How can I convert a hierarchical table to xml in SQL - sql-server

Question

I have a requirement to transform the result of a query into an xml fragment using t-SQL.
My table/query looks like this:
[Col1] [Col2] [Col3] qty
A group1 mod1 5
A group1 mod2 7
A group1 NULL 12
A group2 mod3 8
A group2 mod4 5
A group2 NULL 13
A NULL NULL 25
and the desired XML output should look like this
<A qty="25">
<group1 qty="12">
<mod1 qty="5"/>
<mod2 qty="7"/>
</group1>
<group2 qty="13">
<mod3 qty="8"/>
<mod4 qty="5"/>
</group2>
</A>
Is there a way to achieve this without pivoting the table and using dynamic sql which would be very messy

Answer 1

Good day, Please confirm that this solve your need. We can next work on improving performance. It is important to have the right indexes
DDL+DML
DROP TABLE IF EXISTS T
GO
CREATE TABLE T(
[Col1] VARCHAR(10),
[Col2] VARCHAR(10),
[Col3] VARCHAR(10),
qty VARCHAR(10)
)
GO
INSERT T([Col1],[Col2],[Col3],qty) VALUES
('A', 'group1', 'mod1', 5 ),
('A', 'group1', 'mod2', 7 ),
('A', 'group1', NULL , 12),
('A', 'group2', 'mod3', 8 ),
('A', 'group2', 'mod4', 5 ),
('A', 'group2', NULL , 13),
('A', NULL , NULL , 25)
GO
SELECT [Col1],[Col2],[Col3],qty
FROM T
GO
Solution
;with l1 as (
SELECT COL1, qty
FROM T
where col2 is null and col3 is null
)
, l2 as (
SELECT Col1, COL2, qty
FROM T
where col2 is not null and col3 is null
)
, l3 as (
SELECT Col1, COL2, COL3, qty
FROM T
where col2 is not null and col3 is not null
)
, a1 as (
SELECT COL2, X = STRING_AGG('<' + COL3 + ' qty="' + qty + '"/>' ,'')
FROM l3
GROUP BY Col2
)
, a2 as (
select X = '<' + a1.Col2 + ' qty="' + l2.qty + '">' + a1.X + '</' + a1.Col2 + '>', l2.Col1
from a1
LEFT JOIN l2 on a1.Col2 = l2.Col2
)
select CONVERT(XML,
'<' + l1.Col1 + ' qty="' + l1.qty + '">' + STRING_AGG(a2.X,'') + '</' + l1.Col1 + '>'
)
from a2
left join l1 on l1.Col1 = a2.Col1
group by l1.Col1, l1.qty
GO
Note! at the end I convert the value to XML in order to check that the result fit. In production there is no reason for this step, which cost resources.

Answer 2

#RonenAriely your solution inspired me. I've modified it a little. Because I could not run it.
My solution looks a little dirty and complicated and not elegant but the result is correct.
DECLARE #Tbl TABLE (Col1 CHAR(1), Col2 CHAR(6), Col3 CHAR(4), qty INT);
INSERT INTO #Tbl
VALUES ('A', 'group1', 'mod1', 5),
('A', 'group1', 'mod2', 7),
('A', 'group1', '', 12),
('A', 'group2', 'mod3', 8),
('A', 'group2', 'mod4', 5),
('A', 'group2', '', 13),
('A', '', '', 25);
WITH l1
AS (SELECT Col1,
Col2,
SUM(qty) AS qty FROM #Tbl WHERE Col2 <> ''
AND Col3 <> ''
AND Col2 <> ''
GROUP BY Col1,
Col2),
l2
AS (SELECT Col2,
SUM(qty) AS qty FROM #Tbl WHERE Col3 <> ''
GROUP BY Col2),
l3
AS (SELECT Col2,
Col3,
SUM(qty) AS qty FROM #Tbl
GROUP BY Col2,
Col3),
a1
AS (SELECT Col2,
'<' + Col3 + ' qty="' + CAST(qty AS VARCHAR(10)) + '"/>' AS c FROM l3 WHERE Col3 <> ''),
a2
AS (SELECT l1.Col1,
'<' + aa.Col2 + ' qty="' + CAST(l2.qty AS VARCHAR(10)) + '">' +
STUFF((SELECT DISTINCT '' + c FROM a1 WHERE Col2 = aa.Col2
FOR XML PATH ('')), 1, 0, '')
+ '</' + aa.Col2 + '>' AS c FROM a1 aa
LEFT JOIN l1
ON l1.Col2 = aa.Col2
LEFT JOIN l2
ON aa.Col2 = l2.Col2
GROUP BY l1.Col1,
aa.Col2,
l2.qty)
SELECT REPLACE(REPLACE(REPLACE('<' + aa.Col1 + ' qty="' + CAST(l1.qty AS VARCHAR(10)) + '">'
+ STUFF((SELECT DISTINCT '' + c FROM a2 WHERE Col1 = aa.Col1
FOR XML PATH ('')), 1, 0, '')
+ '</' + aa.Col1 + '>', '&', '&'), '<', '<'), '>', '>') AS c
FROM a2 aa
LEFT OUTER JOIN (SELECT Col1, SUM(qty) AS qty FROM l1 GROUP BY Col1) l1
ON l1.Col1 = aa.Col1
GROUP BY aa.Col1,
l1.qty;

Answer 3

Replying to my own post.
Thank you all for your help. In the end, I have implemented a solution using dynamic SQL as I have many blocks to process and I need a generic solution. My final code looks like this:
DECLARE #SQL nvarchar(4000)
SELECT #SQL = 'SELECT ' +
STUFF((SELECT ',' + qty + ' as '''
+ Col1
+ ISNULL('/'+ Col2,'')
+ ISNULL('/'+ Col3,'')
+ '/#qty''' FROM T ORDER BY Col1, Col2, Col3 FOR XML PATH(''))
,1,1,'') + ' FOR XML PATH('''')'
EXEC(#SQL)
Unfortunately, it cannot be implemented as a function (which was my initial goal) because it requires a physical table and because of the string size limitation. This would also not work for a large dataset with many sublevels. I will have to live with it!