Easiest way to eliminate NULLs in SELECT DISTINCT? - sql

Question

I am working on a query that is fairly similar the following:
CREATE TABLE #test (a char(1), b char(1))
INSERT INTO #test(a,b) VALUES
('A',NULL),
('A','B'),
('B',NULL),
('B',NULL)
SELECT DISTINCT a,b FROM #test
DROP TABLE #test
The result is, unsurprisingly,
a b
-------
A NULL
A B
B NULL
The output I would like to see in actuality is:
a b
-------
A B
B NULL
That is, if a column has a value in some records but not in others, I want to throw out the row with NULL for that column. However, if a column has a NULL value for all records, I want to preserve that NULL.
What's the simplest/most elegant way to do this in a single query?
I have a feeling that this would be simple if I weren't exhausted on a Friday afternoon.

Answer 1

Try this:
select distinct * from test
where b is not null or a in (
select a from test
group by a
having max(b) is null)
You can get the fiddle here.
Note if you can only have one non-null value in b, this can be simplified to:
select a, max(b) from test
group by a

Answer 2

Try this:
create table test(
x char(1),
y char(1)
);
insert into test(x,y) values
('a',null),
('a','b'),
('b', null),
('b', null)
Query:
with has_all_y_null as
(
select x
from test
group by x
having sum(case when y is null then 1 end) = count(x)
)
select distinct x,y from test
where
(
-- if a column has a value in some records but not in others,
x not in (select x from has_all_y_null)
-- I want to throw out the row with NULL
and y is not null
)
or
-- However, if a column has a NULL value for all records,
-- I want to preserve that NULL
(x in (select x from has_all_y_null))
order by x,y
Output:
X Y
A B
B NULL
Live test: http://sqlfiddle.com/#!3/259d6/16
EDIT
Seeing Mosty's answer, I simplified my code:
with has_all_y_null as
(
select x
from test
group by x
-- having sum(case when y is null then 1 end) = count(x)
-- should have thought of this instead of the code above. Mosty's logic is good:
having max(y) is null
)
select distinct x,y from test
where
y is not null
or
(x in (select x from has_all_y_null))
order by x,y
I just prefer CTE approach, it has a more self-documenting logic :-)
You can also put documentation on non-CTE approach, if you are conscious of doing so:
select distinct * from test
where b is not null or a in
( -- has all b null
select a from test
group by a
having max(b) is null)

Answer 3

;WITH CTE
AS
(
SELECT DISTINCT * FROM #test
)
SELECT a,b
FROM CTE
ORDER BY CASE WHEN b IS NULL THEN 9999 ELSE b END ;

Answer 4

SELECT DISTINCT t.a, t.b
FROM #test t
WHERE b IS NOT NULL
OR NOT EXISTS (SELECT 1 FROM #test u WHERE t.a = u.a AND u.b IS NOT NULL)
ORDER BY t.a, t.b

Answer 5

This is a really weird requirement. I wonder how you need it.
SELECT DISTINCT a, b
FROM test t
WHERE NOT ( b IS NULL
AND EXISTS
( SELECT *
FROM test ta
WHERE ta.a = t.a
AND ta.b IS NOT NULL
)
)
AND NOT ( a IS NULL
AND EXISTS
( SELECT *
FROM test tb
WHERE tb.b = t.b
AND tb.a IS NOT NULL
)
)

Answer 6

Well, I don't particularly like this solution, but it seems the most appropriate to me. Note that your description of what you want sounds exactly like what you get with a LEFT JOIN, so:
SELECT DISTINCT a.a, b.b
FROM #test a
LEFT JOIN #test b ON a.a = b.a
AND b.b IS NOT NULL

Answer 7

SELECT a,b FROM #test t where b is not null
union
SELECT a,b FROM #test t where b is null
and not exists(select 1 from #test where a=t.a and b is not null)
Result:
a b
---- ----
A B
B NULL

Answer 8

I'll just put here a mix of two answers that solved my issue, because my View was more complex
--IdCompe int,
--Nome varchar(30),
--IdVanBanco int,
--IdVan int
--FlagAtivo bit,
--FlagPrincipal bit
select IdCompe
, Nome
, max(IdVanBanco)
, max(IdVan)
, CAST(MAX(CAST(FlagAtivo as INT)) AS BIT) FlagAtivo
, CAST(MAX(CAST(FlagPrincipal as INT)) AS BIT) FlagPrincipal
from VwVanBanco
where IdVan = {IdVan} or IdVan is null
group by IdCompe, Nome order by IdCompe asc
Thanks to mosty mostacho and
kenwarner