How to substract rows from one table from another only once - sql

Question

I'm working for a university project, and I have the following question:
I have 2 tables in a Oracle DB... I need to select those rows from table1, which are not included in table2... But the main problem is that I need to exclude that rows from table2 wich was selected once... For example:
Table1 Table2 ResultTable
id | Number | Letter id | Number | Letter id | Number | Letter
_____________________ _____________________ _____________________
1 4 S 1 6 G 2 2 P
2 2 P 2 8 B 3 5 B
3 5 B 3 4 S 4 4 S
4 4 S 4 1 A 6 2 P
5 1 A 5 1 H
6 2 P 6 2 X
So, how you see it, if one row from Table1 has a "twin" in Table2, they both are excluded.

Answer 1

Probably the most thorough query is this:
SELECT table1.id,
table1.digit,
table1.letter
FROM ( SELECT id,
digit,
letter,
ROW_NUMBER() OVER (PARTITION BY digit, letter ORDER BY id) rn
FROM table1
) table1
LEFT
JOIN ( SELECT id,
digit,
letter,
ROW_NUMBER() OVER (PARTITION BY digit, letter ORDER BY id) rn
FROM table2
) table2
ON table2.digit = table1.digit
AND table2.letter = table1.letter
AND table2.rn = table1.rn
WHERE table2.id IS NULL
ORDER
BY table1.id
;
which gives each record in table1 and table2 a "row number" within its group of "twins". For example, this:
SELECT id,
digit,
letter,
ROW_NUMBER() OVER (PARTITION BY digit, letter ORDER BY id) rn
FROM table1
ORDER
BY table1.id
;
returns this:
ID DIGIT LETT RN
---------- ---------- ---- ----------
1 4 S 1
2 2 P 1
3 5 B 1
4 4 S 2 -- second row with 4 S
5 1 A 1
6 2 P 2 -- second row with 2 P
That said, if you know that no (digit, letter) can ever appear more than once in table2, you can simplify this considerably by using EXISTS instead of ROW_NUMBER():
SELECT id,
digit,
letter
FROM table1 table1a
WHERE EXISTS
( SELECT 1
FROM table1
WHERE digit = table1a.digit
AND letter = table1a.letter
AND id < table1a.id
)
OR NOT EXISTS
( SELECT 1
FROM table2
WHERE digit = table1a.digit
AND letter = table1a.letter
)
;

Answer 2

Break it into parts.
Perhaps you have an EOR - Exclusive OR.
So you might have
(condition1
OR
condition2)
AND NOT
(condition1 AND condition2).

Answer 3

Use the Oracle MINUS keyword, which does exactly what you're asking for. See http://oreilly.com/catalog/mastorasql/chapter/ch07.html for more detail.

Answer 4

I can't see how to do what you want with one SQL SELECT.
I think you'll need a temporary table, and several statements.
Call it tmpResults, with id1 and id2, which match the id from Table1 and the id from Table2 respectively.
-- get matched rows - this is too many, we'll delete some later.
INSERT INTO tmpResults (id1, id2)
SELECT Table1.id id1, Table2.id id2
FROM Table1 INNER JOIN Table2
ON Table1.Number = Table2.Number AND Table1.Letter = Table2.Letter;
-- Delete where Table1 has matched more than 1 row
DELETE tmpResults
WHERE rowid IN
(SELECT tmpResults.RowId
FROM tmpResults
INNER JOIN
(SELECT id1, MAX(id2) id2m FROM tmpResults GROUP BY id1 HAVING count(*) > 1) m1
ON tmpResults.id1 = m1.id1 AND tmpResults.id2 = m1.id2m );
-- Delete where Table2 has matched more than 1 row
DELETE tmpResults
WHERE rowid IN
(SELECT tmpResults.RowId
FROM tmpResults
INNER JOIN
(SELECT MAX(id1) id1m, id2 FROM tmpResults GROUP BY id2 HAVING count(*) >1) m2
ON tmpResults.id1 = m2.id1m AND tmpResults.id2 = m2.id2 );
-- now tmpResults should have unique matches only, so we want Table1 where there is no match
SELECT Table1.*
FROM Table1
LEFT JOIN tmpResults
ON table1.id = tmpResults.id1
WHERE tmpResults.id2 IS NULL;