SQL Server Update Trigger, Get Only modified fields

Inside the trigger, you can use COLUMNS_UPDATED() like this in order to get updated value

-- Get the table id of the trigger
--
DECLARE @idTable      INT

SELECT  @idTable = T.id 
FROM    sysobjects P JOIN sysobjects T ON P.parent_obj = T.id 
WHERE   P.id = @@procid

-- Get COLUMNS_UPDATED if update
--
DECLARE @Columns_Updated VARCHAR(50)

SELECT  @Columns_Updated = ISNULL(@Columns_Updated + ', ', '') + name 
FROM    syscolumns 
WHERE   id = @idTable   
AND     CONVERT(VARBINARY,REVERSE(COLUMNS_UPDATED())) & POWER(CONVERT(BIGINT, 2), colorder - 1) > 0

But this snipet of code fails when you have a table with more than 62 columns.. Arth.Overflow...

Here is the final version which handles more than 62 columns but give only the number of the updated columns. It's easy to link with 'syscolumns' to get the name

DECLARE @Columns_Updated VARCHAR(100)
SET     @Columns_Updated = ''   

DECLARE @maxByteCU INT
DECLARE @curByteCU INT
SELECT  @maxByteCU = DATALENGTH(COLUMNS_UPDATED()), 
        @curByteCU = 1

WHILE @curByteCU <= @maxByteCU BEGIN
    DECLARE @cByte INT
    SET     @cByte = SUBSTRING(COLUMNS_UPDATED(), @curByteCU, 1)

    DECLARE @curBit INT
    DECLARE @maxBit INT
    SELECT  @curBit = 1, 
            @maxBit = 8
    WHILE @curBit <= @maxBit BEGIN
        IF CONVERT(BIT, @cByte & POWER(2,@curBit - 1)) <> 0 
            SET @Columns_Updated = @Columns_Updated + '[' + CONVERT(VARCHAR, 8 * (@curByteCU - 1) + @curBit) + ']'
        SET @curBit = @curBit + 1
    END
    SET @curByteCU = @curByteCU + 1
END

I've another completely different solution that doesn't use COLUMNS_UPDATED at all, nor does it rely on building dynamic SQL at runtime. (You might want to use dynamic SQL at design time but thats another story.)

Basically you start with the inserted and deleted tables, unpivot each of them so you are just left with the unique key, field value and field name columns for each. Then you join the two and filter for anything that's changed.

Here is a full working example, including some test calls to show what is logged.

-- -------------------- Setup tables and some initial data --------------------
CREATE TABLE dbo.Sample_Table (ContactID int, Forename varchar(100), Surname varchar(100), Extn varchar(16), Email varchar(100), Age int );
INSERT INTO Sample_Table VALUES (1,'Bob','Smith','2295','[email protected]',24);
INSERT INTO Sample_Table VALUES (2,'Alice','Brown','2255','[email protected]',32);
INSERT INTO Sample_Table VALUES (3,'Reg','Jones','2280','[email protected]',19);
INSERT INTO Sample_Table VALUES (4,'Mary','Doe','2216','[email protected]',28);
INSERT INTO Sample_Table VALUES (5,'Peter','Nash','2214','[email protected]',25);

CREATE TABLE dbo.Sample_Table_Changes (ContactID int, FieldName sysname, FieldValueWas sql_variant, FieldValueIs sql_variant, modified datetime default (GETDATE()));

GO

-- -------------------- Create trigger --------------------
CREATE TRIGGER TriggerName ON dbo.Sample_Table FOR DELETE, INSERT, UPDATE AS
BEGIN
    SET NOCOUNT ON;
    --Unpivot deleted
    WITH deleted_unpvt AS (
        SELECT ContactID, FieldName, FieldValue
        FROM 
           (SELECT ContactID
                , cast(Forename as sql_variant) Forename
                , cast(Surname as sql_variant) Surname
                , cast(Extn as sql_variant) Extn
                , cast(Email as sql_variant) Email
                , cast(Age as sql_variant) Age
           FROM deleted) p
        UNPIVOT
           (FieldValue FOR FieldName IN 
              (Forename, Surname, Extn, Email, Age)
        ) AS deleted_unpvt
    ),
    --Unpivot inserted
    inserted_unpvt AS (
        SELECT ContactID, FieldName, FieldValue
        FROM 
           (SELECT ContactID
                , cast(Forename as sql_variant) Forename
                , cast(Surname as sql_variant) Surname
                , cast(Extn as sql_variant) Extn
                , cast(Email as sql_variant) Email
                , cast(Age as sql_variant) Age
           FROM inserted) p
        UNPIVOT
           (FieldValue FOR FieldName IN 
              (Forename, Surname, Extn, Email, Age)
        ) AS inserted_unpvt
    )

    --Join them together and show what's changed
    INSERT INTO Sample_Table_Changes (ContactID, FieldName, FieldValueWas, FieldValueIs)
    SELECT Coalesce (D.ContactID, I.ContactID) ContactID
        , Coalesce (D.FieldName, I.FieldName) FieldName
        , D.FieldValue as FieldValueWas
        , I.FieldValue AS FieldValueIs 
    FROM 
        deleted_unpvt d

            FULL OUTER JOIN 
        inserted_unpvt i
            on      D.ContactID = I.ContactID 
                AND D.FieldName = I.FieldName
    WHERE
         D.FieldValue <> I.FieldValue --Changes
        OR (D.FieldValue IS NOT NULL AND I.FieldValue IS NULL) -- Deletions
        OR (D.FieldValue IS NULL AND I.FieldValue IS NOT NULL) -- Insertions
END
GO
-- -------------------- Try some changes --------------------
UPDATE Sample_Table SET age = age+1;
UPDATE Sample_Table SET Extn = '5'+Extn where Extn Like '221_';

DELETE FROM Sample_Table WHERE ContactID = 3;

INSERT INTO Sample_Table VALUES (6,'Stephen','Turner','2299','[email protected]',25);

UPDATE Sample_Table SET ContactID = 7 where ContactID = 4; --this will be shown as a delete and an insert
-- -------------------- See the results --------------------
SELECT *, SQL_VARIANT_PROPERTY(FieldValueWas, 'BaseType') FieldBaseType, SQL_VARIANT_PROPERTY(FieldValueWas, 'MaxLength') FieldMaxLength from Sample_Table_Changes;

-- -------------------- Cleanup --------------------
DROP TABLE dbo.Sample_Table; DROP TABLE dbo.Sample_Table_Changes;

So no messing around with bigint bitfields and arth overflow problems. If you know the columns you want to compare at design time then you don't need any dynamic SQL.

On the downside the output is in a different format and all the field values are converted to sql_variant, the first could be fixed by pivoting the output again, and the second could be fixed by recasting back to the required types based on your knowledge of the design of the table, but both of these would require some complex dynamic sql. Both of these might not be an issue in your XML output. This question does something similar to getting the output back in the same format.

Edit: Reviewing the comments below, if you have a natural primary key that could change then you can still use this method. You just need to add a column that is populated by default with a GUID using the NEWID() function. You then use this column in place of the primary key.

You may want to add an index to this field, but as the deleted and inserted tables in a trigger are in memory it might not get used and may have a negative effect on performance.