行列转换(Oracle)
假如有如下表,其中各个i值对应的行数是不定的
SQL> select * from t;
I A D
---------- ---------- -------------------
1 b 2008-03-27 10:55:42
1 a 2008-03-27 10:55:46
1 d 2008-03-27 10:55:30
2 z 2008-03-27 10:55:55
2 t 2008-03-27 10:55:59
I A D
---------- ---------- -------------------
1 b 2008-03-27 10:55:42
1 a 2008-03-27 10:55:46
1 d 2008-03-27 10:55:30
2 z 2008-03-27 10:55:55
2 t 2008-03-27 10:55:59
要获得如下结果,注意字符串需要按照D列的时间排序:
1 d,b,a
2 z,t
2 z,t
这是一个比较典型的行列转换,有好几种实现方法
1.自定义函数实现
create or replace function my_concat(n number)
return varchar2
is
type typ_cursor is ref cursor;
v_cursor typ_cursor;
v_temp varchar2(10);
v_result varchar2(4000):= '';
v_sql varchar2(200);
begin
v_sql := 'select a from t where i=' || n ||' order by d';
open v_cursor for v_sql;
loop
fetch v_cursor into v_temp;
exit when v_cursor%notfound;
v_result := v_result ||',' || v_temp;
end loop;
return substr(v_result,2);
end;
SQL> select i,my_concat(i) from t group by i;
I MY_CONCAT(I)
---------- --------------------
1 d,b,a
2 z,t
return varchar2
is
type typ_cursor is ref cursor;
v_cursor typ_cursor;
v_temp varchar2(10);
v_result varchar2(4000):= '';
v_sql varchar2(200);
begin
v_sql := 'select a from t where i=' || n ||' order by d';
open v_cursor for v_sql;
loop
fetch v_cursor into v_temp;
exit when v_cursor%notfound;
v_result := v_result ||',' || v_temp;
end loop;
return substr(v_result,2);
end;
SQL> select i,my_concat(i) from t group by i;
I MY_CONCAT(I)
---------- --------------------
1 d,b,a
2 z,t
虽然这种方式可以实现需求,但是如果表t的数据量很大,i的值又很多的情况下,因为针对每个i值都要执行一句select,扫描和排序的次数和i的值成正比,性能会非常差。
2.使用sys_connect_by_path
select i,ltrim(max(sys_connect_by_path(a,',')),',') a
from
(
select i,a,d,min(d) over(partition by i) d_min,
(row_number() over(order by i,d))+(dense_rank() over (order by i)) numid
from t
)
start with d=d_min connect by numid-1=prior numid
group by i;
from
(
select i,a,d,min(d) over(partition by i) d_min,
(row_number() over(order by i,d))+(dense_rank() over (order by i)) numid
from t
)
start with d=d_min connect by numid-1=prior numid
group by i;
从执行计划上来看,这种方式只需要扫描两次表,比自定义函数的方法,效率要高很多,尤其是表中数据量较大的时候:
3.使用wm_sys.wm_concat
这个函数也可以实现类似的行列转换需求,但是似乎没有办法做到直接根据另外一列排序,所以需要先通过子查询或者临时表排好序
SQL> select i,wmsys.wm_concat(a) from t group by i;
I WMSYS.WM_CONCAT(A)
---------- --------------------
1 b,a,d
2 z,t
SQL> select i,wmsys.wm_concat(a)
2 from
3 (select * from t order by i,d)
4 group by i;
I WMSYS.WM_CONCAT(A)
---------- --------------------
1 d,b,a
2 z,t
I WMSYS.WM_CONCAT(A)
---------- --------------------
1 b,a,d
2 z,t
SQL> select i,wmsys.wm_concat(a)
2 from
3 (select * from t order by i,d)
4 group by i;
I WMSYS.WM_CONCAT(A)
---------- --------------------
1 d,b,a
2 z,t
执行计划上看,只需要做一次表扫描就可以了,但是这个函数是加密过的,执行计划并不能显示函数内部的操作。
不知道大家还有没有更加高效的实现方式,欢迎指教^_^