '') dtstr
from dbo.Number
where N <= DATEDIFF(mm, @startDate, @endDate)

以上的结果里面所有的月以行的显示在结果里面。现在需要把这些行转换成我需要的字符串。这个如下：

declare @startDate datetime = '2005-12-5',
	@endDate datetime = '2006-1-2';	

set @endDate = DATEADD(MM, DATEDIFF(mm,  0, @enddate)+1, 0)									
declare @part1 nvarchar(max) = 'select SomeLetters2,' + char(10) ;
declare @temp1 nvarchar(4000) = 'coalesce([tmpRep], 0) [tmpRep],' + char(10);
declare @temp2 nvarchar(4000) = '[tmpRep],';
declare @part2 nvarchar(max) = '';
	
	select @part1 = @part1 + dtpart,  @part2 = @part2 + dtpart2
	from(
		select REPLACE(@temp1, 'tmpRep', d.dtstr) dtpart, REPLACE(@temp2, 'tmpRep', d.dtstr) dtpart2
		from(
			select N, stuff(convert(char(11), DATEADD(mm, N-1, @startDate), 100), 4, 3, '') dtstr
			from dbo.Number
			where N <= DATEDIFF(mm, @startDate, @endDate)
		) as d
	) as d2
	
set @part1 = @part1 + REPLACE(@temp1, 'tmpRep', 'ALL')
set @part2 = @part2 + REPLACE(@temp2, 'tmpRep', 'ALL')

--remove the last chart ','  @part1 have char(10) and ',' in the right, remove 2 chars
set @part1 = LEFT(@part1, len(@part1) - 2)
set @part2 = LEFT(@part2, len(@part2) -1 )

print @part1
print @part2

@part1是select list里面的脚本，如下：

select SomeLetters2,
coalesce([Dec 2005], 0) [Dec 2005],
coalesce([Jan 2006], 0) [Jan 2006],
coalesce([ALL], 0) [ALL]

@part2是pivot for in list里面的脚本，如下：

[Dec 2005],[Jan 2006],[ALL]

这样所有的动态的部分都已经实现了，余下的就是静态的部分了。

declare @part3 nvarchar(max) = '
from (
	select  case when grouping(SomeLetters2)  = 1 then 
			' + quotename('1-Total', '''')  + ' else SomeLetters2 end as  SomeLetters2, 
			case when GROUPING( DATEADD(mm, datediff(mm, 0, SomeDate), 0)) = 1 then' + quotename('ALL', '''')+ ' else
					stuff(convert(char(11), DATEADD(mm, datediff(mm, 0, SomeDate), 0), 100), 4, 3, '''')  end as dt,
		SUM(somemoney) sm
	from dbo.jbmtest
	where SomeDate >=  ' + quotename(@startDate, '''') + ' 
	and SomeDate < ' +quotename(@endDate, '''') + '
	group by SomeLetters2, DATEADD(mm, datediff(mm, 0, SomeDate), 0) with cube
) as d
pivot(
	sum(sm)
	for dt in(tmpRep)
) as p
order by SomeLetters2
';

set @part3 = @part1 +  REPLACE(@part3, 'tmpRep', @part2)
print @part3

exec (@part3)

在静态的部分里面需要注意，条件SomeDate >= '12 -5-2005' and SomeDate <= '4-2-2006' 需要转换，如果直接使用变量，会导致datatime无法转换成字符串类型的错误，使用quotename函数；其二，所有’’需要使用’’’’代替,否则会导致错误。

最后调用exec，可以得到结果。

最终所有的脚本如下：

declare @startDate datetime = '2005-12-5',
	@endDate datetime = '2006-1-2';	

set @endDate = DATEADD(MM, DATEDIFF(mm,  0, @enddate)+1, 0)									
declare @part1 nvarchar(max) = 'select SomeLetters2,' + char(10) ;
declare @temp1 nvarchar(4000) = 'coalesce([tmpRep], 0) [tmpRep],' + char(10);
declare @temp2 nvarchar(4000) = '[tmpRep],';
declare @part2 nvarchar(max) = '';
	
select @part1 = @part1 + dtpart,  @part2 = @part2 + dtpart2
from(
	select REPLACE(@temp1, 'tmpRep', d.dtstr) dtpart, REPLACE(@temp2, 'tmpRep', d.dtstr) dtpart2
	from(
		select N, stuff(convert(char(11), DATEADD(mm, N-1, @startDate), 100), 4, 3, '') dtstr
		from dbo.Number
		where N <= DATEDIFF(mm, @startDate, @endDate)
	) as d
) as d2
	
set @part1 = @part1 + REPLACE(@temp1, 'tmpRep', 'ALL')
set @part2 = @part2 + REPLACE(@temp2, 'tmpRep', 'ALL')

--remove the last chart ','  @part1 have char(10) and ',' in the right, remove 2 chars
set @part1 = LEFT(@part1, len(@part1) - 2)
set @part2 = LEFT(@part2, len(@part2) -1 )

--print @part1
--print @part2
 
declare @part3 nvarchar(max) = '
from (
	select  case when grouping(SomeLetters2)  = 1 then 
			' + quotename('1-Total', '''')  + ' else SomeLetters2 end as  SomeLetters2, 
			case when GROUPING( DATEADD(mm, datediff(mm, 0, SomeDate), 0)) = 1 then' + quotename('ALL', '''')+ ' else
					stuff(convert(char(11), DATEADD(mm, datediff(mm, 0, SomeDate), 0), 100), 4, 3, '''')  end as dt,
		SUM(somemoney) sm
	from dbo.jbmtest
	where SomeDate >=  ' + quotename(@startDate, '''') + ' 
	and SomeDate < ' +quotename(@endDate, '''') + '
	group by SomeLetters2, DATEADD(mm, datediff(mm, 0, SomeDate), 0) with cube
) as d
pivot(
	sum(sm)
	for dt in(tmpRep)
) as p
order by SomeLetters2
';

set @part3 = @part1 +  REPLACE(@part3, 'tmpRep', @part2)
print @part3

exec (@part3)