Problem with MINUS and sub queries with ORDER BYs - oracle

Question

select salary
from (
(select salary
from employees
where rownum<=10
order by salary desc)
minus
(select salary
from employees
where rownum<=4
order by salary desc)
);

Answer 1

You cannot use ORDER BY there.
Try this instead:
select salary from (
select salary, row_number() over ( order by salary desc ) rn
from employees )
where rn between 5 and 10;
On Oracle 12c or later, you can also do this:
select salary from employees
order by salary desc
offset 4 rows fetch next 6 rows only;

Answer 2

You've got several issues in what you've written. The immediate problem is that you'll get an error from having an order by in the first branch of your union, but just removing that won't help you much.
You're making a (fairly common) mistake with ordering and rownum; looking just at the first subquery you have:
select salary
from employees
where rownum<=10
order by salary desc
The rownum filter will be applied before the order-by, so what this will actually produce is 10 indeterminate rows from the table, which are then ordered. If I run that I get:
SALARY
----------
24000
13000
12000
10000
8300
6500
6000
4400
2600
2600
but you'll see different values, even from the same sample schema. If you look at the whole table you'll see higher values than those; and even running the second query will show something isn't as you expect - for me that gets:
SALARY
----------
13000
4400
2600
2600
which are not the first four rows from the previous query. (Again, you'll see different results, but hopefully the same effect; if not, look at the whole table ordered by salary.)
You need to order the whole table - in a subquery - and then filter:
select salary
from (
select salary
from employees
order by salary desc
)
where rownum<=10
which gives a much more sensible - and consistent - result. You can then minus the two queries:
select salary
from (
select salary
from employees
order by salary desc
)
where rownum<=10
minus
select salary
from (
select salary
from employees
order by salary desc
)
where rownum<=4
order by salary desc;
SALARY
----------
13500
13000
12000
11500
You may be expecting to see six values there, but there are three employees with a salary of 12000, and minus eliminates duplicates so that is only reported once. #Matthew's approach (or #Jeff's!) will give you all six, including duplicates, if that is what you want. It also stops you having to hit the table multiple times.
A further problem is with ties - if the 4th highest was the same as the 5th highest, what would you expect to happen? Using minus would exclude that value; #Matthew's approach would preserve it.
You need to define what you actually want to get - the 5th to 10th highest salary values? The salaries of the 5th to 10th highest-paid people (a subtle but important difference)? Do you really only want the numbers, or who those employees are - in which case how you deal with ties is even more important? Etc. Once you know what you actually need to find you can decide the best way to get that result.

Answer 3

It doesn't make sense to order rows in two sets that are subsequently operated upon because sets don't have order. If you need a solution that can execute on older versions and you want to return the bottom 6 ranked out of the top 10 ranked, then this will work. If you can use newer features, then you may want to because it's possible they'll require fewer machine instruction executions.
After making the obvious changes that escaped me in my haste...
select salary
from (
select rownum rn, salary
from (
select salary
from employees
order by salary desc
)
)
where rn between 5 and 10