SQL Server: Using conditional group by based on parameter and STUFF() - sql-server

Question

I have a piece of code that I would like to turn into a function. The purpose of the code is to group certain records together on variable criteria and create a grouped concatenation using STUFF(). I would like to be able to toggle the parameters on which the group by happens (and therefore also the parameters for the STUFF).
However, the following gives me errors where the optional parameters (e.g. OwnerName in the example below) are invalid in the select list because they are not contained in either an aggregation function or the GROUP BY clause.
Consider a simplified example like the below (the real life version has a lot of parameters, hence why I would like to be able to get these all into one query):
SELECT CarMake, CarModel, CASE WHEN #FlagOwnerName = 1 THEN OwnerName ELSE NULL END AS [OwnerName], SUM(CarValue),
LicenseIDs = STUFF((SELECT ',' + CONVERT(VARCHAR(20),Cars2.LicenseID) AS [text()]
FROM DB.dbo.Cars Cars2
WHERE Cars2.CarMake = Cars1.CarMake
AND Cars2.CarModel = Cars1.CarModel
AND (#FlagOwnerName = 0 OR Cars2.OwnerName = Cars1.OwnerName)
FOR XML PATH('')), 1, 1, '')
FROM DB.dbo.Cars Cars1
GROUP BY CarMake,
CarModel,
CASE WHEN #FlagOwnerName = 1 THEN OwnerName ELSE NULL END
EDIT: if I change the below, then it 'seems' to return the correct concatenation except if it is NULL, then the concatenation is NULL itself. Additionally, if I try to change the values to ISNULL(Cars1.OwnerName, 'Placeholder') or similarly with COALESCE, it gives me the same error (not valid in select statement as above).
AND (#FlagOwnerName = 0 OR Cars2.OwnerName = Cars1.OwnerName)
to
AND CASE WHEN #FlagOwnerName = 1 THEN Cars1.OwnerName = Cars2.OwnerName

Answer 1

Based on your comments, I don't think using STUFF with FOR XML if the best way to approach this. Typically the best way to concatenate multiple rows into a single string is to use a recursive Common Table Expression (CTE).
There are some examples of using CTEs (and some alternative approaches) here.
I've adapted one of the CTE options there to do something similar to what you've described.
First, I've set up a simple table that resembles the data you've described:
create table #cars (CarMake varchar(50), CarModel varchar(50), CarValue INT, OwnerName varchar(50), LicenseID varchar(50));
insert into #cars(CarMake, CarModel, CarValue, OwnerName, LicenseID) values ('Toyota','Camry', 12000, 'Steve','ABC123');
insert into #cars(CarMake, CarModel, CarValue, OwnerName, LicenseID) values ('Toyota','Camry', 12000, 'Bob','HED999');
insert into #cars(CarMake, CarModel, CarValue, OwnerName, LicenseID) values ('Toyota','Camry', 19000, 'Helen','WKS444');
insert into #cars(CarMake, CarModel, CarValue, OwnerName, LicenseID) values ('Ford','Mustang',30000, 'Amy','JKJL88');
insert into #cars(CarMake, CarModel, CarValue, OwnerName, LicenseID) values ('Ford','Mustang',30000, 'Billy-Bob','EZ1111');
insert into #cars(CarMake, CarModel, CarValue, OwnerName, LicenseID) values ('Aston Martin','Vantage',90000, 'Mike','HY7733');
I've then used a CTE to build a dataset where the car licenses and values are appended/aggregated by make/model. The variable #FlagOwnerName controls whether these values from the CTE or the base values from the source table are used in the final SELECT statement:
DECLARE #FlagOwnerName bit = 1;
WITH cte (CarMake, CarModel, CarValueTotal, Car_Val, LicenseList, License_ID, length_)
AS
(
SELECT
CarMake, CarModel, 0, 0, CAST( '' AS VARCHAR(8000) ), CAST( '' AS VARCHAR(8000) ), 0
FROM #cars
GROUP BY CarMake, CarModel
UNION ALL
SELECT c.CarMake, c.CarModel, cte.CarValueTotal + c.CarValue, c.CarValue,
CAST(cte.LicenseList + CASE WHEN length_ = 0 THEN '' ELSE ', ' END + c.LicenseID AS VARCHAR(8000) ),
CAST( LicenseID AS VARCHAR(8000)),
length_ + 1
FROM cte
INNER JOIN #cars c ON cte.CarMake = c.CarMake AND cte.CarModel = c.CarModel
WHERE c.LicenseID > cte.License_ID
)
SELECT
cars.CarMake,
cars.CarModel,
CASE WHEN #FlagOwnerName = 1 THEN cars.OwnerName ELSE 'ALL' END as OwnerName,
CASE WHEN #FlagOwnerName = 1 THEN cars.CarValue ELSE totals.CarValueTotal END as CarValue,
CASE WHEN #FlagOwnerName = 1 THEN cars.LicenseID ELSE totals.LicenseList END as LicenseID
FROM #cars cars
INNER JOIN
(
SELECT CarMake, CarModel, LicenseList, CarValueTotal
FROM (
SELECT CarMake, CarModel, LicenseList, CarValueTotal,
RANK() OVER ( PARTITION BY CarMake, CarModel ORDER BY length_ DESC )
FROM CTE
) D ( CarMake, CarModel, LicenseList, CarValueTotal, rank )
WHERE rank = 1
) totals ON cars.CarMake = totals.CarMake AND cars.CarModel = totals.CarModel
GROUP BY
cars.CarMake,
cars.CarModel,
CASE WHEN #FlagOwnerName = 1 THEN cars.OwnerName ELSE 'ALL' END,
CASE WHEN #FlagOwnerName = 1 THEN cars.CarValue ELSE totals.CarValueTotal END,
CASE WHEN #FlagOwnerName = 1 THEN cars.LicenseID ELSE totals.LicenseList END
So when #FlagOwnerName = 1 we get:
CarMake CarModel OwnerName CarValue LicenseID
Aston Martin Vantage Mike 90000 HY7733
Ford Mustang Amy 30000 JKJL88
Ford Mustang Billy-Bob 30000 EZ1111
Toyota Camry Bob 12000 HED999
Toyota Camry Helen 19000 WKS444
Toyota Camry Steve 12000 ABC123
And when #FlagOwnerName = 0 we get:
CarMake CarModel OwnerName CarValue LicenseID
Aston Martin Vantage ALL 90000 HY7733
Ford Mustang ALL 60000 EZ1111, JKJL88
Toyota Camry ALL 43000 ABC123, HED999, WKS444
Note that in your comment you implied you didn't want OwnerName to be returned when #FlagOwnerName = 0, while this is possible inside a stored procedure (i.e. executing different queries based on a parameter) I wouldn't recommend it. Better to have a consistent set of columns returned, if you're using a reporting tool over the top of it then you can potentially include some logic there to hide the column based on the parameter value.