Reason for NOT IN behaviour [duplicate]

Question

This question already has an answer here:

Why does NOT IN with a set containing NULL always return FALSE/NULL? (1 answer)

Closed 3 years ago.

sql> SELECT * FROM runners;
+----+--------------+
| id | name         |
+----+--------------+
|  1 | John Doe     |
|  2 | Jane Doe     |
|  3 | Alice Jones  |
|  4 | Bobby Louis  |
|  5 | Lisa Romero  |
+----+--------------+

sql> SELECT * FROM races;
+----+----------------+-----------+
| id | event          | winner_id |
+----+----------------+-----------+
|  1 | 100 meter dash |  2        |
|  2 | 500 meter dash |  3        |
|  3 | cross-country  |  2        |
|  4 | triathalon     |  NULL     |
+----+----------------+-----------+

What will be the result of the query below?

SELECT * FROM runners WHERE id NOT IN (SELECT winner_id FROM races);

The above query will return nothing because the races table contain null winner_id. But what is the reason behind this?

Answer 1

SELECT * FROM runners WHERE id NOT IN (SELECT winner_id FROM races);

Is something like this (actual values from races):

SELECT * FROM runners WHERE
id <> 2 and
id <> 3 and
id <> null;

id <> null is unknown by definition, so the whole where clause evaluates to either false (for ids: 2, 3) or unknown (for ids: 1, 4, 5) because of the and operators, thus no result returned.

If you want to run the query disregarding the nulls in races, you can use:

SELECT * FROM runners 
WHERE id NOT IN 
      (SELECT winner_id FROM races WHERE winner_id IS NOT NULL);

Answer 2

I would advice not doing what Balazs is doing. This is what NOT EXISTS is for, it handles null. Let's take a look at this..

Create sample data

CREATE TABLE foo AS
SELECT * FROM generate_series(1,100) AS t(x);

CREATE TABLE bar AS
SELECT * FROM generate_series(1,10) AS t(x)
  UNION SELECT null;

Refresher of problem

Now, as per the earlier answer, and the docs,

"NOT IN is equivalent to <> ALL()` Note that if there are no failures but at least one right-hand row yields null for the operator's result, the result of the ALL construct will be null, not true. This is in accordance with SQL's normal rules for Boolean combinations of null values.

SELECT *
FROM foo
WHERE foo.x NOT IN ( SELECT x FROM bar );

If any of bar.x is NULL the whole condition is NULL and thus returns 0 rows.

Solutions

The best solution is to skip NOT IN entirely if the set can be null and use a NOT EXISTS test.

SELECT *
FROM foo
WHERE NOT EXISTS (
  SELECT 1 FROM bar WHERE bar.x = foo.x
);

This will return 90 rows.

Alternatively this can be written as this..

SELECT *
FROM foo
LEFT OUTER JOIN bar
  ON foo.x = bar.x
WHERE bar.x IS NULL;