SQL Query - Indirect joining of two tables - sql

Question

I have two tables like the following
Table1
COL1 COL2 COL3
A 10 ABC
A 11 ABC
A 1 DEF
A 2 DEF
B 10 ABC
B 11 ABC
B 1 DEF
C 3 DEF
C 12 ABC
C 21 GHI
Table2
COL1 GHI ABC DEF
A1 21 10 1
A2 21 12 1
A3 21 10 1
A4 23 10 1
A5 25 11 3
A6 21 14 3
A7 25 11 1
A8 23 10 1
A9 29 10 2
A10 21 12 3
I have created another temporary table that returns all the distinct values from tbl1.col1
The values of col3 in tbl1 are columns in tbl2, which are populated by some values.
What I need is for each of these distinct values of table1.column1, (A, B, C) in this case, return a combination of table2.column1 and table1.column1 such that
the ABC value of table2.column1 matches any of the ABC value of the "group" from table1,
AND the DEF value of table2.column1 matches any of the DEF value of the "group" from table1,
AND IF THE GROUP CONTAINS GHI VALUES, the GHI value of table2.column1 matches any of the GHI value of the "group" from table1
So, I would need something like the following
Output Table
Table2.COL1 Table1.Col1
A1 A
A3 A
A4 A
A7 A
A8 A
A9 A
A1 B
A3 B
A4 B
A7 B
A8 B
A10 C
I tried something like this, but Im not sure if this is the right way of approaching
select table2.col1, temp_distinct_table.column1
from table2, temp_distinct_table
where table2.def IN (SELECT col2
FROM table1
WHERE table1.col1 = temp_distinct_table.col1
AND table1.col3 = 'DEF')
AND table2.abc IN (SELECT col2
FROM table1
WHERE table1.col1 = temp_distinct_table.col1
AND table1.col3 = 'ABC')
AND (
table2.ghi IN (SELECT col2
FROM table1
WHERE table1.col1 = temp_distinct_table.col1
AND table1.col3 = 'GHI')
OR NOT EXISTS (SELECT col2
FROM table1
WHERE table1.col1 = temp_distinct_table.col1
AND table1.col3 = 'GHI')
)
where temp_distinct_table contains of all the distinct values from table1.col1
Could someone guide me on the matter?

Answer 1

Another approach, counting how many matches there are for each t1.col/t2.col combination after joining all the possible matches:
select distinct t2_col1, t1_col1
from (
select t2.col1 as t2_col1, t1.col1 as t1_col1, t1.ghi_count as t1_ghi_count,
count(case when t1.col3 = 'ABC' then 1 end)
over (partition by t1.col1, t2.col1) as abc_matches,
count(case when t1.col3 = 'DEF' then 1 end)
over (partition by t1.col1, t2.col1) as def_matches,
count(case when t1.col3 = 'GHI' then 1 end)
over (partition by t1.col1, t2.col1) as ghi_matches
from (
select t1.*,
count(case when t1.col3 = 'GHI' then 1 end)
over (partition by t1.col1) as ghi_count
from table1 t1
) t1
join table2 t2
on (t1.col3 = 'ABC' and t2.abc = t1.col2)
or (t1.col3 = 'DEF' and t2.def = t1.col2)
or (t1.col3 = 'GHI' and t2.ghi = t1.col2)
)
where abc_matches > 0
and def_matches > 0
and (t1_ghi_count = 0 or ghi_matches > 0)
order by t1_col1, t2_col1;
Which with your sample data gets:
T2_COL T1_COL
------ ------
A1 A
A3 A
A4 A
A7 A
A8 A
A9 A
A1 B
A3 B
A4 B
A7 B
A8 B
A10 C
Not sure if the efficiency of that will be significantly different to MTO's cross join with your real data.

Answer 2

This becomes quite simple when you use collections (and you only need to do one table scan for each table):
Oracle Setup:
CREATE TYPE intlist AS TABLE OF INT;
/
Query:
SELECT t2.col1 AS t2_col1,
t1.col1 AS t1_col1
FROM (
SELECT col1,
CAST( COLLECT( CASE col3 WHEN 'ABC' THEN col2 END ) AS INTLIST ) AS abc,
CAST( COLLECT( CASE col3 WHEN 'DEF' THEN col2 END ) AS INTLIST ) AS def,
CAST( COLLECT( CASE col3 WHEN 'GHI' THEN col2 END ) AS INTLIST ) AS ghi
FROM table1
GROUP BY col1
) t1
INNER JOIN table2 t2
ON ( t2.abc MEMBER OF t1.abc
AND t2.def MEMBER OF t1.def
AND ( t2.ghi MEMBER OF t1.ghi OR t1.ghi IS EMPTY ) );
Output:
t2_col1 t1_col1
------- -------
A1 A
A3 A
A4 A
A7 A
A8 A
A9 A
A1 B
A3 B
A4 B
A7 B
A8 B
A10 C
Update
An alternative query without using collections (it is going to be more efficient than your query but probably less efficient than collections):
SELECT t2.col1,
t1.col1
FROM table1 t1
CROSS JOIN
table2 t2
GROUP BY t1.col1, t2.col1
HAVING COUNT( CASE WHEN t1.col2 = t2.abc AND t1.col3 = 'ABC' THEN 1 END ) > 0
AND COUNT( CASE WHEN t1.col2 = t2.def AND t1.col3 = 'DEF' THEN 1 END ) > 0
AND ( COUNT( CASE WHEN t1.col2 = t2.ghi AND t1.col3 = 'GHI' THEN 1 END ) > 0
OR COUNT( CASE t1.col3 WHEN 'GHI' THEN 1 END ) = 0 )
ORDER BY t1.col1, t2.col1;
Update 2:
Changed from CROSS JOIN to INNER JOIN:
SELECT t2.col1 AS t2_col1,
t1.col1 AS t1_col1
FROM (
SELECT t1.*,
COUNT( CASE col3 WHEN 'GHI' THEN 1 END )
OVER ( PARTITION BY col1 ) AS has_ghi
FROM table1 t1
) t1
INNER JOIN table2 t2
ON ( t1.col3 = 'ABC' AND t2.abc = t1.col2 )
OR ( t1.col3 = 'DEF' AND t2.def = t1.col2 )
OR ( t1.col3 = 'GHI' AND t2.ghi = t1.col2 )
GROUP BY t1.col1, t2.col1, t1.has_ghi
HAVING COUNT( CASE t1.col3 WHEN 'ABC' THEN 1 END ) > 0
AND COUNT( CASE t1.col3 WHEN 'DEF' THEN 1 END ) > 0
AND ( COUNT( CASE t1.col3 WHEN 'GHI' THEN 1 END ) > 0 OR has_ghi = 0 )
ORDER BY t1.col1, t2.col1;