Combine multiple rows into list for multiple columns - sql-server

Question

I'm aware that the "combine multiple rows into list" question has been answered a million times, and here's a reference to an awesome article: Concatenating row values in transact sql
I have a need to combine multiple rows into lists for multiple columns at the same time
ID | Col1 | Col2 ID | Col1 | Col2
------------------ => ------------------
1 A X 1 A X
2 B Y 2 B,C Y,Z
2 C Z
I tried to use the xml method, but this has proven to be very slow over large tables
SELECT DISTINCT
[ID],
[Col1] = STUFF((SELECT ',' + t2.[Col1]
FROM #Table t2
WHERE t2.ID = t.ID
FOR XML PATH(''), TYPE).value('.', 'nvarchar(max)'),1,1,''),
[Col2] = STUFF((SELECT ',' + t2.[Col2]
FROM #Table t2
WHERE t2.ID = t.ID
FOR XML PATH(''), TYPE).value('.', 'nvarchar(max)'),1,1,''),
FROM #Table t
My current solution is to use a stored procedure that builds each ID row separately. I'm wondering if there's another approach I could use (other than using a loop)
For each column, rank the rows to combine (partition by the key column)
End up with a table like
ID | Col1 | Col2 | Col1Rank | Col2Rank
1 A X 1 1
2 B Y 1 1
2 C Z 2 2
Create a new table containing top rank columns for each ID
ID | Col1Comb | Col2Comb
1 A X
2 B Y
Loop through each remaining rank in increasing order (in this case 1 iteration)
for irank = 0; irank <= 1; irank++
update n set
n.col1Comb = n.Col1Comb + ',' + o.Col1, -- so append the rank 2 items
n.col2comb = n.Col2Comb + ',' + o.Col2 -- if they are not null
from #newtable n
join #oldtable o
on o.ID = n.ID
where o.col1rank = irank or o.col2rank = irank

Answer 1

A CTE trick can be used where you update the CTE.
Method 1: a new parallel table to which the data is copied and then concatenated:
CREATE TABLE #Table1(ID INT, Col1 VARCHAR(1), Col2 VARCHAR(1), RowID INT IDENTITY(1,1));
CREATE TABLE #Table1Concat(ID INT, Col3 VARCHAR(MAX), Col4 VARCHAR(MAX), RowID INT);
GO
INSERT #Table1 VALUES(1,'A','X'), (2,'B','Y'), (2,'C','Z');
GO
INSERT #Table1Concat
SELECT * FROM #Table1;
GO
DECLARE #Cat1 VARCHAR(MAX) = '';
DECLARE #Cat2 VARCHAR(MAX) = '';
; WITH CTE AS (
SELECT TOP 2147483647 t1.*, t2.Col3, t2.Col4, r = ROW_NUMBER()OVER(PARTITION BY t1.ID ORDER BY t1.Col1, t1.Col2)
FROM #Table1 t1
JOIN #Table1Concat t2 ON t1.RowID = t2.RowID
ORDER BY t1.ID, t1.Col1, t1.Col2
)
UPDATE CTE
SET #Cat1 = Col3 = CASE r WHEN 1 THEN ISNULL(Col1,'') ELSE #Cat1 + ',' + Col1 END
, #Cat2 = Col4 = CASE r WHEN 1 THEN ISNULL(Col2,'') ELSE #Cat2 + ',' + Col2 END;
GO
SELECT ID, Col3 = MAX(Col3)
, Col4 = MAX(Col4)
FROM #Table1Concat
GROUP BY ID
Method 2: Add the concatenation columns directly to the original table and concatenate the new columns:
CREATE TABLE #Table1(ID INT, Col1 VARCHAR(1), Col2 VARCHAR(1), Col1Cat VARCHAR(MAX), Col2Cat VARCHAR(MAX));
GO
INSERT #Table1(ID,Col1,Col2) VALUES(1,'A','X'), (2,'B','Y'), (2,'C','Z');
GO
DECLARE #Cat1 VARCHAR(MAX) = '';
DECLARE #Cat2 VARCHAR(MAX) = '';
; WITH CTE AS (
SELECT TOP 2147483647 t1.*, r = ROW_NUMBER()OVER(PARTITION BY t1.ID ORDER BY t1.Col1, t1.Col2)
FROM #Table1 t1
ORDER BY t1.ID, t1.Col1, t1.Col2
)
UPDATE CTE
SET #Cat1 = Col1Cat = CASE r WHEN 1 THEN ISNULL(Col1,'') ELSE #Cat1 + ',' + Col1 END
, #Cat2 = Col2Cat = CASE r WHEN 1 THEN ISNULL(Col2,'') ELSE #Cat2 + ',' + Col2 END;
GO
SELECT ID, Col1Cat = MAX(Col1Cat)
, Col2Cat = MAX(Col2Cat)
FROM #Table1
GROUP BY ID;
GO

Answer 2

Try this one -
Query1:
DECLARE #temp TABLE
(
ID INT
, Col1 VARCHAR(30)
, Col2 VARCHAR(30)
)
INSERT INTO #temp (ID, Col1, Col2)
VALUES
(1, 'A', 'X'),
(2, 'B', 'Y'),
(2, 'C', 'Z')
SELECT
r.ID
, Col1 = STUFF(REPLACE(REPLACE(CAST(d.x.query('/t1/a') AS VARCHAR(MAX)), '<a>', ','), '</a>', ''), 1, 1, '')
, Col2 = STUFF(REPLACE(REPLACE(CAST(d.x.query('/t2/a') AS VARCHAR(MAX)), '<a>', ','), '</a>', ''), 1, 1, '')
FROM (
SELECT DISTINCT ID
FROM #temp
) r
OUTER APPLY (
SELECT x = CAST((
SELECT
[t1/a] = t2.Col1
, [t2/a] = t2.Col2
FROM #temp t2
WHERE r.ID = t2.ID
FOR XML PATH('')
) AS XML)
) d
Query 2:
SELECT
r.ID
, Col1 = STUFF(REPLACE(CAST(d.x.query('for $a in /a return xs:string($a)') AS VARCHAR(MAX)), ' ,', ','), 1, 1, '')
, Col2 = STUFF(REPLACE(CAST(d.x.query('for $b in /b return xs:string($b)') AS VARCHAR(MAX)), ' ,', ','), 1, 1, '')
FROM (
SELECT DISTINCT ID
FROM #temp
) r
OUTER APPLY (
SELECT x = CAST((
SELECT
[a] = ',' + t2.Col1
, [b] = ',' + t2.Col2
FROM #temp t2
WHERE r.ID = t2.ID
FOR XML PATH('')
) AS XML)
) d
Output:
ID Col1 Col2
----------- ---------- ----------
1 A X
2 B,C Y,Z

Answer 3

One solution, one that is at least syntactically straight-forward, is to use a User-Defined Aggregate to "Join" the values together. This does require SQLCLR and while some folks are reluctant to enable it, it does provide for a set-based approach that does not need to re-query the base table per each column. Joining is the opposite of Splitting and will create a comma-separated list of what was individual rows.
Below is a simple example that uses the SQL# (SQLsharp) library which comes with a User-Defined Aggregate named Agg_Join() that does exactly what is being asked for here. You can download the Free version of SQL# from http://www.SQLsharp.com/ and the example SELECTs from a standard system view. (And to be fair, I am the author of SQL# but this function is available for free).
SELECT sc.[object_id],
OBJECT_NAME(sc.[object_id]) AS [ObjectName],
SQL#.Agg_Join(sc.name) AS [ColumnNames],
SQL#.Agg_Join(DISTINCT sc.system_type_id) AS [DataTypes]
FROM sys.columns sc
GROUP BY sc.[object_id]
I recommend testing this against your current solution(s) to see which is the fastest for the volume of data you expect to have in at least the next year or two.