Suggestions for this sql.

Question

Hi,
Here is my code.
drop table t;
drop table p;
create table t (site_id number,
                state varchar2(2),
                county varchar2(100));
create table p(state varchar2(2),
               county varchar2(100),
               org1 number,
               org2 number,
               org3 number);
insert into t values (1, 'TX', 'DALLAS');
insert into t values (2, 'TX', 'DENTON');
insert into t values (3, 'CA','ORANGE');
insert into t values (4, 'PA','ERIE');
insert into t values (5, 'FL','SARASOTA');
insert into p values ('TX','DALLAS',121,122,123);
insert into p values ('TX',NULL, 124,125,126);
insert into p values ('CA','ORANGE',127,128,129);
insert into p values (null,null,130,131,132);
SELECT *
  FROM (SELECT t.site_id,
               t.state,
               t.county,
               p.org1,
               p.org2,
               p.org3,
               row_number() over(PARTITION BY t.state, t.county ORDER BY p.state, p.county NULLS LAST) rown
          FROM t,
               p
         WHERE t.state = nvl(p.state, t.state)
           AND t.county = nvl(p.county, t.county))
WHERE rown = 1
ORDER BY site_id't' is my driving table.
1) If I find an exact match on the join on 'state' and 'county', then, I should pick that row for the orgs in 'p'.
2) If no match in point 1, then, look for the same 'state' and null as the county in p.
3) If no match in point 2, then, look for the null state and null count in p
to pick the rows.
We came up with this sql.
1). Is this sql right? (first glance, it seems to give us the right intended data).
2). Could this have been written in a different way?
Thank you,
Rahul.

Answer 1

Frank Kulash wrote:
Hi, Rahul,
The query looks pretty good to me. I can't think of any better alternatives, or of any situations where that wouldn't get the results you want.
What is unique in those tables?
If t.site_id is unique, then I would suggest changing the PARTITION BY clause to "PARTITION BY t.site_id", especially if (t.state, t.county) is not unique.
Yes Frank, site_id is unique in my driving view. I will change it accordingly.
In table t, can state or county be NULL? If so, what results would you want? You might have to change the join conditions.
In table t, no, we DO NOT have state and county as nulls. (Good point, I will keep that in mind if not so).
Are you certain that p will always contain a row where both state and county are NULL? If not, you might need to make this an outer join.
Yeap, that is right.
Actually, there are no nulls in the 'state' and/or 'county' fields either. The data was actually a star '*'. I didn't know how to write this sql for * as the value, so, I used NVL in the join.
NULLS LAST is the default in ORDER BY. There's nothing wrong in explicitly saying NULLS LAST, That is what I thought, but, then, I think I saw in one case, my IDE (PL/SQL Developer in this case) showed NULLS FIRST if I just did an order by. I could have been wrong.
but if you're doing so for p.county, why not for p.state as well?Yes, you are right. I forgot to do NULLS LAST for state as well. that phrase is indeed per column.
So, for now, I will change this query to this:
SELECT *
  FROM (SELECT t.site_id,
               t.state,
               t.county,
               p.org1,
               p.org2,
               p.org3,
               row_number() over(PARTITION BY t.site_id, t.state, t.county ORDER BY p.state, p.county) rown
          FROM t,
               p
         WHERE t.state = nvl(p.state, t.state)
           AND t.county = nvl(p.county, t.county))
WHERE rown = 1
ORDER BY site_idEdited by: RPuttagunta on Jan 26, 2011 9:52 PM
Edited by: RPuttagunta on Jan 26, 2011 9:55 PM
EDIT: Love Chrome OS laptop, but, the trackpad feedback is a mess.