bug: typing issue with `filter` method on the table column

[fwojciec]

What happened?

Here is a repo with the reproduction of the issue on the screenshot: https://github.com/fwojciec/ibis-types-test

Legitimate uses of the filter method result in type errors reported by pyright and pylance. This is because the filter method expects the predicates to be ir.BooleanValue | Sequence[ir.BooleanValue] | IfAnyAll and the actual type of the predicates in the above examples is just a plain ol' bool.

Interestingly, the equals method provided by Ibis returns b bool for its result, which seems like an inconsistency.

I appreciate that static typing is not a priority for the project, so I'm happy to make a PR to try and fix any typing issues I come across, but I wanted to get some guidance first on how these types of issues should be addressed - in this case, is the solution basically adding bool as a type that's accepted by the filter method in addition to the internal Ibis types mentioned above? Or maybe this is just something that can be worked around by somehow casting the bool predicate into an instance of BooleanValue? Are there any documents/references I should look up to understand how static typing is supposed to work in Ibis?

What version of ibis are you using?

ibis-framework[duckdb]>=9.0.0 specifically - on 9.0.0 right now

What backend(s) are you using, if any?

DuckDB

Relevant log output

No response

Code of Conduct

• [X] I agree to follow this project's Code of Conduct

[jcrist]

Hi - thanks for opening this. Two things:

• Our type annotations aren't great. Ibis code is pretty dynamic, generally you can trust the output of the method/function/whatever to be properly typed, but the types of arguments are likely stricter than things that are actually supported. In this case here filter is annotated to take a BooleanValue (which will accept a BooleanScalar or BooleanColumn), but it also will work with a standard bool (ibis will automatically coerce to a BooleanScalar). Going through the APIs to improve the accuracy of the type annotations and add tests to ensure they're accurate is a large project and not one we're prioritizing at the moment. Right now I recommend either submitting PRs to fix annotations for bugs that you run into, or (better, more feasible suggestion) just turn off typing for ibis.

• In your specific case this is also a case of user error. The .equals method on an expression checks for expression-level equality (whether the left and right side are identical expressions). t.filter(t.a == 0) is what you want here.

[fwojciec]

Thanks for the feedback - it all makes sense. I'm happy to close this as this is not really a bug, and I see if I can come up with a PR that addresses the specific issue here.