Oracle join view for best matches - sql

Question

I wish to create a view which joins two tables together.
T1 =
Col1 Col2
AA BB
EE FF
YY ZZ
11 00
T2 =
Col1 Col2 Col3
AA BB 1
AA CC 2
CC BB 3
GG FF 4
GG HH 5
EE HH 6
XX YY 7
XX WW 8
YY RR 9
The rules for this view are a Best match scenario based upon the following rules:
1. Return Col3 from T2 if T1.Col1 & T1.Col2 = T2.Col1 & T2.Col2
ELSE
2. Return Col3 if T1.Col2 = T2.Col2
ELSE
3. Return Col3 if T1.Col = T2.Col1
ELSE
4. Return NULL
So in these examples I would expect the final view to contain:
AA BB 1 (Rule 1 match)
EE FF 4 (Rule 2 match)
YY ZZ 9 (Rule 3 match)
11 00 NULL (Rule 4 match)
The difficulty I am having is in the cases where it hits multiple rules (e.g. Rows 1 and 3 where rules 1 and 2 are hit or Rows 4 and 6 where rules 2 and 3 are hit separately).
I realise in this example that Rule 3 is hit multiple times - this is fine as the idea is it will only hit rule 3 when the other rules aren't true which should only ever yield 1 result (like in example 3).
Is there a way to do a similar union to cater for these cascading rules or will additional views need to creating with pre-filtering (such as having count < 2)?
A formula for this in excel would be:
=IF(AND(A3=$F$2,B3=$G$2),"Rule1",IF((B3=$G$2),"Rule 2",IF((A3=$F$2),"Rule 3","NULL")))
Where A3 = T2.Col1, B3 = T2.Col2 G2 = T1.Col2 and F2 = T1.Col1.

Answer 1

I'd do it like this:
with t1 as (select 'AA' col1, 'BB' col2 from dual union all
select 'EE' col1, 'FF' col2 from dual union all
select 'YY' col1, 'ZZ' col2 from dual union all
select '11' col1, '00' col2 from dual),
t2 as (select 'AA' col1, 'BB' col2, 1 col3 from dual union all
select 'AA' col1, 'CC' col2, 2 col3 from dual union all
select 'CC' col1, 'BB' col2, 3 col3 from dual union all
select 'GG' col1, 'FF' col2, 4 col3 from dual union all
select 'GG' col1, 'HH' col2, 5 col3 from dual union all
select 'EE' col1, 'HH' col2, 6 col3 from dual union all
select 'XX' col1, 'YY' col2, 7 col3 from dual union all
select 'XX' col1, 'WW' col2, 8 col3 from dual union all
select 'YY' col1, 'RR' col2, 9 col3 from dual),
res as (select t1.col1,
t1.col2,
t2.col3,
case when t1.col1 = t2.col1 and t1.col2 = t2.col2 then 1
when t1.col2 = t2.col2 then 2
when t1.col1 = t2.col1 then 3
end join_level,
min (case when t1.col1 = t2.col1 and t1.col2 = t2.col2 then 1
when t1.col2 = t2.col2 then 2
when t1.col1 = t2.col1 then 3
end) over (partition by t1.col1, t1.col2) min_join_level
from t1
left outer join t2 on (t1.col1 = t2.col1 or t1.col2 = t2.col2))
select col1,
col2,
col3
from res
where join_level = min_join_level
or join_level is null;
COL1 COL2 COL3
---- ---- ----------
11 00
AA BB 1
EE FF 4
YY ZZ 9
Ie. do the joins first (in this case, t1 left outer join t2 on (t2.col1 = t1.col1 or t2.col2 = t1.col2) includes rows where t1.col1 = t2.col1 and t1.col2 = t2.col2), and then filter the results based on which join condition takes precedence.
Here's a slightly different alternative, using aggregates instead of analytic functions like the above answer:
with t1 as (select 'AA' col1, 'BB' col2 from dual union all
select 'EE' col1, 'FF' col2 from dual union all
select 'YY' col1, 'ZZ' col2 from dual union all
select '11' col1, '00' col2 from dual),
t2 as (select 'AA' col1, 'BB' col2, 1 col3 from dual union all
select 'AA' col1, 'CC' col2, 2 col3 from dual union all
select 'CC' col1, 'BB' col2, 3 col3 from dual union all
select 'GG' col1, 'FF' col2, 4 col3 from dual union all
select 'GG' col1, 'HH' col2, 5 col3 from dual union all
select 'EE' col1, 'HH' col2, 6 col3 from dual union all
select 'XX' col1, 'YY' col2, 7 col3 from dual union all
select 'XX' col1, 'WW' col2, 8 col3 from dual union all
select 'YY' col1, 'RR' col2, 9 col3 from dual)
select t1.col1,
t1.col2,
min(t2.col3) keep (dense_rank first order by case when t1.col1 = t2.col1 and t1.col2 = t2.col2 then 1
when t1.col2 = t2.col2 then 2
when t1.col1 = t2.col1 then 3
end) col3
from t1
left outer join t2 on (t1.col1 = t2.col1 or t1.col2 = t2.col2)
group by t1.col1,
t1.col2;
COL1 COL2 COL3
---- ---- ----------
11 00
AA BB 1
EE FF 4
YY ZZ 9
N.B. These could return different results if there happened to be more than one row that met the highest priority available join condition. The first query would return each row with a (potentially) different col3, whereas the second query would return just one row, with the lowest available col3 value.
What would you expect to see if T2 contained:
COL1 COL2 COL3
---- ---- ----------
AA BB 1
AA CC 2
CC BB 3
GG FF 4
GG HH 5
EE HH 6
XX YY 7
XX WW 8
YY RR 9
YY SS 10
The first query will give you:
COL1 COL2 COL3
---- ---- ----------
11 00
AA BB 1
EE FF 4
YY ZZ 10
YY ZZ 9
The second query will give you:
COL1 COL2 COL3
---- ---- ----------
11 00
AA BB 1
EE FF 4
YY ZZ 9

Answer 2

Perhaps this method, which chains together the result sets of three common table expressions, each of which implements a different join and checks whether the rowid of the row in T1 has already been projected from a successful join:
with
first_join as (
select t1.col1,
t1.col2,
t2.col3,
t1.rowid
from t1 join t2 on t1.col1 = t2.col1 and t1.col2 = t2.col2),
second_join as (
select t1.col1,
t1.col2,
t2.col3,
t1.rowid
from t1 join t2 on t1.col2 = t2.col2
where t1.rowid not in (select rowid from first_join)),
third_join as (
select t1.col1,
t1.col2,
t2.col3,
from t1 join t2 on t1.col1 = t2.col1
where t1.rowid not in (select rowid from first_join union all
select rowid from second_join))
select col1, col2, col3 from first_join union all
select col1, col2, col3 from second_join union all
select col1, col2, col3 from third_join