compare columns only when NOT NULL

Question

This should be easy, but I'm just missing something. I have the following:

IF OBJECT_ID('LAST_NM') IS NOT NULL
    DROP TABLE LAST_NM

CREATE TABLE LAST_NM (
    ID int NOT NULL IDENTITY(1,1),
    LAST_NM_ORIGINAL varchar(255) NOT NULL,
    LAST_NM_1 varchar(255)NULL,
    LAST_NM_2 varchar(255)NULL,
    LAST_NM_3 varchar(255)NULL,
    LAST_NM_4 varchar(255)NULL,
    PRIMARY KEY (ID)
);

INSERT INTO LAST_NM
(LAST_NM_ORIGINAL, LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)
VALUES
('SMITH', 'HARRIS', NULL, 'HARRIS', NULL),
('JONES', 'FUTURE', 'FUTURE', 'FUTURE', 'FUTURE'),
('SMITH', 'ALPHA', 'ALPHA', 'ALPHA', NULL),
('SMITH', 'BETA', 'BETA', 'GEORGE', NULL),
('SMITH', 'SMITH', NULL, 'SMITH', NULL),
('DOPE', NULL, NULL, NULL, 'CURLS')

What I want to do is SELECT from this table where:

• the last_nm_# IS NOT NULL
• the NOT NULL last_nm_# all have the same value
• those last_nm_# are not the same as the LAST_NM_ORIGINAL

I've tried playing with CASE and SWITCH and I got a messy version working if I hard code the heck out of statements and union them like this:

SELECT * FROM (
    SELECT ID, LAST_NM_ORIGINAL, LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4
    FROM LAST_NM
    WHERE       LAST_NM_1 IS NOT NULL
            AND LAST_NM_2 IS NOT NULL
            AND LAST_NM_3 IS NOT NULL
            AND LAST_NM_4 IS NOT NULL
            AND LAST_NM_1 = LAST_NM_2
            AND LAST_NM_1 = LAST_NM_3
            AND LAST_NM_3 = LAST_NM_4
            AND LAST_NM_1 <> LAST_NM_ORIGINAL
    UNION
    SELECT ID, LAST_NM_ORIGINAL, LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4
    FROM LAST_NM
    WHERE       LAST_NM_1 IS NOT NULL
            AND LAST_NM_2 IS NOT NULL
            AND LAST_NM_3 IS NOT NULL
            AND LAST_NM_4 IS NULL
            AND LAST_NM_1 = LAST_NM_2
            AND LAST_NM_1 = LAST_NM_3
            AND LAST_NM_1 <> LAST_NM_ORIGINAL
    /*
    WRITE OUT EACH POSSIBLE WAY AND UNION ALL OF THEM
    .
    .
    .
    */
    UNION
    SELECT ID, LAST_NM_ORIGINAL, LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4
    FROM LAST_NM
    WHERE       LAST_NM_1 IS NULL
            AND LAST_NM_2 IS NULL
            AND LAST_NM_3 IS NULL
            AND LAST_NM_4 IS NOT NULL
            AND LAST_NM_4 <> LAST_NM_ORIGINAL
    ) AS RESULT_SET

To summarize, I want to select out the rows if the LAST_NM_# is not NULL and is the same as all the other NOT NULL LAST_NM_# and is different than the LAST_NM_ORIGINAL

So in my example, I should get back rows 1, 2, 3, and 6. But not rows 4 (the 'new' names don't agree) or row 5 (the new names are the same as the old one).

There has to be a better way than just writing each of those out and UNIONING them.. right?

Answer 1

Here it is using UNPIVOT

;WITH NAMES
AS (
    SELECT DISTINCT ID
        ,LAST_NM_ORIGINAL
        ,LAST_NM_NEW
    FROM (
        SELECT ID
            ,LAST_NM_ORIGINAL
            ,LAST_NM_1
            ,LAST_NM_2
            ,LAST_NM_3
            ,LAST_NM_4
        FROM LAST_NM
        ) AS X
    UNPIVOT(LAST_NM_NEW FOR LAST_NM_NEWS IN (
                LAST_NM_1
                ,LAST_NM_2
                ,LAST_NM_3
                ,LAST_NM_4
                )) AS Y
    )
SELECT ID
    ,LAST_NM_ORIGINAL
    ,LAST_NM_NEW
FROM NAMES
WHERE ID IN (
        SELECT ID
        FROM NAMES
        GROUP BY ID
        HAVING COUNT(ID) = 1
        )
    AND LAST_NM_ORIGINAL <> LAST_NM_NEW

Answer 2

This question immediately screams UNPIVOT to me.

You can see the link or just google for syntax and examples, and then use it to get a derived table that looks like this:

ID NM_Orig   NM_Number  NM_Value
1  Smith     1          Harris
1  Smith     2          NULL
1  Smith     3          Harris
1  Smith     4          NULL
2  Jones     1          Future
etc...

From that derived table you would query to get IDs WHERE NM_Value is NOT NULL AND NM_Value <> NM_Orig AND WHERE NOT EXISTS a correlated ROW With a NON-NULL NM_Value that is different from the correlated NM_Value.

Answer 3

Here's another way to do it.

SELECT
    *
FROM
    LAST_NM
WHERE
    (LAST_NM_1 IS NULL OR LAST_NM_1 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_2 IS NULL OR LAST_NM_2 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_3 IS NULL OR LAST_NM_3 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_4 IS NULL OR LAST_NM_4 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_1 IS NULL OR LAST_NM_1 <> LAST_NM_ORIGINAL) AND
    (LAST_NM_2 IS NULL OR LAST_NM_2 <> LAST_NM_ORIGINAL) AND
    (LAST_NM_3 IS NULL OR LAST_NM_3 <> LAST_NM_ORIGINAL) AND
    (LAST_NM_4 IS NULL OR LAST_NM_4 <> LAST_NM_ORIGINAL)

Edit. Can be shortened to the query below since you want at least one of the LAST_NM_# to not be null:

SELECT
    *
FROM
    LAST_NM
WHERE
    (LAST_NM_1 IS NULL OR LAST_NM_1 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_2 IS NULL OR LAST_NM_2 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_3 IS NULL OR LAST_NM_3 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_4 IS NULL OR LAST_NM_4 = COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4)) AND
    (LAST_NM_ORIGINAL <> COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4))

Answer 4

Here is one way to do this.

select ID
    , LAST_NM_ORIGINAL
    , LAST_NM_1
    , LAST_NM_2
    , LAST_NM_3
    , LAST_NM_4
from LAST_NM
where replace(isnull(LAST_NM_1, '') + isnull(LAST_NM_2, '') + isnull(LAST_NM_3, '') + isnull(LAST_NM_4, ''), LAST_NM_ORIGINAL, '') > ''
    AND replace(isnull(LAST_NM_1, '') + isnull(LAST_NM_2, '') + isnull(LAST_NM_3, '') + isnull(LAST_NM_4, ''), coalesce(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4), '') = ''

Answer 5

Your original query does have some superfluous predicates. Any row meeting LAST_NM_1 = LAST_NM_2 is guaranteed to have both LAST_NM_1 NOT NULL and LAST_NM_2 NOT NULL for example.

But it can be made quite concise by using VALUES to get the 4 columns in tabular form as below.

SELECT *
FROM   LAST_NM
WHERE  EXISTS (SELECT *
               FROM   (VALUES(LAST_NM_1),
                             (LAST_NM_2),
                             (LAST_NM_3),
                             (LAST_NM_4)) V(LAST_NM_N)
               HAVING MAX(LAST_NM_N) = MIN(LAST_NM_N) /*exactly one NOT NULL value among the 4 columns*/
                      AND MAX(LAST_NM_N) <> LAST_NM_ORIGINAL) 

Answer 6

Here is another way to do it

SELECT * FROM
#LAST_NM 
WHERE
COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4) = COALESCE(LAST_NM_2, LAST_NM_3, LAST_NM_4,LAST_NM_1)
AND COALESCE(LAST_NM_2, LAST_NM_3, LAST_NM_4,LAST_NM_1) = COALESCE(LAST_NM_3, LAST_NM_4,LAST_NM_1,LAST_NM_2)
AND COALESCE(LAST_NM_3, LAST_NM_4,LAST_NM_1,LAST_NM_2) = COALESCE(LAST_NM_4,LAST_NM_1,LAST_NM_2,LAST_NM_3)
AND (LAST_NM_ORIGINAL <> COALESCE(LAST_NM_1, LAST_NM_2, LAST_NM_3, LAST_NM_4))