feat(python)!: Improve some error types and messages

[aminalaee]

Partially addresses #10421

Changes:

• Improve error messages to be in accordance with the guidelines (see linked issue).
• Fix the error types of some functions. This could be considered a breaking change.

[stinodego]

Thanks for the effort! Though I definitely see some issues still - messages that start with a capital letter, or mistakenly added newlines.

I'll put this on draft, just mark it ready-to-review when you've made a pass to verify your changes!

[alexander-beedie]

Nice one! I was going to look at this but you're off to a great start - go for it :))

[mcrumiller]

Out of curiosity, would we benefit from an error-helper function, or is this too much indirection? Something like

pl.exceptions.raise(
    "ValueError",
    header="polars can not currently guarantee zero-copy conversion from Arrow for categorical columns",
    desc="Set `allow_copy=True` or cast categorical columns to string first."
)

Which could then ensure proper formatting, and insert newlines where appropriate?

[alexander-beedie]

Out of curiosity, would we benefit from an error-helper function, or is this too much indirection? Something like

Perhaps, but it would find its way into the reported stack which isn't ideal 🤔

[mcrumiller]

@alexander-beedie there are ways to modify the traceback with the traceback module, Jinja is well-known for doing this. But it's probably more effort than it's worth, especially if we want to be compatible on multiple versions.

[alexander-beedie]

@alexander-beedie there are ways to modify the traceback with the traceback module, Jinja is well-known for doing this. But it's probably more effort than it's worth, especially if we want to be compatible on multiple versions.

Indeed; doing "clever" things with your exceptions IN your exception handling code is something I'm not a big fan of ;)

[stinodego]

Nice improvement! Congrats on your first contribution 🎉

I did have to fix a few false positives, and there are many messages that still do not conform to the guidelines. Plus the same improvements should be done on the Rust side. So the linked issue remains open for now.

If you wish to continue to work on the issue, I would advise a more concentrated approach: pick a module and make sure all error messages are perfect. That way, it's easier for us to review.

[mcrumiller]

@aminalaee 55 files changed! Good job here. We all encounter error messages every day so this is a nice improvement :)

[aminalaee]

Thanks for the review and kind words.
Yes I would love to keep working on the issue on both Python and Rust side and as you said it is better to go module by module as there are cases that are harder to find. For example I saw this by going through the files:

msg = “Invalid operation”
raise ValueError(msg)