Group by with having and merge back to original table - sql

Question

I have a table like this
A_Count
B_Count
A
B
C
1
0
A
NULL
C1
0
1
NULL
B
C1
1
1
A
B
C2
1
1
A
B
C2
and I want to have a result table (only need to show column A and B) like:
A_Count
B_Count
A
B
C
1
1
A
B
C1
1
1
A
B
C2
1
1
A
B
C2
So my goal is to merge two row having the following condiction:
both rows belong to same group C and only merge when one row has A being null and one row has B being null.
so its like:
group by C
having sum(A_COUNT) =1 AND sum(B_COUNT) =1
but the problem is, I want to keep those rows that are not merged (ROW 3 & 4) , can someone tell me how to do that? many thanks!

Answer 1

You can use conditional analytical function and group by as follows:
Select max(a) as a, max(b) as b, c from
(Select a, b, c,
case when nulla = 1 and nullb = 1 and (a is null or b is null)
then 0
else row_number() over (partition by c order by 1)
end as rn
from (Select a, b, c,
count(case when a is null then 1 end) over(partition by c) as nulla,
count(case when b is null then 1 end) over (partition by c) as nullb
From your_table t
)
)
Group by c, rn
DB<>Fiddle Thanks to MT0. Used the sample data from MT0's fiddle.

Answer 2

If you were using Oracle 12 then you could use MATCH_RECOGNIZE:
SELECT a_count, b_count, a, b, c
FROM (
SELECT t.*,
NVL2(
A,
ROW_NUMBER() OVER ( PARTITION BY C ORDER BY NVL2( B, 1, 0 ) DESC, ROWNUM ),
ROW_NUMBER() OVER ( PARTITION BY C ORDER BY NVL2( A, 1, 0 ) DESC, ROWNUM )
) AS rn
FROM table_name t
)
MATCH_RECOGNIZE (
PARTITION BY C
ORDER BY rn, A NULLS LAST
MEASURES
FIRST( a_count ) AS a_count,
LAST( b_count ) AS b_count,
FIRST( a ) AS a,
LAST( b ) AS b
PATTERN ( a b? )
DEFINE
a AS a.a IS NOT NULL,
b AS a.b IS NULL AND b.a IS NULL AND b.b IS NOT NULL
)
Before that Oracle version, you can get a similar effect using analytic functions to determine which rows to aggregate:
SELECT SUM( a_count ) AS a_count,
SUM( b_count ) AS b_count,
MAX( a ) AS a,
MAX( b ) AS b,
c
FROM (
SELECT t.*,
NVL2(
A,
ROW_NUMBER() OVER ( PARTITION BY C ORDER BY NVL2( B, 1, 0 ) DESC, ROWNUM ),
ROW_NUMBER() OVER ( PARTITION BY C ORDER BY NVL2( A, 1, 0 ) DESC, ROWNUM )
) AS rn
FROM table_name t
)
GROUP BY c, rn
Which, for the sample data (in an unordered state, with additional rows to demonstrate grouping additional pairs of rows):
CREATE TABLE table_name ( A_Count, B_Count, A, B, C ) AS
SELECT 1, 0, 'A', NULL, 'C1' FROM DUAL UNION ALL
SELECT 0, 1, NULL, 'B', 'C1' FROM DUAL UNION ALL
SELECT 1, 1, 'A', 'B', 'C2' FROM DUAL UNION ALL
SELECT 0, 1, NULL, 'B', 'C2' FROM DUAL UNION ALL -- Added row
SELECT 1, 0, 'A', NULL, 'C2' FROM DUAL UNION ALL -- Added row
SELECT 1, 0, 'A', NULL, 'C2' FROM DUAL UNION ALL -- Added row
SELECT 1, 1, 'A', 'B', 'C2' FROM DUAL UNION ALL
SELECT 0, 1, NULL, 'B', 'C2' FROM DUAL -- Added row
Both output:
A_COUNT | B_COUNT | A | B | C
------: | ------: | :- | :- | :-
1 | 1 | A | B | C1
1 | 1 | A | B | C2
1 | 1 | A | B | C2
1 | 1 | A | B | C2
1 | 1 | A | B | C2
db<>fiddle here

Answer 3

You can do this with join:
select (t1.a_count + coalesce(t2.a_count, 0)) as a_count,
(t1.b_count + coalesce(t2.b_count, 0)) as b_count,
coalesce(t1.a, t2.a) as a,
coalesce(t1.b, t2.b) as b,
t1.c
from t t1 left join
t t2
on t1.c = t2.c and
t1.a is not null and t2.a is null and
t1.b is null and t2.b is not null
where t1.a is not null;
As you've described the problem, aggregation doesn't seem necessary.
Here is a db<>fiddle with your original data.