导航

[SQL]合并字符串

Posted on 2014-04-23 20:59  beeone  阅读(3621)  评论(0编辑  收藏  举报
--带符号合并行列转换

--有表t,其数据如下:
/*
  a b
  1 1
  1 2
  1 3
  2 1
  2 2
  3 1
--如何转换成如下结果:
  a b
  1 1,2,3
  2 1,2
  3 1 
*/
drop table tb
create table tb
(
   a int,
   b int
)
insert into tb(a,b) values(1,1)
insert into tb(a,b) values(1,2)
insert into tb(a,b) values(1,3)
insert into tb(a,b) values(2,1)
insert into tb(a,b) values(2,2)
insert into tb(a,b) values(3,1)
go

if object_id('pubs..f_hb') is not null
   drop function f_hb
go

--创建一个合并的函数
create function f_hb(@a int)
returns varchar(8000)
as
begin
  declare @str varchar(8000)
  set @str = ''
  select @str = @str + ',' + cast(b as varchar) from tb where a = @a 
  set @str = right(@str , len(@str) - 1)
  return(@str)
End
go

--调用自定义函数得到结果:
select distinct a ,dbo.f_hb(a) as b from tb

drop table tb

--结果
/*
a           b     
----------- ------
1           1,2,3
2           1,2
3           1

(所影响的行数为 3 行)
*/

/*
多个前列的合并
数据的原始状态如下:
ID  PR   CON  OP    SC 
001 p    c    差    6
001 p    c    好    2
001 p    c    一般  4
002 w    e    差    8
002 w    e    好    7
002 w    e    一般  1
===========================
用SQL语句实现,变成如下的数据
ID  PR   CON  OPS
001 p    c    差(6),好(2),一般(4)
002 w    e    差(8),好(7),一般(1)
*/

if object_id('pubs..tb') is not null
   drop table tb
go

create table tb
(
id varchar(10),
pr varchar(10),
con varchar(10),
op varchar(10),
sc int
)
 
insert into tb(ID,PR,CON,OP,SC) values('001', 'p',    'c',    '',    6)
insert into tb(ID,PR,CON,OP,SC) values('001', 'p',    'c',    '',    2)
insert into tb(ID,PR,CON,OP,SC) values('001', 'p',    'c',    '一般',  4)
insert into tb(ID,PR,CON,OP,SC) values('002', 'w',    'e',    '',    8)
insert into tb(ID,PR,CON,OP,SC) values('002', 'w',    'e',    '',    7)
insert into tb(ID,PR,CON,OP,SC) values('002', 'w',    'e',    '一般',  1)
go

if object_id('pubs..test') is not null
   drop table test
go
select ID,PR,CON , OPS = op + '(' + cast(sc as varchar(10)) + ')' into test from tb

--创建一个合并的函数
if object_id('pubs..f_hb') is not null
   drop function f_hb
go
create function f_hb(@id varchar(10),@pr varchar(10),@con varchar(10))
returns varchar(8000)
as
begin
  declare @str varchar(8000)
  set @str = ''
  select @str = @str + ',' + cast(OPS as varchar) from test where id = @id and @pr = pr and @con = con
  set @str = right(@str , len(@str) - 1)
  return(@str)
End
go

--调用自定义函数得到结果:
select distinct id ,pr , con , dbo.f_hb(id,pr,con) as OPS from test

drop table tb
drop table test

--结果
/*
id         pr         con        OPS                
---------- ---------- ---------- -------------------
001        p          c          差(6),好(2),一般(4)
002        w          e          差(8),好(7),一般(1)

(所影响的行数为 2 行)
*/


if object_id('pubs..b') is not null
   drop table b
go
create table b
(col varchar(20))

insert b values ('a')
insert b values ('b')
insert b values ('c')
insert b values ('d')
insert b values ('e')


declare @sql varchar(1024)
set @sql=''
select @sql=@sql+b.col+',' from (select col from b) as b
set @sql='select '''+@sql+''''
exec(@sql)

--解答二
create table A(SID int,   F1 int,  F2 int)
insert into A values(1,    2,    3)
insert into A values(2,    2,    1)
insert into A values(3,    3,    2)

create table B(BID int, SID int,  T int,  V int)
insert into B values(1,    1,     1,    4)
insert into B values(2,    1,     2,   2)
insert into B values(3,    2,     1,    3)
insert into B values(4,    2,     2,    4)
insert into B values(5,    3,     1,    5)
insert into B values(6,    3,     2,    2)
go

--静态SQL,即t的值固定为1,2
select sid,f1,f2,
  max(case when t = 1 then v else 0 end) v1,
  max(case when t = 2 then v else 0 end) v2
from
(
  select a.*,b.t,b.v from a,b where a.sid = b.sid
) t
group by sid,f1,f2

--动态sql,即T的值不固定
declare @sql varchar(8000)
set @sql = 'select SID ,  F1 , F2'
select @sql = @sql + ' , sum(case t when ''' + cast(t as varchar) + ''' then v else 0 end) [V' + cast(t as varchar) + ']'
from (select distinct t from (select a.sid,a.f1,a.f2,b.t,b.v from a,b where a.sid = b.sid) m) as a
set @sql = @sql + ' from (select a.sid,a.f1,a.f2,b.t,b.v from a,b where a.sid = b.sid) m group by SID ,  F1 , F2'
exec(@sql)

drop table A,B

/*
sid         f1          f2          v1          v2          
----------- ----------- ----------- ----------- ----------- 
1           2           3           4           2
2           2           1           3           4
3           3           2           5           2

(所影响的行数为 3 行)

SID         F1          F2          V1          V2          
----------- ----------- ----------- ----------- ----------- 
1           2           3           4           2
2           2           1           3           4
3           3           2           5           2
*/