Outer Join Negation Question

Question

I'm retrofitting an existing plsql report into BusinessObjects. I have a question about
the sql I'm looking at. There is an employee table and a case table. Employees have cases. There are two other tables...a case activity table and a case results table. There may or may not be activity and results for a given case. The query is this...
select count(distinct case_id)
employee.employee_id = case.employee_id
and case.case_id = case_activity.case_id (+)
and case.case_id = case_results.case_id (+)
so far so good. The next part is
and (case_activity.activity_date <= mm/dd/yyyy
or case_results.activity_date <= mm/dd/yyyy)
That cancels the outer join, right? I contend that there is no
reason for the outer join symbols. He says they're needed. He thinks the OR is making a difference. It looks to me like you've lost the guarantee that all cases will be included. Can someone please splain this to me :-)
Thanks so much!
Cindy

Answer 1

Cindy,
I hope there's not too much of an arguement brewing on that query, but both of you are correct (to a degree). The OR in that second clause makes all the difference. I have some sql to show some test results based on your query. Basically remeber this, you always get one row back when making a query, if ALL conditions are met you get that record back. So even if the first part of the OR failes and the second one is true then u get the row back.
Here's where that other fellow is having a problem though. The problem your going to have is when neither of thoes dates return ture then you get NO data back regardless of the outter joins. Thoes two table Activity and Results will become related if thought about like this. What I mean, is if there is a date in the Activity table that matches, then you will get the Result data as well, and vice-versa. However if there is neither Activity or Result date matches you loose the whole case record even though you have the outter join specified. UNLESS thats the whole point of that DATE range. If neither match then I dont want any case records back. I'm just assuming since your outter joining you want the case back the case regardless if there are matches or not in Activity and Results, but maybe this isn't so.
Anyway, here some SQL for you and the data that produced it ...
Tables (t1,t2,t3,t4) : t1is employee, t2 is case t3 is activity and t4 is result.
Columns (c1,n1,d1) : varchar ID , number ID and a date
All dates are set for today (Arpil 5, 2002)
Data is the same in ALL the tables, only interested in C1= A and N1 = 1 from main table T1
Table T1 has 1 record where C1= 'A' and N1= 1
SQL> select c1,n1,d1 from t1 where c1='A' and n1=1;
C1 N1 D1
A 1 05-APR-02
Case #1 (BOTH t3 and t4 have no matches)
SQL> select DISTINCT
2 t1.c1 "t1_c1",t1.n1 "t1_n1",
3 t2.c1 "t2_c1",t2.d1 "t2_d1",t2.n1 "t2_n1",
4 t3.c1 "t3_c1",t3.d1 "t3_d1",t3.n1 "t3_n1",
5 t4.c1 "t4_c1",t4.d1 "t4_d1",t4.n1 "t4_n1"
6 from t1,t2,t3,t4
7 where
8 (
9 t1.C1 = t2.C1 and
10 t2.c1 = t3.c1 (+) and
11 t2.c1 = t4.c1 (+) and
12 (T3.D1 <= to_date('2002/04/04','yyyy/mm/dd') or T4.D1 <= to_date('2002/04/04','yyyy/mm/dd'))
13 )
14 ;
no rows selected
This SQL produces no data as BOTH dates in t3 (Activity) and t4 (Results) have no matches since all D1 are set to today. Even though there is an outter join the whole [ ( date match t3 OR date match t4 ) ] fails so no records (or cases) get produced).
The below Case #2 I changed the columns selected to a COUNT since this query would produce too much data to make of any sence in there. Besides, and I hope you oracle folks read this, the FONT makes any kind of query output impossible read. Ahhhhh I'm making something better at the moment for the public anyway.
SQL> select DISTINCT
2 count(t1.c1),count(t2.c1),count(t3.c1),count(t4.c1)
3 from t1,t2,t3,t4
4 where
5 (
6 t1.C1 = t2.C1 and
7 t2.c1 = t3.c1 (+) and
8 t2.c1 = t4.c1 (+) and
9 (T3.D1 <= to_date('2002/04/04','yyyy/mm/dd') or T4.D1 <= to_date('2002/04/04','yyyy/mm/dd'))
10 )
11 AND t1.c1 ='A' AND T1.N1 = 1;
COUNT(T1.C1) COUNT(T2.C1) COUNT(T3.C1) COUNT(T4.C1)
0 0 0 0
So as you can see with BOTH dates not having any matches you get no data back at all even with the outer joins. So yes in this case it does not matter if your outer joinning or not, the whole date clause falied and since it id ANDed to the whole where clause , the whole where clause fails and you get no data.
Case #2 (table t3 has a date match but non in t4) : Only change to SQL was change the DAY from 04 to 05 ;being today). Bascially you get ALL the data from t1,t2,t3 AND t4. The record entries in T4 still show up even though there are no matches and the data from t4 shows up, it is not null.
SQL> select DISTINCT
2 count(t1.c1),count(t2.c1),count(t3.c1),count(t4.c1)
3 from t1,t2,t3,t4
4 where
5 (
6 t1.C1 = t2.C1 and
7 t2.c1 = t3.c1 (+) and
8 t2.c1 = t4.c1 (+) and
9 (T3.D1 <= to_date('2002/04/05','yyyy/mm/dd') or T4.D1 <= to_date('2002/04/04','yyyy/mm/dd'))
10 )
11 AND t1.c1 ='A' AND T1.N1 = 1;
COUNT(T1.C1) COUNT(T2.C1) COUNT(T3.C1) COUNT(T4.C1)
8 8 8 8
And no I didn't just chage the count numbers to 8. Notice though you still get t4 matches since the ID join is true, even though the dates are not. If you only wanted T3 records out then you would have to split out the t3 date and the t4 date from that OR clause. The same results occur if T4 has date mathces and T3 does not. Same result if BOTH have matches in t3 and t4.
Case #3 : both t3 and t4 have no matches on date (both are set to yesterday, same as case #1) but you will want the data returned. Since the ID exists in both t3 and t4, but the dates do not, by adding another OR clause your OUTER joins will work ...
SQL> select DISTINCT
2 count(t1.c1),count(t2.c1),count(t3.c1),count(t4.c1)
3 from t1,t2,t3,t4
4 where
5 (
6 t1.C1 = t2.C1 and
7 t2.c1 = t3.c1 (+) and
8 t2.c1 = t4.c1 (+) and
9 (T3.D1 <= to_date('2002/04/04','yyyy/mm/dd') or T4.D1 <= to_date('2002/04/04','yyyy/mm/dd'))
10 )
11 OR
12 (
13 t1.C1 = t2.C1 and
14 t2.c1 = t3.c1 (+) and
15 t2.c1 = t4.c1 (+)
16 )
17 AND t1.c1 ='A' AND T1.N1 = 1;
COUNT(T1.C1) COUNT(T2.C1) COUNT(T3.C1) COUNT(T4.C1)
8 8 8 8
You can try this your self if you like, but overall if any part of and AND clause failes, no matter if part of it had 100 OR clauses in one set of brackets, it all fails, even if you have outer joins.
Have a kick butt weekend and good luck with your sql, email if you need anything :)
Tyler
[email protected]