Pretty printer inlines modules

[chris-morgan]

a.rs:

// a.rs
mod b;

b.rs:

// b.rs
fn b() { }

Expected output of rustc --pretty normal a.rs:

// a.rs
mod b;

Actual output of rustc --pretty normal a.rs:

// a.rs
mod b {
    fn b() { }
}

Note the two problems here: (a) the module is inlined, and (b) the comment in the module is lost. But by fixing the first, the second becomes irrelevant.

[steveklabnik]

Triage: the output here has gotten weirder:

// a.rs
mod b {

    fn b() { }
}

I'm guessing the blank line comes from dropping the comment when b gets inlined, but who knows.

[steveklabnik]

Triage: it is now un-weird, and back to the original way it was reported in the ticket.

[Mark-Simulacrum]

I personally thought that --pretty expanding modules was a feature. Is that incorrect? It seems like it's a good idea, not a bad one--unexpanded modules would make --pretty expanded less useful on multi-file codebases, since it wouldn't show expansion of anything but the crate root, which is non-ideal.

[kennytm]

But the original report is about --pretty=normal, not --pretty=expanded... @Mark-Simulacrum

[Mark-Simulacrum]

Oh. Yeah, that's true. I'm not even really sure what the intended purpose of --pretty=normal is, I can't really think of any cases where it'd currently be the right solution. rustfmt is better if you just want pretty code...

[tinco]

I made a PR that fixes this behavior, turned out the parser simply forgot where module items come from, so it didn't know how to not expand modules, the fix was to track this information in the Ast.