Search each value of Column A in Column B in SQL Oracle

Let’s say I have this type of dataset:

row     Num_A   Num_B    Num_C    Reg_1    Reg_2
------  -----   -----    -----    -----    -----
   1    9400    9500     9700     AAA      CCC
   2    9700    9400     9500     BBB      DDD
   3    9500    9400     9200     EEE      AAA
   4    9100    9300     9900     AAA      DDD

Consider the following conditions:

In row 1, Num_B "95000" appears in Num_A (row 3) and the correspondent Num_A (row 1) "94000" is equal to its Num_B (row 3) (which is the case `”94000″)

So the new column Comm_1 will have the combination of Num_A and Num_B as its value. Same goes for Comm_2

As for Comm_3, the condition are same as Comm_1 or Comm_2, and an additional condition that Reg_1 (or Reg_2) of row 1 'AAA' (or 'CCC') is equal to Reg_1 or Reg_2 of row 3 (which is the case).

I want to construct a query so that the result would look like this:

    row     Num_A   Num_B    Num_C    Reg_1    Reg_2   Comm_1     Comm_2     Comm_3
    ------  -----   -----    -----    -----    -----   -----     -----     -----
       1    9400    9500     9700     AAA      CCC     94009500  94009700  94009500
       2    9700    9400     9500     BBB      DDD     97009400  97009500  NULL
       3    9500    9400     9200     EEE      AAA     95009400  NULL      95009400
       4    9100    9300     9900     AAA      DDD     NULL      NULL      NULL

I tried the following basic syntax but it isn’t working:

SELECT 
       IF(Num_B in (select distinct Num_A from mydata) , concat(Num_A,Num_B), Null) as Comm_1,
       IF(Num_C in (select distinct Num_A from mydata) , concat(Num_A,Num_C), Null) as Comm_2
FROM mydata

So please provide any hint and point me to reading about the right point in pl/sql or what ever procedure I need to learn.

Answer

You can get what (I think) you want with outer self joins:

-- CTE for sample data, including row_num to maintain display order
with mydata (row_num, Num_A, Num_B, Num_C, Reg_1, Reg_2) as (
  select 1, 94000, 95000, 97000, 'AAA', 'CCC' from dual
  union all select 2, 97000, 94000, 95000, 'BBB', 'DDD' from dual
  union all select 3, 95000, 94000, 92000, 'EEE', 'AAA' from dual
  union all select 4, 91000, 93000, 99000, 'AAA', 'DDD' from dual
)
-- actual query
select m.*,
  case when m1.num_a is not null then m.num_a || m1.num_a end as comm_1,
  case when m2.num_a is not null then m.num_a || m2.num_a end as comm_2,
  case when m3.num_a is not null then m.num_a || m3.num_a end as comm_3
from mydata m
left join mydata m1 on m1.num_a = m.num_b
left join mydata m2 on m2.num_a = m.num_c
left join mydata m3 on m3.num_a = m.num_b and (m3.reg_1 = m.reg_2 or m3.reg_2 = m.reg_1)
order by m.row_num;

   ROW_NUM      NUM_A      NUM_B      NUM_C REG REG COMM_1     COMM_2     COMM_3    
---------- ---------- ---------- ---------- --- --- ---------- ---------- ----------
         1      94000      95000      97000 AAA CCC 9400095000 9400097000 9400095000
         2      97000      94000      95000 BBB DDD 9700094000 9700095000           
         3      95000      94000      92000 EEE AAA 9500094000            9500094000
         4      91000      93000      99000 AAA DDD                                 

(I’ve left in row_num just to keep the display order the same, it isn’t used otherwise; I assume that isn’t actually a column in your table.)

The comm_1 value is based on a left join to another row which has its num_a matching this row’s num_b.

The comm_2 value is based on a left join to another row which has its num_a matching this row’s num_c.

The comm_3 value is the same as comm_1 but has to also match its reg_1 with the other row’s reg_2 or vice versa. But, this calculates row 3’s value as 9500094000 rather than the 94009500 shown in the question – not clear which is right though.

This will also get duplicate rows if there is more than one match for any of the outer joins; that isn’t the case with your sample data, but something to watch out for. I have no idea how you would want to handle that if it can happen.

---

You could also do it with a single outer join and more logic/aggregation in the column expressions:

select m.row_num, m.num_a, m.num_b, m.num_c, m.reg_1, m.reg_2,
  max(case when m1.num_a = m.num_b then m.num_a || m1.num_a end) as comm_1,
  max(case when m1.num_a = m.num_c then m.num_a || m1.num_a end) as comm_2,
  max(case when m1.num_a = m.num_b and (m1.reg_1 = m.reg_2 or m1.reg_2 = m.reg_1)
    then m.num_a || m1.num_a end) as comm_3
from mydata m
left join mydata m1 on (m1.num_a = m.num_b) or (m1.num_a = m.num_c)
group by m.row_num, m.num_a, m.num_b, m.num_c, m.reg_1, m.reg_2
order by m.row_num;

   ROW_NUM      NUM_A      NUM_B      NUM_C REG REG COMM_1     COMM_2     COMM_3    
---------- ---------- ---------- ---------- --- --- ---------- ---------- ----------
         1      94000      95000      97000 AAA CCC 9400095000 9400097000 9400095000
         2      97000      94000      95000 BBB DDD 9700094000 9700095000           
         3      95000      94000      92000 EEE AAA 9500094000            9500094000
         4      91000      93000      99000 AAA DDD