fix: check column references in ORDER BY (#1411)

[akutschera]

MySQL and Sqlite store the information for the order by columns in a WindowClause while PostgreSQL stores it in a SortClause.
That's why I added two new if-statements.

Refs: #1411

[kyleconroy]

Adding additional validation like this is great, but it will cause queries that used to work to fail. If the validation code is wrong, users won't be able to get their queries working.

Let's add a new configuration option here called validate_order_by that defaults to true (make it a pointer to a bool). Then we can update the error message telling users they can turn it off by setting validate_order_by: false.

[kyleconroy]

I take a bit of a different approach to testing sqlc, opting mainly for end-to-end tests that don't verify internal behavior. Can you take these tests and move them into internal/endtoend/testdata/order_by_non_existing_column?

[akutschera]

That I can do. We actually only need these tests for order by. The group by tests are in internal/endtoend/testdata/invalid_group_by_reference already

[akutschera]

Adding additional validation like this is great, but it will cause queries that used to work to fail. If the validation code is wrong, users won't be able to get their queries working.

Let's add a new configuration option here called validate_order_by that defaults to true (make it a pointer to a bool). Then we can update the error message telling users they can turn it off by setting validate_order_by: false.

If we default it to "true" and we break existing queries, we need to either give people the possibility to also configure this in v1-files or we force them to upgrade their config to v2. I think both are valid options.
We could also default it to false (similar to "strict_function_checks"). This way we won't break existing uses and everyone who wants to use this feature needs to upgrade to v2.

[kyleconroy]

sqlc getting more strict over time is a good thing. strict_function_checks works in V1 as well, so we can put validate_order_by in the same spot.

[akutschera]

I added the validate_order_by config option. It works even though it feels a bit awkward to pass this value through multiple function calls. I am open to suggestions to improve this before it gets merged.
I will do the documentation in a separate PR because I also want to add documentation for strict_function_checks in V1 and I don't think this belongs to this PR.

[Stumble]

Hey @akutschera ,

There might be a minor bug in checking ambiguous references when joning tables while using alias, see #2398