I'm trying to use the update function in T-sql in a trigger to check if a column has been changed or not. This is what I'm doing:
declare column_name cursor for select name from sys.columns where object_id = object_id(N'tblKit')
open column_name
fetch column_name into @colname
while (@@Fetch_status=0)
begin
if(Update(@colname))
I get an error saying non-boolean type specified where a condition is expected. This is the syntax used in the msdn forum. So is there anything wrong with this?
I'm using Microsoft SQL server 2008 with Management Studio
If you simply want to log changes of individual columns in your trigger, you could try unpivoting, possibly in conjunction with full join. The idea is, you unpivot both inserted and deleted then join them on the table's key and the column containing the unpivoted names, filtering out the rows where the values are same.
Here's an example illustration of the method.
First, the table definitions:
CREATE TABLE TestTable (
ID int IDENTITY PRIMARY KEY,
Attr1 int,
Attr2 int,
Attr3 int
);
CREATE TABLE TestTableLog (
ID int IDENTITY PRIMARY KEY,
TableID int,
AttrName sysname,
OldValue int,
NewValue int,
Timestamp datetime DEFAULT GETDATE()
);
Next, a trigger for logging changes. This one will catch all the operations: insert, update, and delete:
CREATE TRIGGER trTestTable ON TestTable
AFTER INSERT, UPDATE, DELETE
AS BEGIN
WITH inserted_unpivot AS (
SELECT
ID,
AttrName,
Value
FROM inserted i
UNPIVOT (Value FOR AttrName IN (Attr1, Attr2, Attr3)) u
),
deleted_unpivot AS (
SELECT
ID,
AttrName,
Value
FROM deleted d
UNPIVOT (Value FOR AttrName IN (Attr1, Attr2, Attr3)) u
)
INSERT INTO TestTableLog (TableID, AttrName, OldValue, NewValue)
SELECT
ISNULL(i.ID, d.ID),
ISNULL(i.AttrName, d.AttrName),
d.Value,
i.Value
FROM inserted_unpivot i
FULL JOIN deleted_unpivot d
ON i.ID = d.ID AND i.AttrName = d.AttrName
WHERE CASE i.Value WHEN d.Value THEN 0 ELSE 1 END = 1
END
Now let's fill the TestTable with some data:
WHILE (SELECT COUNT(*) FROM TestTable) < 15
INSERT INTO TestTable
SELECT RAND() * 1000, RAND() * 1000, RAND() * 1000
;
Here's its contents before the subsequent changes:
ID Attr1 Attr2 Attr3
----------- ----------- ----------- -----------
1 820 338 831
2 795 881 453
3 228 430 719
4 36 236 105
5 246 115 649
6 488 657 438
7 990 360 15
8 668 978 724
9 872 385 562
10 460 396 462
11 62 599 630
12 145 815 439
13 595 7 54
14 587 85 655
15 80 606 407
And now let's perform some modifications to the contents:
UPDATE TestTable SET Attr2 = 35 WHERE ID = 3;
UPDATE TestTable SET Attr3 = 0 WHERE ID BETWEEN 6 AND 10;
INSERT INTO TestTable VALUES (1, 1, 1);
DELETE FROM TestTable WHERE ID = 14;
Here's what we've got in TestTable afterwards:
ID Attr1 Attr2 Attr3
----------- ----------- ----------- -----------
1 820 338 831
2 795 881 453
3 228 35 719
4 36 236 105
5 246 115 649
6 488 657 0
7 990 360 0
8 668 978 0
9 872 385 0
10 460 396 0
11 62 599 630
12 145 815 439
13 595 7 54
15 80 606 407
16 1 1 1
And this is what has been logged:
ID TableID AttrName OldValue NewValue Timestamp
----------- ----------- ----------- ----------- ----------- -----------------------
1 3 Attr2 430 35 2011-08-22 20:12:19.217
2 10 Attr3 462 0 2011-08-22 20:12:19.227
3 9 Attr3 562 0 2011-08-22 20:12:19.227
4 8 Attr3 724 0 2011-08-22 20:12:19.227
5 7 Attr3 15 0 2011-08-22 20:12:19.227
6 6 Attr3 438 0 2011-08-22 20:12:19.227
7 16 Attr1 NULL 1 2011-08-22 20:12:19.227
8 16 Attr3 NULL 1 2011-08-22 20:12:19.227
9 16 Attr2 NULL 1 2011-08-22 20:12:19.227
10 14 Attr1 587 NULL 2011-08-22 20:12:19.230
11 14 Attr2 85 NULL 2011-08-22 20:12:19.230
12 14 Attr3 655 NULL 2011-08-22 20:12:19.230
The setup, of course, has been somewhat simplified. In particular, all the main table's columns that are to be logged are of the same type, and so there's no need to convert data to some generic type to embrace various kinds of data. But maybe that's just what you need. And if not, I believe this can provide a good start to implementing the ultimate solution.
