Full Outer Self Join [duplicate]

Question

The problem is to return the rows which contain nulls as well. Below is SQL code to create table and populate it with sample data.

I'm expecting below, but query does not show the two rows with null values.

src_t1  id1_t1  id2_t1  val_t1     src_t2  id1_t2  id2_t2  val_t2

                                        b       z      z        4
a        w       w      100             b       w      w        1
a        x       x      200             b       x      x        2
a        y       y      300

Data:

CREATE TABLE sample (
    src VARCHAR(6)
    ,id1 VARCHAR(6)
    ,id2 VARCHAR(6)
    ,val FLOAT
);

INSERT INTO sample (src, id1, id2, val)
VALUES ('a', 'w', 'w', 100)
      ,('b', 'w', 'w', 1) 
      ,('a', 'x', 'x', 200)
      ,('b', 'x', 'x', 2) 
      ,('a', 'y', 'y', 300)
      ,('b', 'z', 'z', 4) 
;

This is my test query. It does not show results when t1.src = 'a' and t1.id1 = 'y' or when t2.src = 'b' and t2.id1 = 'z'.

Why?

What's the correct query?

SELECT t1.src, t1.id1, t1.id2, t1.val
  ,t2.src as src2, t2.id1, t2.id2, t2.val
FROM sample t1 FULL OUTER JOIN sample t2
  ON t1.id1 = t2.id1 AND t1.id2 = t2.id2
WHERE (t1.src = 'a' AND t2.src = 'b') 
  OR (t1.src IS NULL AND t1.id1 IS NULL AND t1.id2 IS NULL)
  OR (t2.src IS NULL AND t2.id1 IS NULL AND t2.id2 IS NULL)

I've also tried moving the conditions in the WHERE clause to the ON clause as well.

TIA.

Answer 1

The WHERE clause evaluates too late, effectively converting your query into an inner join.

Instead, write your query like this using proper JOIN syntax:

SELECT t1.src, t1.id1, t1.id2, t1.val
  ,t2.src as src2, t2.id1, t2.id2, t2.val
FROM (
    select * from sample
    where src='a'
) t1 FULL OUTER JOIN (
    select * from sample 
    where src='b'
)  t2
  ON t1.id1 = t2.id1 AND t1.id2 = t2.id2

yielding this result set:

src  id1  id2  val         src2 id1  id2  val
---- ---- ---- ----------- ---- ---- ---- -----------
a    w    w    100         b    w    w    1
a    x    x    200         b    x    x    2
NULL NULL NULL NULL        b    z    z    4
a    y    y    300         NULL NULL NULL NULL

Update:
Note also the use of two sub-queries to clearly separate the source table into two distinct relvars. I missed this for a minute on my first submission.

Answer 2

Actually, I think the solution is a bit cleaner if a CTE is used:

WITH A AS (
  select * from sample where src='a'
),
B AS (
  select * from sample where src='b'
)
SELECT *
FROM A FULL OUTER JOIN B
    ON A.ID1 = B.ID1 AND A.ID2 = B.ID2
;