difference between "on" and "where" when using left/right join query

时间:2022-10-19 15:07:24

  I used to put the where-condiction in the "on" substatement in a join query,wishing that can help reducing the join count and improving the performence.But totally i was wrong.It seems the on-condiction is not like the where-condiction in a left/right join query.

  For example, there are two tables : 
  order([id],[order_code]), order_detail([id],[order_id],[product_name]), and they have some rows of data: 

order:
id order_code
1 order001
2 order002
3 order003
order_detail:
id order_id product_name
1 1 p001
2 1 p002
3 2 p003
4 3 p001
5 3 p003

  Now i want to know the order_codes of orders which buying the product "p001",what is the query statement probably like?

  In the pass,I may write this sql like this :

select * from [order_detail] left join [order] on [order].id=[order_detail].orderid and order_detail.product_name='p001'

rather than this:

select * from [order_detail] left join [order] on [order].id=[order_detail].orderid where order_detail.product_name='p001'

  How come i prefer the first one?,but not the second one? I thought the first one is faster because it would only join rows which product_name is 'p001' in table [order_detail]. I guess 
the sql server would check all of "on-condiction" and if the condiction is false, sql server would not execute the join operation,that means join operation will only occur twice(because there are only two rows which product_name is "p001" in table [order_detail]). But the second one sucks since it will join all rows of table [order_detail] and then find out which row's product_name is 'p001' only when all join operations get done!

  But the first one is a bad query,it's not giving what i want. In fact it returns result like this:

id order_id product_name id order_code
1 1 p001 1 order001
2 1 p002 null null
3 2 p003 null null
4 3 p001 3 order003
5 3 p003 null null

   But what result i want is like this:

id order_id product_name id order_code
1 1 p001 1 order001
4 3 p001 3 order003

  And only the second sql is correct.
  So what is wrong? What's the matter of the first one?
  In this case, the "on-condiction" is not like what i  think about.In a left/right join query, sql server will select all rows of the basic table no matter the on-condiction is true or false, in other words, the "on-condiction" is not a condiction to selecting rows of a basic table, in fact ,it's just a condiction to joinning rows.If the on-condiction is true, the current row of the basic table will join the row of the secondary table, if not,it won't,remainning the null in the field. But no matter it's true or not, all rows of the basic table are there,no more no less.

  But the second,surely, will get the right result: only two row with the product_name "p001".It will join all rows,and when the join get done,it then find out rows i want.