Remove all records with opposite sign - sql

Question

I'm looking for a SQL query (or even better a LINQ query) to remove people who have cancelled their leave, i.e. remove all records with the same NAME and same START and END and the DAYS_TAKEN values differ only in the sign.
How to get from this
NAME |DAYS_TAKEN |START |END |UNIQUE_LEAVE_ID
--------|-----------|-----------|-----------|-----------
Alice | 2 | 1 June | 3 June | 1 --remove because cancelled
Alice | -2 | 1 June | 3 June | 2 --cancelled
Alice | 3 | 5 June | 8 June | 3 --keep
Bob | 10 | 4 June | 14 June | 4 --keep
Charles | 12 | 2 June | 14 June | 5 --remove because cancelled
Charles | -12 | 2 June | 14 June | 6 --cancelled
David | 5 | 3 June | 8 June | 7 --keep
To this?
NAME |DAYS_TAKEN |START |END |UNIQUE_LEAVE_ID
--------|-----------|-----------|-----------|-----------
Alice | 3 | 5 June | 8 June | 3 --keep
Bob | 10 | 4 June | 14 June | 4 --keep
David | 5 | 3 June | 8 June | 7 --keep
What I've tried
Query1 to find all the cancelled records (not sure if this is correct)
SELECT L1.UNIQUE_LEAVE_ID
FROM LEAVE L1
INNER JOIN LEAVE L2 ON L2.DAYS_TAKEN > 0 AND ABS(L1.DAYS_TAKEN) = L2.DAYS_TAKEN AND L1.NAME= L2.NAME AND L1.START = L2.START AND L1.END = L2.END
WHERE L1.DAYS_TAKEN < 0
Then I use Query1 twice in an inner select like so
SELECT L.* FROM LEAVE L WHERE
L.UNIQUE_LEAVE_ID NOT IN (Query1)
AND L.UNIQUE_LEAVE_ID NOT IN (Query1)
Is there a way to use the inner query only once?
(It's an Oracle database, being called from .NET/C#)

Answer 1

You can use a query like the following:
SELECT NAME, START, END
FROM LEAVE
GROUP BY NAME, START, END
HAVING SUM(DAYS_TAKEN) = 0
in order to get NAME, START, END groups that have been cancelled (assuming DAYS_TAKEN of the cancellation record negates the days of the initial record).
Output:
NAME |START |END
--------|-----------|----------
Alice | 1 June | 3 June
Charles | 2 June | 14 June
Using the above query as a derived table you can get records not being related to 'cancelled' groups:
SELECT L1.NAME, L1.DAYS_TAKEN, L1.START, L1.END, L1.UNIQUE_LEAVE_ID
FROM LEAVE L1
LEFT JOIN (
SELECT NAME, START, END
FROM LEAVE
GROUP BY NAME, START, END
HAVING SUM(DAYS_TAKEN) = 0
) L2 ON L1.NAME = L2.NAME AND L1.START = L2.START AND L1.END = L2.END
WHERE L2.NAME IS NULL
Output:
NAME |DAYS_TAKEN |START |END |UNIQUE_LEAVE_ID
--------|-----------|-----------|-----------|-----------
Alice | 3 | 5 June | 8 June | 3
Bob | 10 | 4 June | 14 June | 4
David | 5 | 3 June | 8 June | 7

Answer 2

You can use not exists:
select l.*
from leave l
where not exists (select 1
from leave l2
where l2.name = l.name and l2.start = l.start and
l2.end = l.name and l2.days_taken = - l.days_taken
);
This query can take advantage of an index on leave(name, start, end, days_taken).

Answer 3

Here is a variation with SUM() OVER:
SELECT x.*
FROM (SELECT l.*, SUM (days_taken) OVER (PARTITION BY name, "START", "END", ABS (days_taken) ORDER BY NULL) s
FROM leave l) x
WHERE s <> 0
And if you have Oracle 12, this give you the canceled:
SELECT l.*
FROM leave l,
LATERAL (SELECT days_taken
FROM leave l2
WHERE l2.name = l.name
AND l2."START" = l."START"
AND l2."END" = l."END"
AND l2.days_taken = -l.days_taken) x
and this what should remain:
SELECT l.*
FROM leave l
OUTER APPLY (SELECT days_taken
FROM leave l2
WHERE l2.name = l.name
AND l2."START" = l."START"
AND l2."END" = l."END"
AND l2.days_taken = -l.days_taken) x
WHERE x.days_taken IS NULL
And something about the column names.Using reserved word in Oracle SQL is not recommended, but if you must do it, use '"' like here.

Answer 4

I used Giorgos answer to come up with this Linq solution. This solution also considers people who cancel / apply their leave multiple times. See Alice and Edgar below.
Sample data
int id = 0;
List<Leave> allLeave = new List<Leave>()
{
new Leave() { UniqueLeaveID=id++, Name="Alice", Start=new DateTime(2016,6,1), End=new DateTime(2016,6,3), Taken=-2 },
new Leave() { UniqueLeaveID=id++,Name="Alice", Start=new DateTime(2016,6,1), End=new DateTime(2016,6,3), Taken=2 },
new Leave() { UniqueLeaveID=id++, Name="Alice", Start=new DateTime(2016,6,1), End=new DateTime(2016,6,3), Taken=2 },
new Leave() { UniqueLeaveID=id++,Name="Alice", Start=new DateTime(2016,6,3), End=new DateTime(2016,6,5), Taken=3 },
new Leave() { UniqueLeaveID=id++,Name="Bob", Start=new DateTime(2016,6,4), End=new DateTime(2016,6,14), Taken=10 },
new Leave() { UniqueLeaveID=id++,Name="Charles", Start=new DateTime(2016,6,2), End=new DateTime(2016,6,14), Taken=12 },
new Leave() { UniqueLeaveID=id++,Name="Charles", Start=new DateTime(2016,6,2), End=new DateTime(2016,6,14), Taken=-12 },
new Leave() { UniqueLeaveID=id++,Name="David", Start=new DateTime(2016,6,3), End=new DateTime(2016,6,8), Taken=5 },
new Leave() { UniqueLeaveID=id++,Name="Edgar", Start=new DateTime(2016,6,3), End=new DateTime(2016,6,8), Taken=5 },
new Leave() { UniqueLeaveID=id++,Name="Edgar", Start=new DateTime(2016,6,3), End=new DateTime(2016,6,8), Taken=5 },
new Leave() { UniqueLeaveID=id++,Name="Edgar", Start=new DateTime(2016,6,3), End=new DateTime(2016,6,8), Taken=5 },
new Leave() { UniqueLeaveID=id++,Name="Edgar", Start=new DateTime(2016,6,3), End=new DateTime(2016,6,8), Taken=5 }
};
Linq Query (watch out for Oracle version 11 vs 12)
var filteredLeave = allLeave
.GroupBy(a => new { a.Name, a.Start, a.End })
.Select(a => new { Group = a.OrderByDescending(b=>b.Taken), Count = a.Count() })
.Where(a => a.Count % 2 != 0)
.Select(a => a.Group.First());
"OrderByDescending" ensures only positive days taken are returned.
Oracle SQL
SELECT
*
FROM
(
SELECT
L1.NAME, L1.START, L1.END, MAX(TAKEN) AS TAKEN, COUNT(*) AS CNT
FROM LEAVE L1
GROUP BY L1.NAME, L1.START, L1.END
) L2
WHERE MOD(L2.CNT,2)<>0 -- replace MOD with % for Microsoft SQL
The condition "WHERE MOD(L2.CNT,2)<>0" (or in Linq "a.Count % 2 != 0") only returns people who applied once or odd number of times (e.g. apply - cancel - apply). But people who apply - cancel - apply - cancel are filtered out.