combining groups in sql containing substrings - sql-server

Question

I apologize in advance for not explaining this very well.
I have a sql database with some data like this:
column1 | groups
3323052 | 3323052,3324794,3324795
3324794 | 3323052,3324794
3324794 | 3324794
3324794 | 3324794,3763369
3353586 | 3353586
3763369 | 3324794,3763369
I want to combine groups so that if a number is in two groups, the groups will combine and the number will only show up once in the list.
For example, the final result would look like this:
groups
3323052,3324794,3324795,3763369
3353586
I have been googling around without much luck. Any help is greatly appreciated.
Thanks.

Answer 1

So you want to recursively replace any items in groups -column with any values found from other rows with that value in column1? At least you can do it this way:
Split the data into rows, so there's just column1 -> group relation
Fetch any values that can be used as root nodes, my approach takes the smallest value because your data has a circle (3323052 -> 3324794 -> 3323052)
Fetch recursively all the value that can be found from the hierarchy under these root nodes
Put it back together into the original format
This example uses DelimitedSplit8k by Jeff Moden:
-- Step 1:
select distinct
d.column1,
convert(int, s.Item) as item
into #tmp
from
data d
cross apply DelimitedSplit8k(d.groups, ',') s
-- Step 2:
select distinct
column1
into #root
from #tmp t1
where not exists
(select 1 from #tmp t2 where t2.item = t1.column1 and t2.item > t2.column1)
-- Step 3:
;with CTE (root, parent, child) as (
select r.column1, r.column1, r.column1 from #root r
union all
select C.root, t.column1, t.item
from CTE C join #tmp t on t.column1 = C.child and t.item > C.parent
)
select distinct * into #results from CTE
-- Step 4:
SELECT r.column1, STUFF((SELECT distinct ', ' + convert(varchar(50), r2.child)
FROM #results r2
WHERE r2.root = r.column1
ORDER BY ', ' + convert(varchar(50), r2.child)
FOR XML PATH(N'')), 1, 2, '') as groups
FROM #root r
GROUP BY column1
ORDER BY column1
Result:
column1 groups
3323052 3323052, 3324794, 3324795, 3763369
3353586 3353586
I used temp. tables to be sure each of the steps is executed just once, but I believe it would be possible to do the whole thing with just one select and using CTEs instead of temp tables.
You can test this in SQL Fiddle