Sort results by number of NOT NULL values

Question

I have a query where each row consists of 3 columns:

1. Name
2. Distance
3. Proximity

I want to sort the rows based on number of NOT NULL (ie present) values exactly as follows:

1. All values are present
2. Two values are present in this order
   • Name and Distance
   • Name and Proximity
   • Distance and Proximity
3. One value is present
   • Name
   • Distance
   • Proximity

Here is sample data (insert statements are sorted in the order i expect):

/*
CREATE TABLE #TEMP (
    Type      VARCHAR(100),
    Name      VARCHAR(100),
    Distance  VARCHAR(100),
    Proximity VARCHAR(100)
);
*/
INSERT INTO #TEMP VALUES ('AIRPORT', 'KBLI', '21mi', 'City')
INSERT INTO #TEMP VALUES ('AIRPORT', 'KBLI', '21mi',  NULL )
INSERT INTO #TEMP VALUES ('AIRPORT', 'KBLI',  NULL , 'City')
INSERT INTO #TEMP VALUES ('AIRPORT',  NULL , '21mi', 'City')
INSERT INTO #TEMP VALUES ('AIRPORT', 'KBLI',  NULL ,  NULL )
INSERT INTO #TEMP VALUES ('AIRPORT',  NULL , '21mi',  NULL )
INSERT INTO #TEMP VALUES ('AIRPORT',  NULL ,  NULL , 'City')

I have had some success with COALESCE statement but I am looking for something efficient and readable. Later I will change to four columns.

Answer 1

Assign a present value as if it were a number (a name=4, a distance=3, a proximity=2), then sum them and sort by that:

select ...
from ...
order by
  case when name is null then 0 else 4 end +
  case when distance is null then 0 else 3 end +
  case when proximity is null then 0 else 2 end desc

The trick here is that 3+2 > 4, so a distance and proximity beats a name.

Answer 2

This doesn't have any cool fancy math, but if you were to have multiple values for the same [type] then mine would sort those in order as well.

SELECT  *
FROM #Temp
ORDER BY    [Type],

            LEN(CONCAT(LEFT(Name,1),LEFT(Distance,1),LEFT(Proximity,1))) DESC, --counts number of non null columns                       
            --LEN(ISNULL(LEFT(Name,1),'') + ISNULL(LEFT(Distance,1),'') + ISNULL(LEFT(Proximity,1),'')) DESC, /*SQL 2008R2 and below alternative for counting non-null columns*/ 

            ISNULL(Name,'zz'), --ISNULL then 'zz' which when ordered, goes at the end
            ISNULL(Distance,'zz'),
            ISNULL(Proximity,'zz')

Answer 3

For the general question, you can use:

order by ((case when name is not null then 1 else 0 end) + 
          (case when distance is not null then 1 else 0 end) + 
          (case when proximity is not null then 1 else 0 end)
         ) desc

EDIT:

I didn't realize you wanted sub-ordering as well. You might have to list out all the possibilities:

order by (case when name is not null and proximity is not null and distance is not null then 1
               when name is not null and distance is not null then 2
               when name is not null and proximity is not null then 3
               when proximity is not null and distance is not null then 4
               when name is not null then 5
               when distance is not null then 6
               when proximity is not null then 7
               else 8
          end)

Answer 4

This should do it:

SELECT t.*
FROM dbo.Tablename t
ORDER BY 
    CASE WHEN Name       IS NOT NULL 
          AND Distance   IS NOT NULL 
          AND Proximity  IS NOT NULL THEN 0 ELSE 1 END ASC,
     CASE WHEN Name      IS NOT NULL 
          AND Distance   IS NOT NULL THEN 0 ELSE 1 END ASC,
     CASE WHEN Name      IS NOT NULL 
          AND Proximity  IS NOT NULL THEN 0 ELSE 1 END ASC,
     CASE WHEN Distance  IS NOT NULL 
          AND Proximity  IS NOT NULL THEN 0 ELSE 1 END ASC,
     CASE WHEN Name      IS NOT NULL THEN 0 ELSE 1 END ASC,
     CASE WHEN Distance  IS NOT NULL THEN 0 ELSE 1 END ASC,
     CASE WHEN Proximity IS NOT NULL THEN 0 ELSE 1 END ASC

Demo with your sample data.

So the most important order criteria is the first and the least important the last.