Pivoting question

Question

I have been reading and experimenting with pivoting and found a nice description at:
http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::p11_question_id:766825833740
My question regards the following code:
scott@DEV816>
scott@DEV816> select job,
2 max( decode( deptno, 10, cnt, null ) ) dept_10,
3 max( decode( deptno, 20, cnt, null ) ) dept_20,
4 max( decode( deptno, 30, cnt, null ) ) dept_30,
5 max( decode( deptno, 40, cnt, null ) ) dept_40
6 from ( select job, deptno, count(*) cnt
7 from emp
8 group by job, deptno )
9 group by job
10 /
Why is the max function needed? It seems that the "decode() AS dept_10", and the rest of the lines, only yield a single response, and so I don't see why the max is needed. Thank you.

Answer 1

Hi,
apex wrote:
I have been reading and experimenting with pivoting and found a nice description at:
http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::p11_question_id:766825833740
My question regards the following code:
scott@DEV816>
scott@DEV816> select job,
2 max( decode( deptno, 10, cnt, null ) ) dept_10,
3 max( decode( deptno, 20, cnt, null ) ) dept_20,
4 max( decode( deptno, 30, cnt, null ) ) dept_30,
5 max( decode( deptno, 40, cnt, null ) ) dept_40
6 from ( select job, deptno, count(*) cnt
7 from emp
8 group by job, deptno )
9 group by job
10 /
Why is the max function needed? It seems that the "decode() AS dept_10", and the rest of the lines, only yield a single response, and so I don't see why the max is needed. Thank you.Try removing MAX and see what happens.
Remember the ABC's of GROUP BY:
When you use a GROUP BY clause and/or an aggregate fucntion, then every item in the SELECT clause must be:
(A) an <b>A</b>ggregate function,
(B) one of the "group <b>B</b>y" expressions,
(C) a <b>C</b>onstant, or
(D) something that <b>D</b>epends entirely on the above. (For example, if you "GROUP BY TRUNC(dt)", you can "SELECT TO_CHAR (TRUNC(dt), 'Mon-DD')").
In this case, you want to say "GROUP BY job" in the main query because that will combine the rows onto one row for each job. Given that you need to "GROUP BY job", how are you going to get a value of cnt in the output? Cnt is not one of the GROUP BY expressions, it's not a constant, and it's not dependent on anything else. The only way you can get cnt into the results is to use it in an aggregate function, like MAX. Since there will only be 1 non-NULL value in each group (if that many), it doesn't matter if you use MIN or MAX to get that value. Since it's a NUMBER, you could even use MEDIAN, SUM or AVG, if you really wanted to be cute.
By the way, there's another way to get those same results without uisng a sub-query:
SELECT       job
,       SUM (CASE WHEN deptno = 10 THEN 1 END)     AS dept_10
,       SUM (CASE WHEN deptno = 20 THEN 1 END)     AS dept_20
,       SUM (CASE WHEN deptno = 30 THEN 1 END)     AS dept_30
,       SUM (CASE WHEN deptno = 40 THEN 1 END)     AS dept_40
FROM       scott.emp
GROUP BY  job
;You could use COUNT instead of SUM, above, to show 0 instead of leaving a column empty.
Using the SELECT ... PIVOT feature that was introduced in Oracle 11, you need a sub-query, but only to remove the columns that play no role in the problem:
WITH     got_relevant_columns     AS
     SELECT     job
     ,     deptno
     FROM     scott.emp
SELECT     *
FROM     got_relevant_columns
PIVOT     (     SUM (1)
     FOR     deptno IN ( 10     AS dept_10
                        , 20     AS dept_20
                        , 30     AS dept_30
                        , 40     AS dept_40
;Edited by: Frank Kulash on Sep 29, 2011 11:29 AM