Outer join problem (ORA-01799)

Question

We have a database design roughly as follows:
- A STAFF table (columns don't matter here).
- Resources have a cost per hour that varies over time, so we have a STAFF_COST table with an effective date and a cost per hour
- A PROJECT table (columns don't matter here).
- Projects can have staff assigned, so we have a PROJECT_STAFF table which has foreign keys to the PROJECT and STAFF table.
- Project staff have a cost per hour, which can vary over time, and be different to the (default) staff costs. So we have a PROJECT_STAFF_COST table which uses the PROJECT_STAFF foreign key, with an effective date and a cost per hour
- Staff work on tasks so we have a TIMESHEET_TASK and TIMESHEET_DAY tables which define a project worked on, task within the project (optional as time can be 'entered against the project', who is recording the time, the hours worked and the day the hours were worked.
So when timesheet information is entered we have three scenario's that we need to cover.
1) The resource is a member of the project and the hours were worked on a day for which we have a project staff cost. That is, the project staff cost table has one or more rows for the staff member assigned to the given project with an effective date before the date the hours were entered against.
2) The resource is a member of the project but the hours were worked on a day for which we do not have a project staff cost. That is, the project staff cost table has one or more entries for the staff member assigned to the given project, but all the effective dates are after the date the hours were entered against.
3) The resource is not a member of the project. That is, the project staff cost table does not have any rows for the staff member. Time was entered 'against the project'.
We need to work out the actual cost of the project. So we need to retrieve every day's timesheet entry, returning the hours assigned and the cost per hour relevant for that day. I have the following query:
select tsh.staff_id, s.full_name, tsd.entry_date, tsd.hours as ProjectHours,
psCOST_INFO.cost_per_hour as ProjectCost
from timesheet_day tsd
inner join timesheet_task tst on tst.timesheet_task_id = tsd.timesheet_task_id
inner join timesheet_header tsh on tst.timesheet_header_id = tsh.timesheet_header_id
inner join staff s on s.staff_id = tsh.staff_id
left join (Select ps.project_id, ps.staff_id, psc.project_staff_id, psc.effective_date, psc.cost_per_hour
from project_staff ps
inner join project_staff_cost psc on ps.project_staff_id = psc.project_staff_id) as psCOST_INFO
on psCOST_INFO.staff_id = tsh.staff_id and psCOST_INFO.project_id = tst.project_id
and psCOST_INFO.effective_date = (select max(ps2.effective_date) from project_staff_cost ps2
where ps2.project_staff_id = psCOST_INFO.project_staff_id
and ps2.effective_date <= tsd.entry_date)
where tst.project_id = 55825
Using the left join covers scenario's 2 and 3 above because I will get null in the cost columns and can then take appropriate action. If I were to use an inner join, then hours in timesheets from scenario's 2 and 3 would be excluded, which is not what we want.
The subselect using the MAX aggregate function is required to get the cost per hour most relevant for the timesheet day. That is, if there are several effective dates for the project staff member before the date in question, we want the most recent one. We can't just use the MAX one, however in case there is an effective date after the particular timesheet date. Sheesh...
This query works fine in SQL Server. It it not allowed in Oracle and returns an ORA-01799 column may not be outer joined to a subquery.
I'm not going to bother to ask why not. I just need a way to do what I want. I've spent days trying to move the code around but I either end up with an inner join (which returns fewer rows than I want) or it just plain don't work.
Can someone help me rework this query to achieve the result I want?
Thanks, Andrew

Answer 1

Thanks for your reply, Laurent. In my experience trying to cut important corners in explaining a problem only serves to make it more difficult to solve. That pretty much was the smallest reproducable query that demonstrates the complexity of the problem I have. I'm not just trying to get which publishers live in the 'CA' state here...
From what I have just read about rank() it serves the same purpose as max() on a given column, and getting the maximum or top ranked one just doesn't cut it. As I said in my original post that provided all the relevant (and no spurious) information on the problem, it is possible that there are effective dates AFTER the date we are interested in and they have to be excluded.
I have to get the project staff cost row with the latest date that is before the timesheet date. That means I have to reference data in the outer query. Oracle seems to have a problem with that when used in an outer join.
We are currently going down the track of 3 UNION'd statement to cover the 3 scenario's. A single query would be more efficient so if anyone can provide guidance I would appreciate it.
Thanks, Andrew