SQL Server 2005 - writing an insert and update trigger for validation - sql

Question

I'm pretty bad at SQL, so I need someone to check my trigger query and tell me if it solves the problem and how acceptable it is. The requirements are a bit convoluted, so please bear with me.
Suppose I have a table declared like this:
CREATE TABLE Documents
(
id int identity primary key,
number1 nvarchar(32),
date1 datetime,
number2 nvarchar(32),
date2 datetime
);
For this table, the following constraints must be observed:
At least one of the number-date pairs should be filled (both the number and the date field not null).
If both number1 and date1 are not null, a record is uniquely identified by this pair. There cannot be two records with the same number1 and date1 if both fields are not null.
If either number1 or date1 is null, a record is uniquely identified by the number2-date2 pair.
Yes, there is a problem of poor normalization, but I cannot do anything about that.
As far as I know, I cannot write unique indexes on the number-date pairs that check whether some of the values are null in SQL Server 2005. Thus, I tried validating the constraints with a trigger.
One last requirement - the trigger should have no inserts of its own, only validation checks. Here's what I came up with:
CREATE TRIGGER validate_requisite_uniqueness
ON [Documents]
FOR INSERT, UPDATE
AS
BEGIN
DECLARE #NUMBER1 NVARCHAR (32)
DECLARE #DATE1 DATETIME
DECLARE #NUMBER2 NVARCHAR (32)
DECLARE #DATE2 DATETIME
DECLARE #DATETEXT VARCHAR(10)
DECLARE inserted_cursor CURSOR FAST_FORWARD FOR SELECT number1, date1, number2, date2 FROM Inserted
IF NOT EXISTS (SELECT * FROM INSERTED)
RETURN
OPEN inserted_cursor
FETCH NEXT FROM inserted_cursor into #NUMBER1, #DATE1, #NUMBER2, #DATE2
WHILE ##FETCH_STATUS = 0
BEGIN
IF (#NUMBER1 IS NULL OR #DATE1 IS NULL)
BEGIN
IF (#NUMBER2 IS NULL OR #DATE2 IS NULL)
BEGIN
ROLLBACK TRANSACTION
RAISERROR ('Either the first or the second number-date pair should be filled.', 10, 1)## Heading ##
END
END
IF (#NUMBER1 IS NOT NULL AND #DATE1 IS NOT NULL)
BEGIN
IF ((SELECT COUNT(*) FROM Documents WHERE number1 = #NUMBER1 AND date1 = #DATE1) > 1)
BEGIN
ROLLBACK TRANSACTION
SET #DATETEXT = CONVERT(VARCHAR(10), #DATE1, 104)
RAISERROR ('A document with the number1 ''%s'' and date1 ''%s'' already exists.', 10, 1, #NUMBER1, #DATETEXT)
END
END
ELSE IF (#NUMBER2 IS NOT NULL AND #DATE2 IS NOT NULL) /*the DATE2 check is redundant*/
BEGIN
IF ((SELECT COUNT(*) FROM Documents WHERE number2 = #NUMBER2 AND date2 = #DATE2) > 1)
BEGIN
ROLLBACK TRANSACTION
SET #DATETEXT = CONVERT(VARCHAR(10), #DATE2, 104)
RAISERROR ('A document with the number2 ''%s'' and date2 ''%s'' already exists.', 10, 1, #NUMBER2, #DATETEXT)
END
END
FETCH NEXT FROM inserted_cursor
END
CLOSE inserted_cursor
DEALLOCATE inserted_cursor
END
Please tell me how well-written and efficient this solution is.
A couple of questions I can come up with:
Will this trigger validate correctly against existing rows and newly inserted/updated rows in case of bulk modifications? It should, because the modifications are already applied to the table in the scope of this transaction, right?
Are the constraint violations handled correctly? Meaning, was I right to use the rollback transaction and raiserror pair?
Is the "IF NOT EXISTS (SELECT * FROM INSERTED) RETURN" statement used correctly?
Is the use of COUNT to check the constraints acceptable, or should I use some other way of checking the uniqueness of number-date pairs?
Can this solution be optimized in terms of execution speed? Should I add non-unique indexes on both number-date pairs?
Thanks.
EDIT:
A solution using a check constraint and an indexed view, based on Damien_The_Unbeliever's answer:
CREATE TABLE dbo.Documents
(
id int identity primary key,
number1 nvarchar(32),
date1 datetime,
number2 nvarchar(32),
date2 datetime,
constraint CK_Documents_AtLestOneNotNull CHECK (
(number1 is not null and date1 is not null) or
(number2 is not null and date2 is not null)
)
);
go
create view dbo.UniqueDocuments
with schemabinding
as
select
CASE WHEN (number1 is not null and date1 is not null)
THEN CAST(1 AS BIT)
ELSE CAST(0 AS BIT)
END as first_pair_filled,
CASE WHEN (number1 is not null and date1 is not null)
THEN number1
ELSE number2
END as number,
CASE WHEN (number1 is not null and date1 is not null)
THEN date1
ELSE date2
END as [date]
from
dbo.Documents
go
create unique clustered index IX_UniqueDocuments on dbo.UniqueDocuments(first_pair_filled,number,[date])
go

Answer 1

I would avoid the trigger, and use a check constraint and an indexed view:
CREATE TABLE dbo.Documents
(
id int identity primary key,
number1 nvarchar(32),
date1 datetime,
number2 nvarchar(32),
date2 datetime,
constraint CK_Documents_AtLestOneNotNull CHECK (
(number1 is not null and date1 is not null) or
(number2 is not null and date2 is not null)
)
);
go
create view dbo.UniqueDocuments
with schemabinding
as
select
COALESCE(number1,number2) as number,
COALESCE(date1,date2) as [date]
from
dbo.Documents
go
create unique clustered index IX_UniqueDocuments on dbo.UniqueDocuments(number,[date])
go
Which has the advantage that, although there is some "trigger-like" behaviour because of the indexed view, it's well-tested code that's already been deeply integrated into SQL Server.

Answer 2

I would use this logic instead (I didn't type it all as it takes ages), and definitely use SELECT 1 FROM ... in the IF EXISTS() statement as it helps performance. Also remove the cursors like marc_s said.
CREATE TRIGGER trg_validate_requisite_uniqueness
ON dbo.[Documents]
AFTER INSERT, UPDATE
AS
DECLARE #Number1 NVARCHAR(100) = (SELECT TOP 1 number1 FROM dbo.Documents ORDER BY Id DESC)
DECLARE #Date1 DATETIME = (SELECT TOP 1 date1 FROM dbo.Documents ORDER BY Id DESC)
DECLARE #Number2 NVARCHAR(100) = (SELECT TOP 1 number2 FROM dbo.Documents ORDER BY Id DESC)
DECLARE #Date2 DATETIME = (SELECT TOP 1 date2 FROM dbo.Documents ORDER BY Id DESC)
DECLARE #DateText NVARCHAR(100)
IF EXISTS (SELECT 1 FROM dbo.Documents AS D
INNER JOIN INSERTED AS I ON D.id = I.id WHERE I.Number1 IS NULL AND I.number2 IS NULL)
BEGIN
ROLLBACK TRANSACTION
RAISERROR ('Either the first or the second number pair should be filled.', 10, 1)
END
ELSE IF EXISTS (SELECT 1 FROM dbo.Documents AS D
INNER JOIN INSERTED AS I ON D.id = I.id WHERE I.Date1 IS NULL AND I.Date2 IS NULL)
BEGIN
ROLLBACK TRANSACTION
RAISERROR ('Either the first or the second date pair should be filled.', 10, 1)
END
ELSE IF EXISTS (SELECT 1 FROM dbo.Documents AS D
GROUP BY D.number1, D.date1 HAVING COUNT(*) >1
)
BEGIN
ROLLBACK TRANSACTION
SET #DateText = (SELECT CONVERT(VARCHAR(10), #Date1, 104))
RAISERROR ('Cannot have duplicate values', 10, 1, #Number1, #DateText )
END