當表建立了物化視圖日志之后，表的DML修改會被記錄到物化視圖日志中，而物化視圖日志則包含了一個修改矢量，來記錄哪個列被修改。
在文章
列的修改矢量可以通過2的N次方來獲得，也就是POWER(2, N)。而N的值，就是列的位置。但是如果存在一些隱藏的列，通過DBA_TAB_COLUMNS來獲取列的位置就存在問題，比如：
SQL> CREATE TABLE T_PERSON OF T_PERSON_TYP
2? (ID PRIMARY KEY);
Table created.
SQL> SELECT COLUMN_ID
2? FROM DBA_TAB_COLUMNS
3? WHERE WNER = USER
4? AND TABLE_NAME = 'T_PERSON'
5? AND COLUMN_NAME = 'AGE';
COLUMN_ID
----------
3
SQL> SELECT INTCOL#
2? FROM SYS.COL$
3? WHERE NAME = 'AGE'
4? AND OBJ# =
5? (SELECT OBJECT_ID
6? FROM DBA_OBJECTS
7? WHERE WNER = USER
8? AND OBJECT_NAME = 'T_PERSON'
9? AND OBJECT_TYPE = 'TABLE');
INTCOL#
----------
5
SQL> CREATE MATERIALIZED VIEW LOG ON T_PERSON;
Materialized view log created.
SQL> INSERT INTO T_PERSON VALUES (1, 'ABC', 18, 'BCD');
1 row created.
SQL> UPDATE T_PERSON SET AGE = 20 WHERE ID = 1;
1 row updated.
SQL> COL CHANGE_VECTOR$$ FORMAT A30
SQL> SELECT * FROM MLOG$_T_PERSON;
ID SNAPTIME$$???? D O CHANGE_VECTOR$$
---------- -------------- - - ------------------------------
1 01-1月-00???? I N FE
1 01-1月-00???? U U 20
SQL> SELECT TO_CHAR(POWER(2, 5), 'XX') FROM DUAL;
TO_
---
20
SQL> SELECT TO_CHAR(POWER(2, 3), 'XX') FROM DUAL;
TO_
---
8
很顯然，物化視圖日志中獲取的列的偏移量是通過SYS.COL$視圖的INTCOL#列獲取的，而不是通過DBA_TAB_COLUMNS視圖。
表SYS.COL$中INTCOL#的值，就是我們要計算的N，得到POWER(2, N)后，通過TO_CHAR轉換為16進制的RAW，就是這個列對應的偏移量。
不過還需要注意一點，CHANGE_VECTOR$$中的偏移量和直接TO_CHAR得到的結果還是有區別的，首先二者的排列順序就有區別：
SQL> DROP TABLE T_PERSON PURGE;
Table dropped.
SQL> DROP TYPE T_PERSON_TYP;
Type dropped.
SQL> CREATE TABLE T_PERSON
2? (ID NUMBER PRIMARY KEY,
3? NAME VARCHAR2(30),
4? ADDR01 VARCHAR2(30),
5? ADDR02 VARCHAR2(30),
6? ADDR03 VARCHAR2(30),
7? ADDR04 VARCHAR2(30),
8? ADDR05 VARCHAR2(30),
9? ADDR06 VARCHAR2(30),
10? ADDR07 VARCHAR2(30),
11? ADDR08 VARCHAR2(30),
12? ADDR09 VARCHAR2(30),
13? ADDR10 VARCHAR2(30),
14? ADDR11 VARCHAR2(30),
15? ADDR12 VARCHAR2(30),
16? ADDR13 VARCHAR2(30),
17? ADDR14 VARCHAR2(30),
18? ADDR15 VARCHAR2(30),
19? ADDR16 VARCHAR2(30),
20? ADDR17 VARCHAR2(30),
21? ADDR18 VARCHAR2(30),
22? AGE NUMBER);
Table created.
SQL> CREATE MATERIALIZED VIEW LOG ON T_PERSON;
Materialized view log created.
SQL> SELECT TO_CHAR(POWER(2, INTCOL#), 'XXXXXX')
2? FROM SYS.COL$
3? WHERE NAME = 'AGE'
4? AND OBJ# =
5? (SELECT OBJECT_ID
6? FROM DBA_OBJECTS
7? WHERE WNER = USER
8? AND OBJECT_NAME = 'T_PERSON'
9? AND OBJECT_TYPE = 'TABLE');
TO_CHAR
-------
200000
SQL> INSERT INTO T_PERSON
2? (ID, NAME, AGE)
3? VALUES (1, 'A', 15);
1 row created.
SQL> UPDATE T_PERSON SET AGE = 18;
1 row updated.
SQL> SELECT * FROM MLOG$_T_PERSON;
ID SNAPTIME$$???? D O CHANGE_VECTOR$$
---------- -------------- - - ------------------------------
1 01-1月-00???? I N FEFFFF
1 01-1月-00???? U U 000020
顯然物化視圖日志中的偏移量是逆向的，好在Oracle的內置函數REVERSE也支持RAW類型，省得自己編寫這個函數了：
SQL> SELECT REVERSE(LTRIM(TO_CHAR(POWER(2, INTCOL#), 'XXXXXX')))
2? FROM SYS.COL$
3? WHERE NAME = 'AGE'
4? AND OBJ# =
5? (SELECT OBJECT_ID
6? FROM DBA_OBJECTS
7? WHERE WNER = USER
8? AND OBJECT_NAME = 'T_PERSON'
9? AND OBJECT_TYPE = 'TABLE');
REVERSE
-------
000002
這里還有問題，REVERSE采用字符串的反轉方法，而這時需要RAW類型的反轉：
SQL> SELECT REVERSE(CAST(LTRIM(TO_CHAR(POWER(2, INTCOL#), 'XXXXXX')) AS RAW(255)))
2? FROM SYS.COL$
3? WHERE NAME = 'AGE'
4? AND OBJ# =
5? (SELECT OBJECT_ID
6? FROM DBA_OBJECTS
7? WHERE WNER = USER
8? AND OBJECT_NAME = 'T_PERSON'
9? AND OBJECT_TYPE = 'TABLE');
REVERSE(CAST(LTRIM(TO_CHAR(POWER(2,INTCOL#),'XXXXXX'))ASRAW(255)))
------------------------------------------------------------------
000020
下面還有一些問題，首先就是TO_CHAR后結果的前綴0問題：
SQL> UPDATE T_PERSON SET ADDR08 = 5;
1 row updated.
SQL> SELECT * FROM MLOG$_T_PERSON;
ID SNAPTIME$$???? D O CHANGE_VECTOR$$
---------- -------------- - - ------------------------------
1 01-1月-00???? I N FEFFFF
1 01-1月-00???? U U 000020
1 01-1月-00???? U U 000400
SQL> SELECT REVERSE(CAST(LTRIM(TO_CHAR(POWER(2, INTCOL#), 'XXXXXX')) AS RAW(255)))
2? FROM SYS.COL$
3? WHERE NAME = 'ADDR08'
4? AND OBJ# =
5? (SELECT OBJECT_ID
6? FROM DBA_OBJECTS
7? WHERE WNER = USER
8? AND OBJECT_NAME = 'T_PERSON'
9? AND OBJECT_TYPE = 'TABLE');
REVERSE(CAST(LTRIM(TO_CHAR(POWER(2,INTCOL#),'XXXXXX'))ASRAW(255)))
------------------------------------------------------------------
0004
在TO_CHAR的時候使用’0X’作為參數可以避免前綴0的問題：
SQL> SELECT REVERSE(CAST(LTRIM(TO_CHAR(POWER(2, INTCOL#), '0XXXXX')) AS RAW(255)))
2? FROM SYS.COL$
3? WHERE NAME = 'ADDR08'
4? AND OBJ# =
5? (SELECT OBJECT_ID
6? FROM DBA_OBJECTS
7? WHERE WNER = USER
8? AND OBJECT_NAME = 'T_PERSON'
9? AND OBJECT_TYPE = 'TABLE');
REVERSE(CAST(LTRIM(TO_CHAR(POWER(2,INTCOL#),'0XXXXX'))ASRAW(255)))
------------------------------------------------------------------
000400
但是另一個麻煩的問題又出現了，就是執行TO_CHAR的時候需要指定多少個X，X的個數如果少了會報錯，個數多了結果又不正確。
其中X的個數由表中總的列數來決定，具體算法為：
SQL> SELECT FLOOR(LOG(256, POWER(2, COUNT(*)))) + 1
2? FROM SYS.COL$
3? WHERE OBJ# =
4? (SELECT OBJECT_ID
5? FROM DBA_OBJECTS
6? WHERE WNER = USER
7? AND OBJECT_NAME = 'T_PERSON'
8? AND OBJECT_TYPE = 'TABLE');
FLOOR(LOG(256,POWER(2,COUNT(*))))+1
-----------------------------------
3
將上面的結果整合在一起：
SQL> SELECT REVERSE(CAST(LTRIM(TO_CHAR(NUM, RPAD('0X', CNT * 2, 'X'))) AS RAW(255)))
2? FROM
3? (
4???? SELECT POWER(2, INTCOL#) NUM,
5???????????? FLOOR(LOG(256, POWER(2, COUNT(*) OVER()))) + 1 CNT,
6???????????? NAME
7???? FROM SYS.COL$
8???? WHERE OBJ# =
9???? (
10???????????? SELECT OBJECT_ID
11???????????? FROM DBA_OBJECTS
12???????????? WHERE WNER = USER
13???????????? AND OBJECT_NAME = 'T_PERSON'
14???????????? AND OBJECT_TYPE = 'TABLE'
15???? )
16? )
17? WHERE NAME = 'ADDR13';
REVERSE(CAST(LTRIM(TO_CHAR(NUM,RPAD('0X',CNT*2,'X')))ASRAW(255)))
-----------------------------------------------------------------
008000
SQL> SELECT REVERSE(CAST(LTRIM(TO_CHAR(NUM, RPAD('0X', CNT * 2, 'X'))) AS RAW(255)))
2? FROM
3? (
4???? SELECT POWER(2, INTCOL#) NUM,
5???????????? FLOOR(LOG(256, POWER(2, COUNT(*) OVER()))) + 1 CNT,
6???????????? NAME
7???? FROM SYS.COL$
8???? WHERE OBJ# =
9???? (
10???????????? SELECT OBJECT_ID
11???????????? FROM DBA_OBJECTS
12???????????? WHERE WNER = USER
13???????????? AND OBJECT_NAME = 'T_PERSON'
14???????????? AND OBJECT_TYPE = 'TABLE'
15???? )
16? )
17? WHERE NAME = 'AGE';
REVERSE(CAST(LTRIM(TO_CHAR(NUM,RPAD('0X',CNT*2,'X')))ASRAW(255)))
-----------------------------------------------------------------
000020
