Question

Let’s say that I have 2 columns:

_______________
| id1  |  id2 |
|------|------|
|  1   |   2  |
|  2   |   1  |
|  3   |   4  |
|  4   |   1  |
|  4   |   3  |
|  1   |   4  |
      ...

If I have a row id1 = 1 and id2 = 2, there will be a row somewhere after it which will be id1 = 2 and id2 = 1.

How can I get the top of each (id1, id2) pair?

---

Update:

The result from the example should be:

row1: 1 | 2
row2: 3 | 4
row3: 4 | 1

Answer 1

Linking back to the original table depends a lot on what is in that table that causes the order. Lets says you have a date field or a sequence field and you want the smallest value…

In this example I assume seqField is unique by row

Then it looks like this:

--drop table #test

 SELECT 1 as seqField, 1 as id1, 2 as id2
 INTO #test
 UNION ALL
 SELECT 2,2,1
 UNION ALL
 SELECT 3,3,4
 UNION ALL
 SELECT 4,4,1
 UNION ALL
 SELECT 5,4,3
 UNION ALL
 SELECT 6,1,4 
 UNION ALL
 SELECT 7,10,20;

WITH norm AS 
( 
   SELECT CASE WHEN id1 > id2 THEN id1 ELSE id2 END as a,
          CASE WHEN id1 < id2 THEN id1 ELSE id2 END as b, * 
   FROM #test
), setList AS
(
  SELECT DISTINCT a, b, min(seqField) as s
  FROM norm
  GROUP BY a, b
)
SELECT #test.* 
FROM #test
JOIN setList ON #test.seqField = setList.s

This gives

seq id1 id2
1   1   2
3   3   4
4   4   1
7   10  20

As sqlchan points out seqField can be replaced with %%physloc%% if you don’t have an existing column to use.