feat: catch Polars exceptions, unify exception raising more

[MarcoGorelli]

Unify exceptions, and re-raise Polars ones as Narwhals ones

This doesn't unify everything...but I think it's a nice step forwards, and should be totally backwards-compatible mostly backwards-compatible (unless anyone catches Polars exceptions specifically in tests)

Towards #1373

[MarcoGorelli]

The only tests failures are from Marimo, where they're checking the Polars error very tightly in their tests

        with pytest.raises(pl.exceptions.ComputeError):
            # Polars doesn't support writing nested lists to csv
>           manager.to_csv()

      with pytest.raises(pl.exceptions.ComputeError):
            # Polars doesn't support writing nested lists to csv
>           complex_data.to_csv()

[MarcoGorelli]

Hey @mscolnick , I just wanted to run this past you

There's been requests to unify the exceptions from Narwhals. We do that somewhat already, but this PR makes another step towards that.

As far as I can't, there'd be no user-facing impact, it would just be matter of updating the expected exception in 2 tests from the Marimo test suite. I'd be happy to make a PR if welcome

Would you be OK with this change? Else no worries, we could always just preserve the current exceptions in the v1 API, I just figured that for something so small it might not be worth it

[mscolnick]

Yea that is fine with us and would appreciate the help with the change, thank you!

[dangotbanned]

@MarcoGorelli I'd be careful with how this impacts performance, since we only get zero-cost exceptions on >=3.11

Somewhat related:

• https://docs.astral.sh/ruff/rules/try-except-in-loop/
• https://bugs.python.org/issue40222
• https://docs.python.org/3/whatsnew/3.11.html#misc

[MarcoGorelli]

Thanks Dan! I'll do some checks before shipping it - I think it shouldn't be an issue because it's only on Dataframe methods, e.g. for

df.filter(super_complicated_expression_1, super_complicated_expression_2, super_complicated_expression_3)

the try-except would only happen when filter is called, and not for each expression

Nonetheless, I would like to time it and be careful with it, so this'll stay out of today's release, we can revisit for the next one when we've done more careful checks