Fix aggregate queries with case expressions

[dauinsight]

Addresses issue: #316

Params in the SELECT clause are interpreted differently than the params in the GROUP BY clause.

Example:
The following query fails

SELECT count(1) , case when name = ? then 1 else 2 end from TABLE group by case when name = ? then 1 else 2 end;

Results:

[42000] [Microsoft][ODBC Driver 17 for SQL Server][SQL Server]Column 'name' is invalid in the select list because it is not contained in either an aggregate function or the GROUP BY clause. (8120) (SQLExecDirectW)"

However, grouping these params avoids the error:

DECLARE @var1 NVARCHAR(64);
SET @var1 = ?
SELECT count(1) , case when name = @var1 then 1 else 2 end 
FROM TABLE 
GROUP BY case when name = @var1 then 1 else 2 end;

[federicoemartinez]

I would add a test like this one:

    def test_group_by_case(self):
        annotated_queryset = Book.objects.annotate(age=Case(
            When(id__gt=1000, then=Value("new")),
            default=Value("old"),
            output_field=CharField())).values('age').annotate(sum=Sum('id'))
        self.assertEqual(list(annotated_queryset.all()), [])

[federicoemartinez]

should we do the same for executemany ?

[dauinsight]

Right, we'll have to update every set of params

[dauinsight]

It'll be difficult to support this logic for multiple queries because we group params if they are the same (and different queries have different params).

In the future we could tighten the constraints for grouping params so that it only applies to case/when expressions.

[federicoemartinez]

if value is too long, we should use nvarchar(max), shouldn't we?

[federicoemartinez]

I think this will generate errors.
For example, if the column where I use this value is of type datetime, it will fail as seen in #340.
Maybe we have to replace None with nulls in our query?

[federicoemartinez]

I just tried with sql_variant and it fails. I didn't try it with a group by query, but I think we can get the same result

1> declare @test sql_variant;
2> set @test = null
3> update TABLE_NAMe set datetime_column = @test;
4> go
Msg 257, Level 16, State 3, Server sqlhost, Line 3
Implicit conversion from data type sql_variant to datetimeoffset is not allowed. Use the CONVERT function to run this query.

[federicoemartinez]

I have not been able to generate this with the ORM, but using raw i was able to make it fail:

query = Comment.objects.raw("SELECT * from testapp_comment where created_at = %s GROUP BY id", params=(None,))
print(list(query))

The generated query is:

DECLARE @var0 INT = ?  
SELECT * from testapp_comment where created_at = @var0 GROUP BY id

And it fails:
pyodbc.DataError: ('22018', '[22018] [Microsoft][ODBC Driver 17 for SQL Server][SQL Server]Operand type clash: datetime2 is incompatible with int (206) (SQLExecDirectW)')

It is a silly example, but I wonder what else we can break

[federicoemartinez]

I think it is ok now

[dauinsight]

I think it is ok now

Thanks @federicoemartinez , we will have an additional team member review before merging.