In SQL Server 2008, I'm using MERGE. Everything is fine except that I have 2 nullable columns. If I pass a null value and the target isn't null, MERGE doesn't see a difference (evals against null = false per BOL). If I use IsNull on both sides (source & target) that works, but has the issue of potentially mis-evaluating a value.
What I mean by the last statement is, if I say:
WHEN MATCHED AND NOT (IsNull(tgt.C, 0) = IsNull(src.C, 0)) THEN
then if tgt.C is null and src.C = 0, no update will be performed. No matter what substitute value I choose, I'll have this problem.
I also tried the "AND NOT (...true...)" syntax since BOL states that evaluations against null result in FALSE. However, it seems they actually result in NULL and do not result in my multi-part statement becoming false.
I thought one solution is to use NaN or -INF or +INF since these are not valid in target. But I can't find a way to express this in the SQL.
Any ideas how to solve this?
EDIT:
The following logic solves the problem, but it's verbose and won't make for fast evals:
declare @i int, @j int
set @j = 0
set @i = 0
if ISNULL(@i, 0) != ISNULL(@j, 0) OR
((@i is null or @j is null) and not (@i is null and @j is null))
print 'update';
You can use
WHEN MATCHED AND EXISTS (SELECT tgt.C EXCEPT SELECT src.C)
See this article for more on this issue.
You can change the ON part of the merge statement, putting in a check for when both source and target are null.
MERGE tgt
USING src
ON ( -- enter non-nullable columns to match on ...
tgt.A = src.A
AND (tgt.C = src.C OR (tgt.C IS NULL AND src.C IS NULL))
)
WHEN MATCHED -- ...
Actually, this works better. Just add another substitution value as an OR :-
WHEN MATCHED AND
(
NOT (IsNull(tgt.C, 0) = IsNull(src.C, 0)) OR NOT (IsNull(tgt.C, 1) = IsNull(src.C, 1))
)
THEN ....
Have you tried SET ANSI_NULLS OFF
, which will make NULL=NULL
return true? This may create additional issues but it could be a script-level workaround (turn it off then on once you run your proc).
WHEN MATCHED AND tgt.c <> src.c OR tgt.c IS NULL AND src.c IS NOT NULL OR tgt.c IS NOT NULL AND src.c IS NULL
This works as well and may be better when you have multiple columns that you want to check if they are different.
MERGE @t2 a
using @t1 b
ON a.PK = b.PK
WHEN MATCHED AND CHECKSUM(a.PK,a.VALUE)!= CHECKSUM(b.pk,b.VALUE)
THEN UPDATE SET a.VALUE = b.VALUE;
Instead of using 0 when the values are null, why not use a value that is highly unlikely to exist?
EG (IsNull(tgt.C, 2093128301).
The datatypes are int so you have a lot to play with......
WHEN MATCHED AND
(
NULLIF(tgt.C, src.C) IS NOT NULL OR NULLIF(src.C, tgt.C) IS NOT NULL
)
THEN