Multiple Joins

Question

Using HR demonstration schema, I am trying to understand how Oracle performs this multiple join:
select r.region_name, c.country_name, l.city, d.department_name
from departments d natural join locations l
natural join countries c natural join regions r
I ran the query using SQL Developer 2.1.1.64.45 and SQLPlus and the result I obtain is the same: 27 rows.
The documentation says that I should obtain a set of 2700 rows:
“The join between DEPARTMENTS and LOCATIONS creates an interim
result set consisting of 27 rows. These tables provide the DEPARTMENT_NAME
and CITY columns. This set is naturally joined to the COUNTRIES table. Since
the interim set does not contain the COUNTRY_ID column, a Cartesian join is
performed. The 27 interim rows are joined to the 25 rows in the COUNTRIES
table, yielding a new interim results set with 675 (27 × 25) rows and three columns: DEPARTMENT_NAME, CITY, and COUNTRY_NAME. This set is naturally joined to the REGIONS table. Once again, a Cartesian join occurs because the REGION_ID column is absent from the interim set. The final result set contains 2700 (675 × 4) rows and four columns.”
I parsed the query like it is suggested and the difference began with the second join:
select c.country_name, l.city, d.department_name
from departments d natural join locations l
natural join countries c;
I obtain a set of 27 rows instead of 675 as the documentation says (and I assume the documentation is correct).
If the result set from first Join it is fed as first term for the second join(from left to right) the result should be a Cartesian join and still my tools returns a set of 27 rows .
I verified the structure and contain of the 4 tables involved and it do match the one describe in documentation.
Why is my result different from the one provide in doc?
It is like it doesn’t matter if I select country_id column or not from first join,
somehow Oracle knows to identify the common column for the second join.

Answer 1

Hi,
Welcome to the forum!
user12290417 wrote:
Using HR demonstration schema, I am trying to understand how Oracle performs this multiple join:
select r.region_name, c.country_name, l.city, d.department_name
from departments d natural join locations l
natural join countries c natural join regions r
I ran the query using SQL Developer 2.1.1.64.45 and SQLPlus That's a very good idea to post the exact version number of your front end!
The version of your database can make a big difference, too. Don't just put an easy-to-miss tag, like "11g" on your message; come right out and say "I'm using Oracle 11.1.0.7.0" (or whatever it is).
and the result I obtain is the same: 27 rows.
The documentation says that I should obtain a set of 2700 rows:I get 2700 rows, using Oracle 10.2.0.1.0 Express Edition. I might be able to test it on Oracle 11 tomorrow.
“The join between DEPARTMENTS and LOCATIONS creates an interim
result set consisting of 27 rows. These tables provide the DEPARTMENT_NAME
and CITY columns. This set is naturally joined to the COUNTRIES table. Since
the interim set does not contain the COUNTRY_ID column, a Cartesian join is
performed. The 27 interim rows are joined to the 25 rows in the COUNTRIES
table, yielding a new interim results set with 675 (27 × 25) rows and three columns: DEPARTMENT_NAME, CITY, and COUNTRY_NAME. This set is naturally joined to the REGIONS table. Once again, a Cartesian join occurs because the REGION_ID column is absent from the interim set. The final result set contains 2700 (675 × 4) rows and four columns.”Can you post a link, or at least a reference, to the source of this quote? Is it describning Oracle 11 specifically?
I parsed the query like it is suggested and the difference began with the second join:
select c.country_name, l.city, d.department_name
from departments d natural join locations l
natural join countries c;
I obtain a set of 27 rows instead of 675 as the documentation says (and I assume the documentation is correct).Again, I get 675 rows, like your source.
If the result set from first Join it is fed as first term for the second join(from left to right) the result should be a Cartesian join and still my tools returns a set of 27 rows .
I verified the structure and contain of the 4 tables involved and it do match the one describe in documentation.
Why is my result different from the one provide in doc?
It is like it doesn’t matter if I select country_id column or not from first join,
somehow Oracle knows to identify the common column for the second join.Yes, the behavior you're seeing does seem to be like that.
When you explicitly give the join conditions in an ON clause, you can use any columns in any table that's been introduced up to that point. You're not limited to columns that are in the SELECT clause. The behavior you're seeing is consistent with that.
For what it's worth, the only people I've ever heard of using NATURAL JOIN are students, and even then, only if they are very diligent about following instructions. In real life, there are so many generic column names, such as QUANTITY, or MODIFY_DATE_TIME, that NATURAL JOIN is often meaningless. Also, it's a very fragile technique. If someone adds a column to a table, suddenly dozens of queries using NATURAL JOINs could start giving bad results. Most people want to write robust code, and minimize the probability of needing to re-write it.
There are lots of good places to invest your time, and lots of useful things to learn in Oracle. Spend your time learning about things that you might use, such as regular expressions, or analytic functions, or recursive sub-queries, before you spend too much time on NATURAL JOIN (or USING).