how to group parent-child rows (TSQL) - tsql

Question

I have a SQL table with transactions some are in a parent-child relationship. But the relationship is only determined by a line and a type column. There is no other reference. I have to build this reference.
DECLARE #tmp TABLE (line INT, type CHAR(1), product VARCHAR(30))
INSERT #tmp VALUES
( 1,' ','22411')
,( 2,' ','22413')
,( 3,'P','27050')
,( 4,'C','22492')
,( 5,'C','22493')
,( 6,'C','22490')
,( 7,' ','22410')
,( 8,' ','22511')
,( 9,'P','27051')
,(10,'C','22470')
,(11,'C','22471')
,(12,'C','22473')
,(13,'C','22474')
,(14,' ','22015')
,(15,' ','22167')
,(16,' ','12411')
,(17,' ','22500')
Line 3 is a parent product. Lines 4 to 6 are the child rows.
Line 9 is another parent product. Lines 10 to 13 are the child rows.
Desired output is something like this, where the parent-child lines are grouped:
line|type|product|group
1 | |22411 |1
2 | |22413 |2
3 |P |27050 |3
4 |C |22492 |3
5 |C |22493 |3
6 |C |22490 |3
7 | |22410 |7
8 | |22511 |8
9 |P |27051 |9
10 |C |22470 |9
11 |C |22471 |9
12 |C |22473 |9
13 |C |22474 |9
14 | |22015 |14
15 | |22167 |15
16 | |12411 |16
17 | |22500 |17
How to achieve this without a cursor?

Answer 1

If input is always valid, you can use below query.
SELECT *, RANK() OVER(ORDER BY gid) AS [group]
FROM
(
SELECT *, SUM(CASE WHEN type = 'C' AND (prev = 'P' OR prev = 'C') THEN 0 ELSE 1 END) OVER(ORDER BY line) AS gid
FROM
(
SELECT *, LAG(type) OVER(ORDER BY line) AS prev
FROM #tmp
) AS withPreviouLine
) AS grouped
It can't handle continuous 'C' without 'P'. LAG is added starting with SQL SERVER 2012.

Answer 2

This will work for almost all version of SQL Server, since you didn't specify what version you're using.
But if you're using at least SQL Server 2012, then you can use #qxg's solution, since it's simpler.
Here's the code that you need, albeit it is not a single query, but it gives you the result you want:
CREATE TABLE #tmp (line INT, type CHAR(1), product VARCHAR(30))
INSERT #tmp VALUES ( 1,' ','22411')
,( 2,' ','22413')
,( 3,'P','27050')
,( 4,'C','22492')
,( 5,'C','22493')
,( 6,'C','22490')
,( 7,' ','22410')
,( 8,' ','22511')
,( 9,'P','27051')
,(10,'C','22470')
,(11,'C','22471')
,(12,'C','22473')
,(13,'C','22474')
,(14,' ','22015')
,(15,' ','22167')
,(16,' ','12411')
,(17,' ','22500')
select t1.*
, case
when t1.type = 'C'
and t3.type = ''
then 'G'
when t1.type = 'P' or t1.type = 'C'
then 'G'
else 'N'
end [same_group]
into #tmp2
from #tmp t1
left join #tmp t2 on t1.line = t2.line + 1
left join #tmp t3 on t1.line = t3.line - 1
order by t1.line
select *
, case
when t.type <> ''
then (select max(line)
from #tmp2
where same_group = 'G'
and type = 'P'
and line <= t.line)
else t.line
end [group_id]
from #tmp2 t
order by line
You probably could refactor it to be a single query, but I don't have the time to do so at the moment.