Change struct expr pretty printing to match rustfmt style

[dtolnay]

This PR backports trailing comma support from https://github.com/dtolnay/prettyplease into rustc_ast_pretty and uses it to improve the formatting of struct expressions.

Example:

macro_rules! stringify_expr {
    ($expr:expr) => {
        stringify!($expr)
    };
}

fn main() {
    println!("{}", stringify_expr!(Struct {
        a: Struct { b, c },
    }));
    println!("{}", stringify_expr!(Struct {
        aaaaaaaaaa: AAAAAAAAAA,
        bbbbbbbbbb: Struct {
            cccccccccc: CCCCCCCCCC,
            dddddddddd: DDDDDDDDDD,
            eeeeeeeeee: EEEEEEEEEE,
        },
    }));
}

🤮 Before:

Struct{a: Struct{b, c,},}
Struct{aaaaaaaaaa: AAAAAAAAAA,
    bbbbbbbbbb:
        Struct{cccccccccc: CCCCCCCCCC,
            dddddddddd: DDDDDDDDDD,
            eeeeeeeeee: EEEEEEEEEE,},}

After:

Struct { a: Struct { b, c } }
Struct {
    aaaaaaaaaa: AAAAAAAAAA,
    bbbbbbbbbb: Struct {
        cccccccccc: CCCCCCCCCC,
        dddddddddd: DDDDDDDDDD,
        eeeeeeeeee: EEEEEEEEEE,
    },
}

[rust-highfive]

r? @cjgillot

(rust-highfive has picked a reviewer for you, use r? to override)

[cjgillot]

I'd like to understand better the purpose of this crate. It seems to be partly redundant with rustfmt. Now that rustfmt is a subtree, could we reuse part of rustfmt to do what rustc_ast_pretty does?

[jyn514]

Rustfmt is less granular than ast_pretty. If you search zulip I had a discussion with @calebcartwright about it a while ago.

[calebcartwright]

Agreed. I think one day it would be good if we can get rustfmt's API to a point where it could be utilized for formatting individual AST nodes, but for today and the very foreseeable future it's very much towards the all-or-nothing end of the spectrum that would make rustfmt a poor fit for trying to format an individual construct

[dtolnay]

@cjgillot, separately, rustfmt prioritizes quality over simplicity and performance, and is significantly slower than rustc_ast_pretty. If rustc needs to dump out megabytes of code from a -Zunpretty=expanded involving lots of macros, it's relevant whether that takes 500ms or 20 seconds to format. rustc_ast_pretty is optimized for simplicity and performance while having a "good enough" formatting that doesn't require heavyweight algorithms, very similar to the design requirements of prettyplease. If rustfmt's output quality is not close to perfect, people would never use it on their codebases, whereas if rustc_ast_pretty's output is not close to perfect, that does not matter.

[cjgillot]

Could you explain what this negative offset does? I have trouble understanding how this code manages to find the correct indentation.

[dtolnay]

The top of https://github.com/rust-lang/rust/blob/1.58.1/compiler/rustc_ast_pretty/src/pp.rs has an overview of the printing algorithm. Very briefly, when you make a sequence of cbox string break string break string break string end calls on the printer, either all of those breaks will turn into linebreaks, or none of them will turn into linebreaks. If they turn into linebreaks, the indentation at each one is the sum of the indentation of all the surrounding boxes.

So if we have:

ibox(4)cbox(0)Struct {breaka: Abreakb: Bbreak}endend

and it does not fit on the line, then you'd expect to get:

Struct {
    a: A
    b: B
    }

That's why every break also holds its own offset which is relative to the sum of the surrounding boxes' indents, so that we can set -4 on the last break and get:

Struct {
    a: A
    b: B
}

[cjgillot]

Thank you for the detailed explanation. I'm surprised such negative offset facility was not used before, but that's maybe for another refactoring.

[dtolnay]

• Rebased to resolve conflict with Factor convenience functions out of main printer implementation #93515

[cjgillot]

Overall, this PR makes a nicer output.
@bors r+ rollup

[cjgillot]

Thank you for the detailed explanation. I'm surprised such negative offset facility was not used before, but that's maybe for another refactoring.

[cjgillot]

@bors r+ rollup

[bors]

📌 Commit ca3057f has been approved by cjgillot