All possible combinations unpivoted - sql

Question

I've been working on a problem that I can't quite figure out. I've tried different combinations of cross joins, CTEs, windowing functions, etc but could never quite get there. I'm also not wanting to go the dynamic SQL route. Can someone please help?
Given a variable set of grouped values produce all possible combinations vertically (derived group, value)
Additional info:
No 2 combinations should have the same set of values, regardless of
order. Example: If you already have (1,2) then don't produce (2,1),
if (1,2,3) then no (1,3,2),(2,1,3),(2,3,1),(3,1,2),(3,2,1)
Values of the same group should not combine
all values are unique, regardless of group. The only reason for the
initial grouping is to apply rule #2
Example:
Given the starting groups and values of
InputGroup Value
1 8
2 7
2 9
3 1
3 6
3 3
Produce this output
OutputGroup Value
1 8
2 7
3 9
4 1
5 6
6 3
7 8
7 7
8 8
8 9
9 8
9 1
10 8
10 6
11 8
11 3
12 7
12 1
13 7
13 6
14 7
14 3
15 9
15 1
16 9
16 6
17 9
17 3
18 8
18 7
18 1
19 8
19 7
19 6
20 8
20 7
20 3
21 8
21 9
21 1
22 8
22 9
22 6
23 8
23 9
23 3
Here's the manual, non-vertical method for producing the output
CREATE TABLE #temp1 (GroupID INT, MyValue INT)
INSERT INTO #temp1 (GroupID, MyValue)
VALUES (1,8),(2,7),(2,9),(3,1),(3,6),(3,3)
--1st set of possibilities
SELECT MyValue
FROM #temp1
--2nd set of possibilities
SELECT a.MyValue, b.MyValue
FROM #temp1 a
JOIN #temp1 b
ON a.GroupID < b.GroupID
--3rd set
SELECT a.MyValue, b.MyValue, c.MyValue
FROM #temp1 a
JOIN #temp1 b
ON a.GroupID < b.GroupID
JOIN #temp1 c
ON b.GroupID < c.GroupID
DROP TABLE #temp1
My problem is that there can be a variable number of starting values
With this in mind, my output needs to be in grouped vertical sets so I'm only returning 2 columns. 1 that groups the numbers together and the number itself.
For this specific example there should be 46 rows with 23 distinct groups as shown above
I wrote CTE that I kept modifying and finally scrapped:
WITH MyCTE
AS (SELECT 1 AS Level, DENSE_RANK() OVER (ORDER BY GroupID, MyValue) AS DgroupID, GroupID, MyValue
FROM #temp1
UNION ALL
SELECT a.Level + 1, DENSE_RANK() OVER (ORDER BY b.GroupID, b.MyValue), b.GroupID, b.MyValue
FROM MyCTE a
JOIN #temp1 b
ON a.GroupID < b.GroupID)
SELECT DENSE_RANK() OVER (ORDER BY Level, DgroupID), MyValue
FROM MyCTE
The obvious problems with this:
1) The windowing function I used to give an incremental value to each row didn't work as expected. This is probably due to the way CTEs work. Good for performance, bad for me. The ROW_NUMBER windowing function does the same thing. All I'm trying to do there is to autoincrement the rows within each iteration so I can identify the group when the table gets "unpivoted". I believe the reason CTEs are so fast is because they're actually set-based operations so even though there's recursion I can't rely on the loop/iteration mode of thinking to produce the intended result. Feel free to correct me in all of my assumptions
2) Unpivoting. I need to take a set of rows and unpivot the columns into rows, with each keeping the identifier of the original row to show they are grouped together. SQL Server has a wonderful command called UNPIVOT which doesn't help me at all because you need to know how many columns you're unpivoting at design time. The whole point of this is to be able to provide a variable number of inputs and produce a predictable output

Answer 1

So, you are trying to group all "Group 1" values with all "Group 2" values and all "Group 3" values, but prevent the duplicates as you stated of ex: 1,2 and 2,1. You manual approach looks ok, however I'm not getting why you are comparing groups as opposed to the "Values" being less than the prior such that..
SELECT a.MyValue, b.MyValue
FROM #temp1 a
JOIN #temp1 b
ON a.MyValue < b.MyValue AND a.GroupID <> b.GroupID
--3rd set
SELECT a.MyValue, b.MyValue, c.MyValue
FROM #temp1 a
JOIN #temp1 b
ON a.MyValue < b.MyValue AND a.GroupID <> b.GroupID
JOIN #temp1 c
ON b.MyValue < c.MyValue AND a.GroupID <> c.GroupID AND b.GroupID <> c.GroupID
Per your feedback, the above adjustments should work, it would just take take extra muscle as it would have to permeate through Group 1, Group 2 and also Group 2, Group 1 since the number 1 could exist in group 2, but group 1 has a low number of 5. If you are always qualifying a.Group less than b.Group, you would never have the value 1 in the first position as group 2 is greater than group 1.
Does that make sense to your scenario?