对于 mysql 中的字符串列类型，NOT IN 未按预期工作

Question

I found something weird between NOT IN and IN where I use a string type column. It works fine for IN but not for NOT IN. Though both works (NOT IN and IN) for integer type column

SELECT * FROM `table1` t1 where t1.email not in (select t2.email from `table2` t2)

doesn't return any result from t1 which exist in t2 though it should as there are eligible records whereas

SELECT * FROM `table1` t1 where t1.email in (select t2.email from `table2` t2)

returns the records from t1 which are in t2.

Answer 1

Here is a simple example to explains this appearance which also occurs in other DBMS like SQL Server and Oracle.

Assume that

select t2.email from `table2` t2

returns

+-------------+
| email       |
+-------------+
| 111@abc.com |
| 222@abc.com |
| NULL        |
+-------------+

When execute

SELECT * FROM `table1` t1 where t1.email not in (select t2.email from `table2` t2)

MySQL translates it into

SELECT * FROM `table1` t1 
where 
t1.email <> '111@abc.com' AND t1.email <> '222@abc.com' AND t1.email <> NULL

The where clause always returns NULL because any value compare to NULL returns NULL and any boolean value do AND operation with NULL also returns NULL . Therefore, the whole SQL query as shown above always return non record.

Similarly,

SELECT * FROM `table1` t1 where t1.email in (select t2.email from `table2` t2)

will be translated into

SELECT * FROM `table1` t1 
where 
t1.email = '111@abc.com' OR t1.email = '222@abc.com' OR t1.email <> NULL

The value of t1.email <> NULL is NULL , which will be ignored if any other comparation is TRUE .

Answer 2

Modifying query to below worked.

SELECT * 
  FROM table1 t1 
 WHERE t1.email NOT IN ( SELECT t2.email 
                           FROM table2 t2 
                          WHERE t2.email IS NOT NULL )

Since table2's email column value was NULL for one row which was returning no records.

Answer 3

Don't use NOT IN with subqueries. The reason is simple: it does not handle NULL values as a person would expect them to be handled.

Instead, just get used to using NOT EXISTS :

SELECT t1.*
FROM `table1` t1 
WHERE NOT EXISTS (SELECT 1
                  FROM `table2` t2
                  WHERE t1.email= t2.email 
                 ) ;

Why does this not work as expected? This is because of how SQL defines NULL values. They have the semantics of an "unknown" value rather than a "missing" value.

Consider the following conditions:

• 1 IN (1, 2) -- evaluates to true
• 3 IN (1, 2) -- evaluates to false

NULL s do not affect this, because it is an exact match:

• 1 IN (1, 2, NULL) -- evaluates to true
• 3 IN (1, 2, NULL) -- evaluates to NULL, which is treated as false

But NULL s do affect NOT IN

• 1 NOT IN (1, 2, NULL) -- evaluates to false
• 3 NOT IN (1, 2, NULL) -- evaluates to NULL because NULL is "unknown"

The key is that NULL does not have a specific meaning -- it means "unknown". So, NULL could be equal to 3. Hence, the result of the last expression is "unknown" (ie NULL ) rather than "true".