[sql] Inner join vs Where

Is there a difference in performance (in oracle) between

Select * from Table1 T1 
Inner Join Table2 T2 On T1.ID = T2.ID

And

Select * from Table1 T1, Table2 T2 
Where T1.ID = T2.ID

?

They're logically identical, but in the earlier versions of Oracle that adopted ANSI syntax there were often bugs with it in more complex cases, so you'll sometimes encounter resistance from Oracle developers when using it.

As kiewik said, the execution plan is the same.

The JOIN statement is only more easy to read, making it easier not to forget the ON condition and getting a cartesian product. These errors can be quite hard to detect in long queries using multiple joins of type : SELECT * FROM t1, t2 WHERE t1.id=t2.some_field.

If you forget only one join condition, you get a very long to execute query returning too many records... really too many. Some poeple use a DISTINCT to patch the query, but it's still very long to execute.

That's accurately why, using JOIN statement is surely the best practice : a better maintainability, and a better readability.

Further more, if I well remember, JOIN is optimized concerning memory usage.

Using JOIN makes the code easier to read, since it's self-explanatory.

There's no difference in speed(I have just tested it) and the execution plan is the same.

In a scenario where tables are in 3rd normal form, joins between tables shouldn't change. I.e. join CUSTOMERS and PAYMENTS should always remain the same.

However, we should distinguish joins from filters. Joins are about relationships and filters are about partitioning a whole.

Some authors, referring to the standard (i.e. Jim Melton; Alan R. Simon (1993). Understanding The New SQL: A Complete Guide. Morgan Kaufmann. pp. 11–12. ISBN 978-1-55860-245-8.), wrote about benefits to adopt JOIN syntax over comma-separated tables in FROM clause.

I totally agree with this point of view.

There are several ways to write SQL and achieve the same results but for many of those who do teamwork, source code legibility is an important aspect, and certainly separate how tables relate to each other from specific filters was a big leap in sense of clarifying source code.

I have an addition to that good answer:

That's what is defined as SQL92 and SQL89 respectively, there is no performance difference between them although you can omit the word INNER (using just JOIN is clear enough and in the simplest query you save 5 keyboard strokes now imagine how many strokes there are in big ones).

Although the identity of two queries seems obvious sometimes some strange things happens. I have come accros the query wich has different execution plans when moving join predicate from JOIN to WHERE in Oracle 10g (for WHERE plan is better), but I can't reproduce this issue in simplified tables and data. I think it depends on my data and statistics. Optimizer is quite complex module and sometimes it behaves magically.

Thats why we can't answer to this question in general because it depends on DB internals. But we should know that answer has to be 'no differences'.

They're both inner joins that do the same thing, one simply uses the newer ANSI syntax.

It is true that, functionally, both queries should be processed the same way. However, experience has shown that if you are selecting from views that use the new join syntax, it is important to structure your queries using it as well. Oracle's optimizer can get confused if a view uses a "join" statement, but a query accessing the view uses the traditional method of joining in the "where" clause.

The performance should be identical, but I would suggest using the join-version due to improved clarity when it comes to outer joins.

Also unintentional cartesian products can be avoided using the join-version.

A third effect is an easier to read SQL with a simpler WHERE-condition.

Functionally they are the same as has been said. I agree though that doing the join is better for describing exactly what you want to do. Plenty of times I've thought I knew how I wanted to query something until I started doing the joins and realized I wanted to do a different query than the original one in my head.

They're both joins and where that do the same thing.

Give a look at In MySQL queries, why use join instead of where?

In PostgreSQL, there's definitely no difference - they both equate to the same query plan. I'm 99% sure that's also the case for Oracle.

The performance should be identical, but I would suggest using the join-version due to improved clarity when it comes to outer joins.

Also unintentional cartesian products can be avoided using the join-version.

A third effect is an easier to read SQL with a simpler WHERE-condition.

If the query optimizer is doing its job right, there should be no difference between those queries. They are just two ways to specify the same desired result.

They should be exactly the same. However, as a coding practice, I would rather see the Join. It clearly articulates your intent,

I don't know about Oracle but I know that the old syntax is being deprecated in SQL Server and will disappear eventually. Before I used that old syntax in a new query I would check what Oracle plans to do with it.

I prefer the newer syntax rather than the mixing of the join criteria with other needed where conditions. In the newer syntax it is much clearer what creates the join and what other conditions are being applied. Not really a big problem in a short query like this, but it gets much more confusing when you have a more complex query. Since people learn on the basic queries, I would tend to prefer people learn to use the join syntax before they need it in a complex query.

And again I don't know Oracle specifically, but I know the SQL Server version of the old style left join is flawed even in SQL Server 2000 and gives inconsistent results (sometimes a left join sometimes a cross join), so it should never be used. Hopefully Oracle doesn't suffer the same issue, but certainly left and right joins can be mcuh harder to properly express in the old syntax.

Plus it has been my experience (and of course this is strictly a personal opinion, you may have differnt experience) that developers who use the ANSII standard joins tend to have a better understanding of what a join is and what it means in terms of getting data out of the database. I belive that is becasue most of the people with good database understanding tend to write more complex queries and those seem to me to be far easier to maintain using the ANSII Standard than the old style.

Don’t forget that in Oracle, provided the join key attributes are named the same in both tables, you can also write this as:

select *
from Table1 inner join Table2 using (ID);

This also has the same query plan, of course.

i had this conundrum today when inspecting one of our sp's timing out in production, changed an inner join on a table built from an xml feed to a 'where' clause instead....average exec time is now 80ms over 1000 executions, whereas before average exec was 2.2 seconds...major difference in the execution plan is the dissapearance of a key lookup... The message being you wont know until youve tested using both methods.

cheers.

[For a bonus point...]

Using the JOIN syntax allows you to more easily comment out the join as its all included on one line. This can be useful if you are debugging a complex query

As everyone else says, they are functionally the same, however the JOIN is more clear of a statement of intent. It therefore may help the query optimiser either in current oracle versions in certain cases (I have no idea if it does), it may help the query optimiser in future versions of Oracle (no-one has any idea), or it may help if you change database supplier.