Add overflow_closures and overflow_match_arms opts

[spinda]

cc #1791

overflow_closures

Allow overflow on closures

• Default value: true
• Possible values: true, false

true:

result.and_then(|maybe_value| match maybe_value {
    None => ...,
    Some(value) => ...,
})

false:

result.and_then(|maybe_value| {
    match maybe_value {
        None => ...,
        Some(value) => ...,
    }
})

overflow_match_arms

Allow overflow on match arms

• Default value: true
• Possible values: true, false

true:

match lorem {
    None => if ipsum {
        println!("Hello World");
    },
    Some(dolor) => ...,
}

false:

match lorem {
    None => {
        if ipsum {
            println!("Hello World");
        }
    }
    Some(dolor) => ...,
}

[topecongiro]

Thank you for you PR! Could you please add tests for overflow_closures as well?

[spinda]

Oops, I thought I'd done that. Will add.

[emilio]

(Totally not a rustfmt expert, so could totally be wrong, but won't this undo any other formatting on everything else inside the expression?)

[spinda]

Look at how rewrite_closure_expr is used:

https://github.com/rust-lang-nursery/rustfmt/blob/6fb17c41301e5a98748882cc6ad6a627def8eb80/src/expr.rs#L629-L648

https://github.com/rust-lang-nursery/rustfmt/blob/6fb17c41301e5a98748882cc6ad6a627def8eb80/src/expr.rs#L650-L653

It's called to attempt to format the closure body expression without wrapping it in a block. If it fails and returns None, rustfmt will fall back to rewrite_closure_block and format the expression with a block surrounding it. rewrite_closure_block will perform any formatting that's still appropriate for the expression in that context.

[emilio]

Yup, ok, fair enough. That happens when drive-by reviewing, thanks for explaining :)

[nrc]

Bikeshedding, I would prefer closures_force_block rather than overflow_closures (and likewise for match arms). "Overflow" suggests something being too full, which is not the case here, we're just using a block (you'd need to flip the true/false too).

[spinda]

@nrc That would seem to imply always forcing a block for closures and match arms, whereas this option only forces a block when the respective constructs "overflow" a single line.

[nrc]

Perhaps multline_closure_forces_block or something similar? The issues with the current names are that they don't indicate what happens on overflow, and "overflow" is not a term we use in Rustfmt at the moment.

[spinda]

I like multiline_closure_forces_block. My original choice of overflow did in fact come from the rustfmt code, though; see https://github.com/rust-lang-nursery/rustfmt/pull/1898/files#diff-180bb3ad456cc10517d878fb7795341cR1708 for example (can_be_overflowed).

[nrc]

Thanks for making the changes!