Oracle 10g 使用REGEXP_SUBSTR 分拆字符串 （转）

SELECT l_count, REGEXP_SUBSTR('add, daddf, dsdf, asdfa, dsfasd, dsfad','[^,]+',1,l_count) AS NAME
  FROM dual
      ,(SELECT LEVEL l_count FROM DUAL CONNECT BY LEVEL<=100)
WHERE l_count <=LENGTH('add, daddf, dsdf, asdfa, dsfasd, dsfad') - LENGTH(REPLACE('add, daddf, dsdf, asdfa, dsfasd, dsfad',','))+1

查询结果为：
lcount   name
1            add
2            daddf
3            dsdf
4            asdfa
5            dsfasd
6             dsfad

一种奇特的字符串拆分方法

create table t_test (id number, names varchar2(200));

insert into t_test values (1,'a1,a2,a3,a4');
insert into t_test values (2,'b1,b2,b3');
insert into t_test values (3,'c1,c2,c3,c4,c5');

目标输出：

ID NAME
--- ----
  1 a1
  1 a2
  1 a3
  1 a4
  2 b1
  2 b2
  2 b3
  3 c1
  3 c2
  3 c3
  3 c4
  3 c5

常规做法：
SELECT id
      ,REGEXP_SUBSTR(names,'[^,]+',1,l) AS NAME
  FROM t_test
      ,(SELECT LEVEL l FROM DUAL CONNECT BY LEVEL<=100)
WHERE l <=LENGTH(names) - LENGTH(REPLACE(names,','))+1
ORDER BY 1,2;

下面这种写法(从ASKTOM看来的)，很奇怪地在T2中可以看到T1的列(如果不用TABLE,CAST,MULTISET就不行), T1和T2两个集合之间没有任何连接条件, 看起来像Correlated Subquery, 但它是写在from的地方:

SELECT id
      ,column_value
  FROM (SELECT id,','||names||',' names FROM t_test) t1     ----- 前后拼上逗号是为了下面定位拆分
      ,TABLE(CAST(MULTISET( SELECT SUBSTR (names            ----- 此处竟然可以看到t1.names
                                          ,INSTR (names, ',', 1, LEVEL  ) + 1
                                          ,INSTR (names, ',', 1, LEVEL+1) - INSTR (names, ',', 1, LEVEL) -1 
                                          ) 
                              FROM DUAL
                            CONNECT BY LEVEL <= LENGTH(names)-LENGTH(REPLACE(names,',',''))-1 
                           )
             AS SYS.ODCIVARCHAR2LIST ) ------ SYS.ODCIVARCHAR2LIST 可以换成任意一个TABLE OF VARCHAR2的嵌套表类型
             ) t2
ORDER BY 1,2;