T-SQL / Unexpected NULL handling when ANSI_NULLS is disabled
I'm just struggling with handling NULL values in SQL Server (tested on version 12.0.5000.0). Basically, my intention is to get all rows that have a <> column value of a static value (like 999). I am NOT looking for an alternative to LIKE "use ISNULL function". The request is generated by a third party engine and I am not going to write a parser and change the operator.
-- All examples with ANSI_NULLS OFF
SET ANSI_NULLS OFF;
GO
--------------------------------------------------------------------------------------------------------
-- "Virtual" example / working as expected
--------------------------------------------------------------------------------------------------------
DECLARE
@One INT = 1,
@Null INT = NULL
SELECT
IIF(@Null = NULL, 1, 0) '@Null = NULL' -- To test if ANSI_NULL is OFF
,IIF(@One <> NULL, 1, 0) '@One <> NULL' -- working with NULL variable
,IIF(1 <> NULL, 1, 0) '1 <> NULL' -- working with NULL value
--------------------------------------------------------------------------------------------------------
-- MSDN Example / NOT working as expected
-- https://docs.microsoft.com/en-us/sql/t-sql/statements/set-ansi-nulls-transact-sql
--------------------------------------------------------------------------------------------------------
-- Create table t1 and insert values.
CREATE TABLE dbo.t1 (a INT NULL);
INSERT INTO dbo.t1 values (NULL),(0),(1);
GO
-- SET ANSI_NULLS to OFF and test.
DECLARE @varname int;
SET @varname = 999;
SELECT a
FROM t1
WHERE a <> @varname; -- working with NULL variable
SELECT a
FROM t1
WHERE a <> 999; -- NOT working with NULL value
-- Drop table t1.
DROP TABLE dbo.t1;
Can anyone explain why the "virtual" example works differently than the MSDN example?
Virtual example:
+--------------+--------------+-----------+
| @Null = NULL | @One <> NULL | 1 <> NULL |
+--------------+--------------+-----------+
| 1 | 1 | 1 |
+--------------+--------------+-----------+
MSDN example:
-- SELECT 1
+------+
| a |
+------+
| NULL |
| 0 |
| 1 |
+------+
-- SELECT 2
+------+
| a |
+------+
| 0 |
| 1 |
+------+
source to share
It looks like the query optimizer is choosing a different comparison operator:
DECLARE @varname int;
SET @varname = 999;
SELECT a
FROM t1
WHERE a <> @varname;
XML Execution Plan:
<Predicate>
<ScalarOperator ScalarString="[fiddle_84f7799901e54a779e8bff464a2d01f3].[dbo].[t1].[a] <> [@varname]">
<Compare CompareOp="IS NOT">
<ScalarOperator>
<Identifier>
<ColumnReference Database="[fiddle_84f7799901e54a779e8bff464a2d01f3]" Schema="[dbo]" Table="[t1]" Column="a"></ColumnReference>
</Identifier>
</ScalarOperator>
<ScalarOperator>
<Identifier>
<ColumnReference Column="@varname"></ColumnReference>
</Identifier>
</ScalarOperator>
</Compare>
</ScalarOperator>
</Predicate>
Compare CompareOp = "IS NOT"
Second hardcoded request:
SELECT a
FROM t1
WHERE a <> 999;
-- same as
DECLARE @varname int = 999;
SELECT a
FROM t1
WHERE a <> (SELECT @varname);
XML Execution Plan:
<Predicate>
<ScalarOperator ScalarString="[fiddle_ac5121a789da473382366733b51ef441].[dbo].[t1].[a]<>(999)">
<Compare CompareOp="NE">
<ScalarOperator>
<Identifier>
<ColumnReference Database="[fiddle_ac5121a789da473382366733b51ef441]" Schema="[dbo]" Table="[t1]" Column="a"></ColumnReference>
</Identifier>
</ScalarOperator>
<ScalarOperator>
<Const ConstValue="(999)"></Const>
</ScalarOperator>
</Compare>
</ScalarOperator>
</Predicate>
Compare CompareOp = "NE"
EDIT:
SET ANSI_NULLS ON affects the comparison only if one of the comparison operands is either a NULL variable or a NULL literal. If both sides of the comparison are columns or compound expressions, the setting does not affect the comparison.
source to share