If positive, sum all items. If negative, return each one

I am needing to find a way to SUM() all of the positive values for num and return the SUM() of all positive numbers and an individual row for each negative number. Below is a sample DDL:

Create Table #Be
(
id int
, salesid int
, num decimal(16,4)
)

Insert Into #BE Values
(1, 1, 12.32), (2, 1, -13.00), (3, 1, 14.00)
, (4, 2, 12.12), (5, 2, 14.00), (6, 2, 21.23)
, (7, 3, -12.32), (8,3, -43.23), (9, 3, -2.32)

And this is my desired output (positive numbers for each salesid SUM() and negatives get an individual line returned):

salesid    num
1          26.32
1          -13.00
2          47.35
3          -12.32
3          -43.23
3          -2.32

Try this:

SELECT   salesid, sum(num) as num
FROM     #BE
WHERE    num > 0
GROUP BY salesid
UNION ALL
SELECT   salesid, num
FROM     #BE
WHERE    num < 0;

If you want both the sum values in one row then you must create a maxValue (and minValue) function and use this as sum(maxValue(0, num)) and sum(minValue(0, num)). This is described in: Is there a Max function in SQL Server that takes two values like Math.Max in .NET?

• I fixed the query. Needed UNION ALL as well, not UNION. – ypercubeᵀᴹ Jun 29 '16 at 12:27

This works too:

SELECT salesid, SUM(num)
FROM #BE
GROUP BY salesid, CASE WHEN num >= 0 THEN 0 ELSE id END;

Assumptions:

• Id starts at 1, hence it can use THEN 0. salesid ELSE salesid+id+1 would work as well
• 0 is considered positive number, hence the >= 0 (Is zero positive or negative?). Although x+0=x seems to make the = sign unnecessary, it helps remember that this case has not been forgotten and how 0 is handled (as a SUM or as an individual row). If the SUM() of all positive numbers means SUM of strictly positive numbers (i.e. >0), then = is not needed.

It must be tested with real data and indexes, but with only 1 table scan, performances may be a little better in some cases.

The absence of an index seems to have a smaller impact with this query on test data below:

SET NO COUNT ON
Create Table #Be(
id int identity(0,1)
,salesid int,num decimal(16,4)
)
INSERT INTO #BE(salesid, num)
SELECT CAST(rand()*10 as int), rand() - rand()
GO 10000 -- or 100.000
• You can simplify your group clause with an iif like this: GROUP BY salesid, iif(num >= 0, 0, id) Cool query. – user2023861 Jun 30 '16 at 15:45
• Yes but OP would have to first install SQL Server 2012. IIF starts with SQL Server 2012: msdn.microsoft.com/en-us/library/hh213574.aspx. OP has tagged his question with SQL Server 2008. – Julien Vavasseur Jun 30 '16 at 15:48