With SQL Server you can however work around the problem with indexed views (as of SQL Server 2000) or filtered indexes (as of SQL Server 2008).

For more details on the SQL Server bug, see https://connect.microsoft.com/SQLServer/feedback/ViewFeedback.aspx?FeedbackID=299229

About the error:
An explicit DROP INDEX is not allowed on index…

Dont use the UI editor
Use query analyser to drop the constraint first:
ALTER TABLE table_name DROP CONSTRAINT FK_constraint_name
and then recreate the constraint(then you can do in the UI editor)

Drop failed for Index ‘IX_TABLE_A’. (Microsoft.SqlServer.Smo)

——————————
ADDITIONAL INFORMATION:

An exception occurred while executing a Transact-SQL statement or batch. (Microsoft.SqlServer.ConnectionInfo)

——————————

An explicit DROP INDEX is not allowed on index ‘dbo.TABLE_A.IX_TABLE_A’. It is being used for FOREIGN KEY constraint enforcement. (Microsoft SQL Server, Error: 3723)

– optional instead of trigger to disable the insert directly into the table
CREATE TRIGGER trg_t1_DisableInsert ON t1
INSTEAD OF INSERT
AS
BEGIN
— use 18 to stop further processing
RAISERROR (40000, 18, 1, ‘Use view dbo.v1 to insert data into table’)
END
GO

CREATE VIEW v1
WITH SCHEMABINDING
AS
SELECT id, title
FROM dbo.t1
WHERE id IS NOT NULL
GO
–Create an index on the view.
CREATE UNIQUE CLUSTERED INDEX CI_V1_ID
ON dbo.v1 (id);
GO

INSERT INTO v1 (id, title)
SELECT 1, ‘title 1′ UNION ALL
SELECT 2, ‘title 2′ UNION ALL
SELECT 3, ‘title 3′ UNION ALL
SELECT 4, ‘title 4′ UNION ALL
SELECT null, ‘title null’

– this fails because it clashes with the clustered index on the view
INSERT INTO v1 (id, title)
SELECT 1, ‘title 5′

– this is ok because it doesn’t clash.
INSERT INTO v1 (id, title)
SELECT NULL, ‘title 5′

– we see that the table can contain only unique values and more than one null
SELECT * FROM t1
SELECT * FROM v1

GO
DROP VIEW dbo.v1
GO
DROP TABLE dbo.t1

user_role
user_role_ID PK
user_id FK
role_id FK

where a combination of user_id and role_id is unique. Is it a good idea to build a composite unique index on those kind of join tables or it is not necessary.

Thanks.

create view xy
with schemabinding
as
select
uniquevalue
from dbo.table
where
uniquevalue is not NULL

create unique clustered index cui on dbo.table(uniquevalue)

IMHO the only reasonable solution

Regards
Christop Ingenhaag

I think that the solution to this would be to use GUIDs (Type: UniqueIdentifier) as the datatype for the Primary Key row. This would ensure that no PK value would ever equal the unique column value. You would still have to put logic into your application to determine whether or not a real value or GUID-in-place-of-null is being used.

1. create a computed column.
2. Pass the ‘unique column’ value when it is not null
3. When it is null copy the Primary key’s value into the computed column.
4. Set the unique constraint on computed column

I wish all of blog reader read this particular thread. This is very interesting and in depth discussion of the subject. I really like the way all of you are participating to help community.

@Imran,

I really like your additional explanation and politeness towards other experts.

Regards,
Pinal Dave ( http://www.SQLAuthority.com )

My first statement says… “I am sure there must be another way of doing this, I would do something like this,”

The question was: There is a Unique Key and this field “alow null”, but when i am going to create a Unique Key the SQLSERVER saw that there were values duplicated and the values are “nulls”.

What I understood from this question was, the column we are talking about itself is unique except the fact that it has more than one null values. Because of this we cannot create a unique constraint on that column and I believe there is no way to create a unique constraint on a column which has more than one null unless you make a composite key, to do this I suggested two ways,

1. Combine this with any other unique column. Here when I say these, I am talking for only null records, not for non null records, that wasn;t even discussed in the question. The only concern of that person was how to solve null values and still make this column unique. no where he said that he want to make this column who unique…. his concern was only for null records.

2. I suggested that he create an identity column, in case if he dont have a uniwue column. now create a composite unique constraint on previous column and newly created identity column, again the concern here was only for null values.

I understand your concern, you are asking if I make a unique constraint, on cola and colb, colb is identity, then I can have duplicate records in cola, but the combination of cola and colb will still be unique. You are correct, but again like I said, I answered this question thinking that the problem is only with null records and column by default have only unique non null records.

I am always ready to correct myself. Thanks for bringing up this fact.

Thanks,
Imran.

If you truly, desperately must have a column of unique non-null values but still allow many null values, one cheesy way would be to add a computed column, defined to contain the same non-null data, and some other, unique, data when the original column is null. E.g. like this:

ALTER TABLE SomeTable
ADD DummyColumn AS CASE
WHEN UniqueColumn IS NULL THEN SomeReallyUniqueColumnOrCombinationsThereofThatAreGuaranteedNeverToOccurInTheOtherUniqueColumn
ELSE UniqueColumn
END PERSISTED

The problem of course is to find a set of deterministic data that is guaranteed to be unique for each row and at the same time guaranteed not to be present in the original column.

However, I don’t think that this will solve the problem. True, it will allow multiple nulls in ColA (since any combo of null + ColB will be unique). But it will also allow multiple non-unique, non-null values in ColA (since ColB is a unique column itself, all of the rows for ColA could equal the same value while still observing the unique constraint). Am I missing something here?