Remove duplicate (combination of 2 columns) values in a row - sql

Question

I have a requirement to remove duplicate values present in a row.
like :
C1 | C2 | C3 | C4 | C5 | C6
----------------------------
1 | 2 | 1 | 2 | 1 | 3
1 | 2 | 1 | 3 | 1 | 4
1 |NULL| 1 |NULL| 1 |NULL
OUTPUT of the query should be:
C1 | C2 | C3 | C4 | C5 | C6
----------------------------
1 | 2 | 1 | 3 |NULL|NULL
1 | 2 | 1 | 3 | 1 | 4
1 |NULL|NULL|NULL|NULL|NULL
As you can see combination of 2 columns should be unique in a row.
in Row 1:
combination of 1/2 is duplicate so its removed and 1/3 is in c5/c6 is moved to c3/c4
in Row 2:
there is no duplicate in the combination of 1/2 , 1/3, 1/4 so no change in the result
in Row 3:
All the 3 combinations are same like 1/NULL is present in all the combinations so c3 to c6 is set to null.
Thanks in advance

Answer 1

Maybe there is a more clever way... but you could convert them to pairs, distinct (union in this case does that), then pivot back.
with pairs as (
select id, c1 as x, c2 as y from mytable
union
select id, c3, c4 from mytable
union
select id, c5, c6 from mytable
)
select id,
max(decode(rn,1,x)) c1,
max(decode(rn,1,y)) c2,
max(decode(rn,2,x)) c3,
max(decode(rn,2,y)) c4,
max(decode(rn,3,x)) c5,
max(decode(rn,3,y)) c6
from (
select id, x, y, row_number() over (partition by id) rn
from pairs
) as foo
group by id

Answer 2

This one works - data included for testing, but might take some time to understand
A tip: un-comment the code snippets under the -- debug lines, copy the script until just these code snippets and paste this part into an SQL prompt to test the intermediate results.
The principle is get a row identifier to "remember" the rows; then to vertically pivot - not 3 columns to one, but 6 columns to 3 pairs of columns; then, use DISTINCT to de-dupe; then get an index within the row identifier of the de-duped intermediate rows; then use that index to pivot horizontally again.
Like so:
WITH
input(c1,c2,c3,c4,c5,c6) AS (
SELECT 1, 2,1, 2,1, 3
UNION ALL SELECT 1, 2,1, 3,1, 4
UNION ALL SELECT 1,NULL::INT,1,NULL::INT,1,NULL::INT
)
,
-- need rowid
input_with_rowid AS (
SELECT ROW_NUMBER() OVER() AS rowid, * FROM input
)
,
-- three groupy of 2 columns, so pivot using 3 indexes
idx3(idx) AS (SELECT 1 UNION SELECT 2 UNION SELECT 3)
,
-- pivot vertically, two columns at a time and de-dupe
pivot_pair AS (
SELECT DISTINCT
rowid
, CASE idx
WHEN 1 THEN c1
WHEN 2 THEN c3
WHEN 3 THEN c5
END AS c1
,
CASE idx
WHEN 1 THEN c2
WHEN 2 THEN c4
WHEN 3 THEN c6
END AS c2
FROM input_with_rowid CROSS JOIN idx3
)
-- debug
-- SELECT * FROM pivot_pair ORDER BY rowid;
,
-- add sequence per rowid
pivot_pair_with_seq AS (
SELECT
rowid
, ROW_NUMBER() OVER(PARTITION BY rowid) AS seq
, c1
, c2
FROM pivot_pair
)
-- debug
-- SELECT * FROM pivot_pair_with_seq;
SELECT
rowid
, MAX(CASE seq WHEN 1 THEN c1 END) AS c1
, MAX(CASE seq WHEN 1 THEN c2 END) AS c2
, MAX(CASE seq WHEN 2 THEN c1 END) AS c3
, MAX(CASE seq WHEN 2 THEN c2 END) AS c4
, MAX(CASE seq WHEN 3 THEN c1 END) AS c5
, MAX(CASE seq WHEN 3 THEN c2 END) AS c6
FROM pivot_pair_with_seq
GROUP BY rowid
ORDER BY rowid
;
rowid|c1|c2|c3|c4|c5|c6
1| 1| 2| 1| 3|- |-
2| 1| 2| 1| 3| 1| 4
3| 1|- |- |- |- |-

Answer 3

Using marcothesane's idea with pivot/unpivot operators. Easier to maintain if more input columns should be deduplicated. This maintains the order of source data (column pairs) - whereas marcothesane's solution might reorder column pairs depening on input data. Also it is a little slower than marcothesane's. It works only in 11R1 and up.
WITH
input(c1,c2,c3,c4,c5,c6) AS (
SELECT 1, 2,1, 2,1, 3 from dual
UNION ALL SELECT 1, 2,1, 3,1, 4 from dual
UNION ALL SELECT 1,NULL ,1,NULL ,1,NULL from dual
)
,
-- need rowid
input_with_rowid AS (
SELECT ROW_NUMBER() OVER (order by 1) AS row_id, input.* FROM input
),
unpivoted_pairs as
(
select row_id, tuple_idx, val1, val2, row_number() over (partition by row_id, val1, val2 order by tuple_idx) as keep_first
from input_with_rowid
UnPivot include nulls(
(val1, val2) --measure
for tuple_idx in ((c1,c2) as 1,
(c3,c4) as 2,
(c5,c6) as 3)
)
)
select row_id,
t1_val1 as c1,
t1_val2 as c2,
t2_val1 as c3,
t2_val2 as c4,
t3_val1 as c5,
t3_val2 as c6
from (
select row_id,
val1, val2, row_number() over (partition by row_id order by tuple_idx) as tuple_order
from unpivoted_pairs
where keep_first = 1
)
pivot (sum(val1) as val1, sum(val2) as val2
for tuple_order in ('1' as t1, '2' as t2, '3' as t3)
)