[10g] Simple question on selecting a column in same record as a min value

Question

I have the following sample data to illustrate my question:
CREATE TABLE     t_1
(     c_a     VARCHAR2(1)
,     c_b     NUMBER(2)
,     c_c     DATE
INSERT INTO     t_1
VALUES('A',1,SYSDATE);
INSERT INTO     t_1
VALUES('A',2,SYSDATE+1);
INSERT INTO     t_1
VALUES('A',3,SYSDATE+2);
INSERT INTO     t_1
VALUES('A',4,SYSDATE+3);
INSERT INTO     t_1
VALUES('B',3,SYSDATE+6);
INSERT INTO     t_1
VALUES('B',4,SYSDATE+7);
INSERT INTO     t_1
VALUES('B',5,SYSDATE+8);I have the following query which allows me to select the value of c_c that corresponds with the minimum value of c_b grouped by c_a. In other words, my goal is: for each value in c_a, I want the value of c_c that corresponds with the minimum value of c_b.
SELECT     t1.c_a
,     t1.c_c
FROM     t_1 t1
     SELECT      c_a
     ,      MIN(c_b)     AS min_c_b
     FROM      t_1
     GROUP BY c_a
     ) t2
WHERE     t1.c_a     = t2.c_a
AND     t1.c_b     = t2.min_c_b
;This query works, but I'm wondering if there is a more efficient way to do this. Something without a subquery? I can't come up with anything else... It just gets a bit ugly with my real data, as there are probably 10 columns of data instead of 3. (Still only 1 minimum value though).

Answer 1

Hi,
Since you're using Oracle 10, you can do this:
SELECT       c_a
,       MIN (c_c) KEEP (DENSE_RANK FIRST ORDER BY c_b)     AS c_of_min_b
FROM       t_1
GROUP BY  c_a
;This won't work in earlier versions.
What do you want in case of a tie?
For example, if we add one more row to the sample data:
INSERT INTO     t_1 (c_a,  c_b,  c_c)
             VALUES ('A',  1,    DATE '2001-01-01');Now there are two rows, one dated today and one dated 2001, that have an equal claim to being the "first" in group 'A', since they have the exact same value of c_b.
The query above would display the row from 2001, that is, the earliest of the contenders. That's what MIN means in the query above. Instead of MIN, you can use any aggregate function, but, of course, that function has to work on your data type.
If you want to show all the contenders in case of a tie, then you can stick with your original query, or you could use the analytic RANK function.
WITH     got_rnk          AS
     SELECT     c_a, c_c
     ,     RANK () OVER ( PARTITION BY  c_a
                           ORDER BY          c_b
                    )   AS rnk
     FROM    t_1
SELECT     c_a, c_c
FROM     got_rnk
WHERE     rnk     = 1
;This approach requires a sub-query, because analytic functions (like RANK) are computed after the WHERE clause is applied. If we want to use the results of an anlytic func tion in a WHERE clause, we have to compute the function in a suib-query first.