Getting the ROWIDs of non-duplicate rows in Oracle 10g

Question

Hi,
given a table TAB1 in which the primary key is composed of these first 3 columns:
================================================
colA____________colB_________________colC____________________colD
================================================
AA-----------------------01-----------------------20080101-----------------------100
BX-----------------------32-----------------------20050325-----------------------366
AA-----------------------01-----------------------20080102-----------------------645
AA-----------------------01-----------------------20080707-----------------------765
AB-----------------------02-----------------------20080806-----------------------336
AB-----------------------02-----------------------20080705-----------------------543
I wish to be able to find the ROWIDs of those rows where colA||colB occurs only once in the table TAB1.
So in the above example, I would want to know the ROWID of the row containing colA=BX and colB=32 as "BX32" occurs only once in this table (AA01 occurs 3 times, AB02 occurs twice so I'm not interested in them).
The following does work, but it is too SLOW:
select ROWID from TAB1 where colA||colB = (select colA||colB from TAB1 HAVING COUNT(*)=1)
Can anyone please suggest an efficient way to find this unique ROWID value in these circumstances?
Edited by: Nemesis on Nov 29, 2008 2:50 AM

Answer 1

Thank you Walter.
This does not give the desired result however. It compares the table, row by row with a 'mirror' of itself and says 'give me the rowid of the row where its mirror does not match itself' (correct me if I am wrong)
(Here I added the '=' to ' AND d.ROWID = d1.ROWID);'
SELECT D.rowid
FROM TAB1 D
WHERE NOT EXISTS (SELECT *
FROM TAB1 d1
WHERE d1.COLA = d.COLA
AND d1.COLB = d.COLB
AND d.ROWID = d1.ROWID);
So it returns nothing as all rows match their mirror rows.
However, it gave me the idea of doing the following:
SELECT D.ROWID FROM TAB1 D WHERE (SELECT COUNT(*) FROM TAB1 D1 WHERE X.colA = D.colA AND X.colB = D.colB )=1;
Which works, and seems a little faster than my original query, so thank you for this.