Strange behaviour of SQL Server query - sql

Question

I have a query in a validation stored procedure. It goes something like this:
SELECT *
FROM Table1
WHERE batchID IN (SELECT id FROM #tempIds)
AND CAST(field1 AS DATE) > CAST(field2 AS DATE)
Both field1 and field2 have valid dates, i.e. doing IdDate on field1/2 returns 1.
The #tempIds table only has one column ID and contains only one row.
When I run the above query, I get this error:
Unable to convert varchar to date
But instead of selecting batch ids from temp table if I put hard-coded ID from the same temp table it works.
Any ideas what could be the issue?

Answer 1

The problem is that you are using varchar to store date (or datetime) values.
Choosing the correct data type for your columns would save you from a lot of problems, this one included. For detailed information, read Aaron Bertrand's Bad habits to kick : choosing the wrong data type.
Now, to address our conversation in the comments - SQL Server does not guarantee the order on which the conditions in the where clause are evaluated. This means that even if all the "date" strings in both your columns are convertible to date values for the specific batchID, you only need one wrong value in one of the columns to raise the "Unable to convert varchar to date" error.
This also means that even if you where to write your query the way Larry B suggested in his answer (now deleted - so for the sake of future readers - this was his suggestion:)
SELECT *
FROM Table1
WHERE batchID IN (SELECT id FROM #tempIds)
AND ISDATE(field1) = 1
AND ISDATE(field2) = 1
AND CAST(field1 AS DATE) > CAST(field2 AS DATE)
There is no guarantee that the last condition (cast(field1 as date) > cast(field2 as date)) will be evaluated after the isdate(field1)=1 condition.
The correct thing to do is to fix the problem - change the data types of the columns to the correct data type.
Assuming that can't be done (if you have no control over the structure of the database, for instance) - you can do a couple of things:
Find all the places where the values in field1 and in field2 can't be converted to dates and fix them. Consider adding a check constraint to validate that the values of these columns can be converted to date (assuming you can).
Separate your query into 2 parts:
;With cte as
(
SELECT *
FROM Table1
WHERE batchID IN (SELECT id FROM #tempIds)
AND ISDATE(field1) = 1
AND ISDATE(field2) = 1
)
SELECT *
FROM cte
WHERE CAST(field1 AS DATE) > CAST(field2 AS DATE)
This will eliminate the error, since you are only casting values where the ISDATE function already returned 1, but might not return some rows you want back, if the value if either field1 or field2 is wrong in these rows.