I noticed that I can have NULL values in columns that have the UNIQUE constraint: UNIQUE(col)
Would that generate any issues in certain situations?
3 Answers
While the following addresses multiple null values, it does not address any "issues" associated with such a design, other than possible database/SQL portability - as such, it should probably not be considered an answer, and is left here merely for reference.
This is actually covered in the SQLite FAQ. It is a design choice - SQLite (unlike SQL Server) chose that multiple NULL values do not count towards uniqueness in an index.
Perhaps you are referring to the following statement from SQL92:
- A unique constraint is satisfied if and only if no two rows in a table have the same non-null values in the unique columns.
That statement is ambiguous, having at least two possible interpretations:
A unique constraint is satisfied if and only if no two rows in a table have the same values and have non-null values in the unique columns.
A unique constraint is satisfied if and only if no two rows in a table have the same values in the subset of unique columns that are not null.
SQLite follows interpretation (1), as does PostgreSQL, MySQL, Oracle, and Firebird. It is true that Informix and Microsoft SQL Server use interpretation (2), however we the SQLite developers hold that interpretation (1) is the most natural reading of the requirement and we also want to maximize compatibility with other SQL database engines, and most other database engines also go with (1), so that is what SQLite does.
See a comparison of NULL handling.
4 Comments
Rei Miyasaka
I can't think of a situation in which (1) would be the useful interpretation.
Johannes Egger
I have one: Users have a key card with a unique id, which means no two users can have the same key. But some users don't have a key card anymore (NULL). And they might have given their key card to another user.
user2864740
@JohannesEgger One of those the times when SQL Server's filtered indices (stackoverflow.com/a/15206061/2864740) becomes very useful for behavior (1) semantics.
DimeCadmium
I can't think of a situation in which (2) would be the useful interpretation. What is the purpose of allowing NULL if there can only be one value with it? It would no longer be special, it would be just another value, and NULL is very much not just another value. As Johannes mentioned, it is useful for any situation where a column may not apply to certain rows (but must be unique if it does apply).
If you want your unique index to throw an error when a two rows would be the same if you disregard NULL columns (and don't want to use triggers from Satyam's answer) you can do something like this
CREATE TABLE `test` (
`Field1` INTEGER,
`Field2` INTEGER
);
CREATE UNIQUE INDEX `ix` ON `test` (
`Field1`,
`Field2`
);
INSERT INTO `test`(`Field1`,`Field2`) VALUES (1,NULL);
INSERT INTO `test`(`Field1`,`Field2`) VALUES (1,NULL); -- This shouldn't be allowed
DROP INDEX IF EXISTS `ix`;
CREATE UNIQUE INDEX `ix2` ON `test` (
`Field1`,
ifnull(`Field2`, 0) --use this instead
); --will fail
4 Comments
Steve V.
The ifnull() solution also causes inserts to fail for me on 'INSERT INTO
test(Field1,Field2) VALUES (2,3);' and I'm not sure why, but it's not what I expected. I solved the problem in the application instead of in the database, but I wanted to put that here for anyone who came after me.user2864740
Another option to to use a filtered index (stackoverflow.com/a/15206061/2864740).
HH321
you're right, here is a better link sqlite.org/partialindex.html, also see the note on when it can be used
Manngo
Nice trick. However, the code sample could be improved on by dropping the unnecessary back-ticks (or at least replacing them with the standard double quotes) and using the standard
coalesce(Field2,0) instead of ifnull().You can create 2 triggers on table to check if a row with column as null exists before any insert or update operation, if so raise exception.
CREATE TRIGGER UniqueColumnCheckNullInsert
BEFORE INSERT
ON 'Tablename'
WHEN NEW.'column_name' IS NULL
BEGIN
SELECT CASE WHEN((
SELECT 1
FROM 'Tablename'
WHERE 'column_name' IS NULL
)
NOTNULL) THEN RAISE(ABORT, "error row exists") END;
END;
CREATE TRIGGER UniqueColumnCheckNullUpdate
BEFORE UPDATE
ON 'Tablename'
WHEN NEW.'column_name' IS NULL
BEGIN
SELECT CASE WHEN((
SELECT 1
FROM 'Tablename'
WHERE 'column_name' IS NULL
)
NOTNULL) THEN RAISE(ABORT, "error row exists") END;
END;
1 Comment
phidrho
This blocks all other row inserts (on all columns) if at least one row has NULL in 'column_name', so it doesn't make much sense because field is nullable. This would make sense only on table with one field.
DELETEoperations should be as fast as finding the records.