Return Result from Shape methods

[camsteffen]

• The Shape modifier methods now return Result<Shape, ExeedsMaxWidthError>
• There are _opt variants of the methods for when an Option is still needed
  • I believe this will remain useful in the long term, since sometimes we want to fallback and never error out
• There is a minor change in behavior for when multiple methods are chained together - the width used in the error will be the last successfully modified width. I think this is okay?

[ding-young]

IMO, using the last successfully modified width is fine. Actually, that might be the more accurate width in some cases.

[calebcartwright]

IMO, using the last successfully modified width is fine. Actually, that might be the more accurate width in some cases.

by the way @ding-young thank you for all your excellent work this year! I know you worked closely with @ytmimi and we didn't get a chance to interact much, but i really appreciate all the work you did and I'm glad that you have expressed an interest in remaining engaged with it

[calebcartwright]

There is a minor change in behavior for when multiple methods are chained together - the width used in the error will be the last successfully modified width. I think this is okay?

@camsteffen - re: this item, i've not looked at the diff yet but if there's a change to the formatting output (even if it's an improvement) then we'll need to edition gate it behind style edition 2027

e.g.

rustfmt/src/items.rs

Line 147 in 777e25a

let init_str = if style_edition >= StyleEdition::Edition2024 {

[camsteffen]

There is a minor change in behavior for when multiple methods are chained together - the width used in the error will be the last successfully modified width. I think this is okay?

@camsteffen - re: this item, i've not looked at the diff yet but if there's a change to the formatting output (even if it's an improvement) then we'll need to edition gate it behind style edition 2027

There should be no change in user-facing behavior, formatting. Just a change to the contents of the error object, which is currently never used IIUC.

[ytmimi]

Really happy with the simplification! Left a few minor comments inline.

[ytmimi]

Would it be better to define sub_width_opt, and really all the *_opt functions in terms of their Result<T, MaxWidthError> versions?

    pub(crate) fn sub_width_opt(&self, n: usize) -> Option<Shape> {
        self.sub_width(n, rustc_span::DUMMY_SP).ok()
    }

[camsteffen]

I considered that and I opted to avoid using DUMMY_SP. But it's not a strong preference for me.

[camsteffen]

Another possibility is to go the other way around with something like sub_width_opt(n).ok_or_else(|| self.max_width_error(span))

[ytmimi]

I think you're second idea would be a nice way to avoid using a DUMMY_SP.

[camsteffen]

done!

[ytmimi]

Thanks for helping simplify the code with these Shape updates!