Evaluate WHERE predicates on analytic functions before other predicates (Oracle analytic functions)

Question

Background

Sample data set

#Employee
Id | Period | Status 
---------------------
1  |  1 |   L    
1  |  2 |   G    
2  |  3 |   L    

I want a simple select query to yield employees' latest record (by period) only if the status='L'.

The results would look like this:

#Desired Results
Id | Period | Status | Sequence
-------------------------------
2  |  3     |   L    |   1

Naive attempt

Obviously, my naive attempt at a query does not work:

#select query
SELECT *, RANK() OVER (PARTITION BY id ORDER BY period ASC) sequence
FROM employees
WHERE   status = 'L' 
 AND    sequence = 1

Which results in the following:

#Naive (incorrect) Results
ID | Period | Status | Sequence
-------------------------------
1  |  1 |   L    |   1
2  |  3 |   L    |   1

Knowing the order that clauses are evaluated in SQL explains why it doesn't work. Here is how my query is evaluated:

• Isolate rows where status='L'
• Rank the rows
• Isolate top rank row

I want the following:

• Rank rows
• Isolate the top ranked rows
• Isolate where status='L'

Questions

• Is possible--with only a simple modification to the SELECT/WHERE clauses and using only basic predicate operators--to ensure that predicates based on analytic functions in the WHERE clause get evaluated before the non-aggregate predicates?

• Anyone have other solutions that can be implemented as an end-user in Oracle Discoverer Plus?

Thanks!

Answer 1

Is it possible to do this without a sub-query

Technically the following is not a sub-query but a derived table

SELECT * 
FROM (
    SELECT *, 
           RANK() OVER (PARTITION BY id ORDER BY period ASC) sequence
    FROM employees
) t
WHERE status = 'L' 
  AND sequence = 1

I can't think of a different solution to your problem.

Answer 2

The classic Group by

SELECT e.id, e.period, e.status, 1 sequence
FROM
(
    SELECT id, min(period) period
    FROM employees
    GROUP BY id
) X
JOIN employees e on e.period=X.period and e.id=X.id
WHERE e.status = 'L'

Exists

select e.id, e.period, e.status, 1 sequence
FROM employees e
WHERE e.status = 'L'
  AND NOT EXISTS (select *
                  from employees e2
                  where e2.id=e.id and e2.period>e.period)

Answer 3

I'll probably have to do a "Dobby" and slam my ear in the oven door and iron my hands for this...

You can create a function which evaluates the current row.
Note that this is inherently non-scalable. But I guess it's better than nothing.

Create the sample data:

--drop table employee purge;

create table employee(
    id     number  not null
   ,period number  not null
   ,status char(1) not null
   ,constraint employee_pk primary key(id, period)
);

insert into employee(id,period, status) values(1, 1, 'L');
insert into employee(id,period, status) values(1, 2, 'G');
insert into employee(id,period, status) values(2, 3, 'L');

commit;

Create the slowest function in the database:

create or replace function i_am_slow(
    ip_id     employee.id%type
   ,ip_period employee.period%type
)
return varchar2
as
   l_count number := 0;
begin
    select count(*)
      into l_count
      from employee e
     where e.id     = ip_id
       and e.period = ip_period
       and e.status = 'L'
       and not exists(
            select 'x'
              from employee e2
             where e2.id = e.id
               and e2.period > e.period);

    if l_count = 1 then
        return 'Y';
    end if;

    return 'N';
end;
/

Demonstrates the use of the function:

select id, period, status
  from employee
 where i_am_slow(id, period) = 'Y';

        ID     PERIOD STATUS
---------- ---------- ------
         2          3 L

Rushes towards the oven...

Answer 4

select * from (SELECT a.*, rank() OVER (ORDER BY period ASC) sequence from (select * from ( select 1 id, 1 period, 'L' status from dual union all select 1 id, 2 period, 'G' status from dual union all select 2 id, 3 period, 'L' status from dual ) where status = 'L' ) a ) where sequence = 1