The issue with using the multiple columns sub-query in WHERE clause

Question

Hi All,
my database version is 10.2.
the problem i am trying to deal with is that when I use multiple column sub-query in the WHERE clause of the SELECT statement, the actual row number returned from the sub-query is different from the whole statement.
And what I found is that, whenever there is NULL in any of those columns returned from the SUB-QUERY, the outer query will just return NULL for that whole row.
Here is an example:
select empno, ename, job, mgr, hiredate, sal, deptno from EMP
intersect
select empno, ename, job,  mgr, hiredate, sal, deptno from t;
7782     CLARK     MANAGER     7839     09-JUN-81     2450     10
7839     KING     PRESIDENT  NULL  17-NOV-81     5000     10
7934     MILLER     CLERK     7782     23-JAN-82     1300     10
select * from EMP where (EMPNO, ENAME, job, MGR, HIREDATE, SAL, DEPTNO) in (
select empno, ename, job, mgr, hiredate, sal, deptno from EMP
intersect
select empno, ename, job,  mgr, hiredate, sal, deptno from t);
7782     CLARK     MANAGER     7839     09-JUN-81     2450          10     
7934     MILLER     CLERK     7782     23-JAN-82     1300          10     If I specifically deal with the NULL situations for the columns which might return NULL, I can get the result right.
select * from EMP where (EMPNO, ENAME, job, NVL(MGR,-1), HIREDATE, SAL, DEPTNO) in (
select empno, ename, job, nvl(mgr,-1), hiredate, sal, deptno from EMP
intersect
select empno, ename, job,  nvl(mgr,-1), hiredate, sal, deptno from t);
7782     CLARK     MANAGER     7839     09-JUN-81     2450          10     
7839     KING     PRESIDENT  null   17-NOV-81     5000          10     
7934     MILLER     CLERK     7782     23-JAN-82     1300          10     the problem is that, I feel this is a very lame way of handling it.
So, I wonder or am asking if there is any better or standard way to do it?
any help would be highly appreciated.
Thanks

Answer 1

Hi,
As you discovered, INTERSECT treats NULL as a value, but IN does not.
What you did with NVL is one way to handle the situation. If there was a chance that any of the columns could be NULL, then you might prefer something like this:
select      *
from      EMP
where      ( EMPNO      || '~' ||
       ENAME      || '~' ||
       job           || '~' ||
       MGR           || '~' ||
       TO_CHAR (HIREDATE, 'DD-MON-YYYY HH24:MI:SS')
                || '~' ||
       SAL           || '~' ||
       DEPTNO
     ) in (
          select  EMPNO      || '~' ||
                 ENAME      || '~' ||
               job     || '~' ||
               MGR     || '~' ||
               TO_CHAR (HIREDATE, 'DD-MON-YYYY HH24:MI:SS')               
                    || '~' ||
               SAL      || '~' ||
               DEPTNO
          from     emp
         intersect
          select  EMPNO      || '~' ||
                 ENAME      || '~' ||
               job     || '~' ||
               MGR     || '~' ||
               TO_CHAR (HIREDATE, 'DD-MON-YYYY HH24:MI:SS')               
                    || '~' ||
               SAL      || '~' ||
               DEPTNO
          from      t
         );This assumes that you can identify some string (I used '~') that never occurs in the strings in these tables.
This is implicitly converting the NUMBERs. That's usually not a good thing to do. but explicitly converting them would make this even more tedious.
You should explicitly convert any DATEs to strings, however. Depending on your default format, and your data, you might get away with implicit conversions even for DATEs, but don't bet on it.
If you had to do this often, you might consider writing a user-defined function:
delimited_string (empno, ename, job, mgr, hiredate, sal, deptno) would return a string like
'7839~KING~PRESIDENT~~17-NOV-1981~5000~10'
This will make the coding easier, but beware: it will make the execution slower.