fix(rust, python): improve concat_list with empty list error message

[josh]

Currently calling pl.concat_list with an empty list panics in Rust. It seems like the expected behavior should just be to return an empty list series.

pl.concat_list([])
# PanicException: index out of bounds: the len is 0 but the index is 0

My current fix is rather high level and only fixes it at the Python API surface. It might be better if this could somehow be patched at a lower level. But because pl.concat_list is implemented on top of a lower lhs.concat([]) interface, a ton of stuff just breaks expecting there to be a lhs. I think I maybe out of my depth there.

Related #7235

[ritchie46]

I think we should raise an error. The [] in this case is not the list we want to concat, but the set of lists that we want to concattenate, which should not be empty.

This also keeps consistency with rust as we can raise the error there.

[josh]

I tweaked the Python API to raise a nicer error. But there's still something inconsistent with this approach at the Rust level. pub fn concat_lst<E, IE>(s: E) -> Expr is defined as infallible. To be more specific about the panic on master, it happens at a very odd place in the execution phase.

index out of bounds: the len is 0 but the index is 0

polars/polars/polars-lazy/polars-plan/src/dsl/function_expr/schema.rs

Line 29 in 86f8be0

let mut fld = fields[0].clone();

If concat_lst should consistency not permit an empty list of expressions, it's signature should probably be pub fn concat_lst<E, IE>(s: E) -> PolarsResult<Expr> instead. And a nicer is_empty guard should be added to that function to throw earlier.

Also appears that pl.concat_str([]) leads to a similar error, and it's Rust API also returns Expr, not PolarsResult<Expr>.

So what do you think about me changing the types on those public functions? Or the other direction is keeping them infallible as is, but returning an empty value respectively.

[ritchie46]

I tweaked the Python API to raise a nicer error. But there's still something inconsistent with this approach at the Rust level. pub fn concat_lst<E, IE>(s: E) -> Expr is defined as infallible. To be more specific about the panic on master, it happens at a very odd place in the execution phase.

Let's make concat_lst fallible then. :)

[josh]

Pushed the error message down to Rust. So that covers it nicely from Python without any changes there. Let me know if NoDataError is the right choice here.

Side note, it would be neat if pl.lit([]) could represent that empty list use case #6608.

[ritchie46]

Almost there. :)

[ritchie46]

Thank you @josh!