Question

I have a table similar to:-

+----+---+---+
| Id | A | B |
+----+---+---+    
| 1  | 1 | 2 |
+----+---+---+
| 2  | 2 | 1 |
+----+---+---+
| 3  | 3 | 4 |
+----+---+---+
| 4  | 0 | 5 |
+----+---+---+
| 5  | 5 | 0 |
+----+---+---+

I want to remove all duplicate pairs of values, regardless of which column contains which value, e.g. after whatever the query might be I want to see:-

+----+---+---+
| Id | A | B |
+----+---+---+    
| 1  | 1 | 2 |
+----+---+---+
| 3  | 3 | 4 |
+----+---+---+
| 4  | 0 | 5 |
+----+---+---+

I’d like to find a solution in Microsoft SQL Server (has to work in <= 2005, though I’d be interested in any solutions which rely upon >= 2008 features regardless).

In addition, note that A and B are going to be in the range 1-100 (but that’s not guaranteed forever. They are surrogate seeded integer foreign keys, however the foreign table might grow to a couple hundred rows max).

I’m wondering whether I’m missing some obvious solution here. The ones which have occurred all seem rather overwrought, though I do think they’d probably work, e.g.:-

• Have a subquery return a bitfield with each bit corresponding to one of the ids and use this value to remove duplicates.
• Somehow, pivot, remove duplicates, then unpivot. Likely to be tricky.

Thanks in advance!

Answer 1

Try:

select min(Id) Id, A, B
from (select Id, A, B from DuplicatesTable where A <= B
      union all
      select Id, B A, A B from DuplicatesTable where A > B) v
group by A, B
order by 1