Question

Using SQL Server Management Studio, I am getting some undesired results (looks like a bug to me..?)

If I use (FIELD rather than field for the other_table):

SELECT * FROM main_table WHERE field IN (SELECT FIELD FROM other_table)

I get all results from main_table.

Using the correct case:

SELECT * FROM main_table WHERE field IN (SELECT field FROM other_table)

I get the expected results where field appears in other.

Running the subquery on it’s own:

SELECT FIELD FROM other_table

I get an invalid column name error.

Surely I should get this error in the first case?

Is this related to collation?
The DB is binary collation.
The server is case insensitive however.
It seems to me like the server component is saying “this code is OK” and not allowing the DB to say the field is the wrong name..?

What are my options for a solution?

Answer 1

Let’s illustrate what is happening using something that doesn’t depend on case sensitivity:

USE tempdb;
GO

CREATE TABLE dbo.main_table(column1 INT);

CREATE TABLE dbo.other_table(column2 INT);

INSERT dbo.main_table SELECT 1 UNION ALL SELECT 2;
INSERT dbo.other_table SELECT 1 UNION ALL SELECT 3;

SELECT column1 FROM dbo.main_table
WHERE column1 IN (SELECT column1 FROM dbo.other_table);

Results:

column1
-------
1
2

Why doesn’t that raise an error? SQL Server is looking at your queries and seeing that the column1 inside can’t possibly be in other_table, so it is extrapolating and "using" the column1 that exists in the outer referenced table (just like you could reference a column that only exists in the outer table without a table reference). Think about this variation:

SELECT [column1] FROM dbo.main_table
WHERE EXISTS (SELECT [column1] FROM dbo.other_table WHERE [column2] = [column1]);

Results:

column1
-------
1

Again SQL Server knows that column1 in the where clause also doesn’t exist in the locally referenced table, but it tries to find it in the outer scope. So in an imaginary world you might consider the query to actually be saying:

SELECT m.[column1] FROM dbo.main_table AS m
WHERE EXISTS (SELECT m.[column1] FROM dbo.other_table AS o WHERE o.[column2] = m.[column1]);

(Which is not how I typed it, but if I do type it that way, it still works.)

It doesn’t make logical sense in some of the cases but this is the way the query engine does it and the rule has to be applied consistently. In your case (no pun intended), you have an extra complication: case sensitivity. SQL Server didn’t find FIELD in your subquery, but it did find it in the outer query. So a couple of lessons:

1. Always prefix your column references with the table name or alias (and always prefix your table references with the schema).
2. Always create and reference your tables, columns and other entities using consistent case. Especially when using a binary or case-sensitive collation.