Double outer join on table

Question

Hi
Let's say I have three tables ORDERS, PRODUCT and CLIENT. I want a query that will give me all PRODUCT and CLIENT combinations and the total sum of orders if it has one. If I try to do this:
select P.NAME, C.NAME, SUM(O.AMOUNT)
from PRODUCT P, CLIENT C, ORDER O
where P.ID = O.PRODUCT_ID(+) and
C.ID = O.CLIENT_ID(+)
I get an error where I cannot double outer join a table.
Any ideas?

Answer 1

3360 wrote:
BluShadow wrote:
and ANSI allows a little more in those terms than regular oracle syntax.It is different, but it cannot do anything that regular equi-joins cannot do. In most cases the optimizer converts ANSI joins to regular joins for execution, and this additional conversion has been subject to numerous bugs.ANSI does allow for doing things that regular Oracle syntax cannot do.
Here's an ANSI way to do it if you only want to see those client/product combinations that exist in orders table:
SQL> with product as (
  2     select 101 id, 'Product 1' name from dual union all
  3     select 102 id, 'Product 2' name from dual union all
  4     select 103 id, 'Product 3' name from dual
  5  ), client as (
  6     select 11 id, 'Client 1' name from dual union all
  7     select 12 id, 'Client 2' name from dual
  8  ), orders as (
  9     select 11 client_id, 101 product_id, 10 amount from dual union all
10     select 11 client_id, 103 product_id, 30 amount from dual union all
11     select 12 client_id, 102 product_id, 20 amount from dual union all
12     select 12 client_id, 102 product_id, 20 amount from dual
13  )
14  --
15  -- end-of-test-data
16  --
17  select p.name
18       , c.name
19       , sum(o.amount)
20    from orders o
21    left outer join product p
22         on p.id = o.product_id
23    left outer join client c
24         on c.id = o.client_id
25   group by
26         p.name
27       , c.name
28   order by
29         p.name
30       , c.name
31  /
NAME      NAME     SUM(O.AMOUNT)
Product 1 Client 1            10
Product 2 Client 2            40
Product 3 Client 1            30(If the datamodel is good with product_id being a foreign key of product.id and the same with client, then outer join is not really necessary, I know. But for the sake of argument assume bad datamodel that needs outer join ;-) )
The above ANSI can be written in regular Oracle syntax as well:
SQL> with product as (
  2     select 101 id, 'Product 1' name from dual union all
  3     select 102 id, 'Product 2' name from dual union all
  4     select 103 id, 'Product 3' name from dual
  5  ), client as (
  6     select 11 id, 'Client 1' name from dual union all
  7     select 12 id, 'Client 2' name from dual
  8  ), orders as (
  9     select 11 client_id, 101 product_id, 10 amount from dual union all
10     select 11 client_id, 103 product_id, 30 amount from dual union all
11     select 12 client_id, 102 product_id, 20 amount from dual union all
12     select 12 client_id, 102 product_id, 20 amount from dual
13  )
14  --
15  -- end-of-test-data
16  --
17  select p.name
18       , c.name
19       , sum(o.amount)
20    from orders o, product p, client c
21   where p.id(+) = o.product_id
22     and c.id(+) = o.client_id
23   group by
24         p.name
25       , c.name
26   order by
27         p.name
28       , c.name
29  /
NAME      NAME     SUM(O.AMOUNT)
Product 1 Client 1            10
Product 2 Client 2            40
Product 3 Client 1            30Many old Oracle coders will probably state that this syntax is much easier. For a simple case like this I find both ways easy. But I have complex cases where I much prefer ANSI where I can easily distinguish join predicates from filter predicates. It is a matter of opinion, I know ;-)
But if we then take the original post as a literal specification, that we want all combinations of products and clients and then the sum of order amount if it exists, null otherwise.
This can be written easily in ANSI notation:
SQL> with product as (
  2     select 101 id, 'Product 1' name from dual union all
  3     select 102 id, 'Product 2' name from dual union all
  4     select 103 id, 'Product 3' name from dual
  5  ), client as (
  6     select 11 id, 'Client 1' name from dual union all
  7     select 12 id, 'Client 2' name from dual
  8  ), orders as (
  9     select 11 client_id, 101 product_id, 10 amount from dual union all
10     select 11 client_id, 103 product_id, 30 amount from dual union all
11     select 12 client_id, 102 product_id, 20 amount from dual union all
12     select 12 client_id, 102 product_id, 20 amount from dual
13  )
14  --
15  -- end-of-test-data
16  --
17  select p.name
18       , c.name
19       , sum(o.amount)
20    from product p
21   cross join client c
22    left outer join orders o
23         on o.product_id = p.id
24        and o.client_id = c.id
25   group by
26         p.name
27       , c.name
28   order by
29         p.name
30       , c.name
31  /
NAME      NAME     SUM(O.AMOUNT)
Product 1 Client 1            10
Product 1 Client 2
Product 2 Client 1
Product 2 Client 2            40
Product 3 Client 1            30
Product 3 Client 2
6 rows selected.ANSI notation allows outer join with more than one table.
Regular Oracle syntax could only do this by creating an inline view of the cartesian join and then outer joining orders with the inline view. Doable, but not as nice (IMHO) as the ANSI method for a case like this ;-)