Async gets removed from extern blocks.

[rodrigorc]

Describe the bug

The async keyword is removed when used inside an extern "C" block.

Currently this is not valid in Rust, because in C there are no async functions. But when using wasm-bindgen it is processes from a macro, and the functions are actually Javascript, not C, and there there are JS promises. Anyway, even if it is not valid, the keyword should not be removed while formatting.

To Reproduce

Format the following code:

extern "C" {
    async fn foo();
}

And see how the async keword disappears.

Expected behavior

It should keep the async keyword.

Meta

• rustfmt version: 1.4.15-beta (530eadf 2020-06-02)
• From where did you install rustfmt?: Rustup
• How do you run rustfmt: From console: rustfmt --edition 2018 test.rs

[calebcartwright]

The root cause is that the fields from the FnHeader aren't leveraged in the format handling for functions in extern blocks so qualifiers like asyncness are dropped. I believe this was done intentionally given that such qualifiers are not valid Rust code (as you noted in the issue description), and rustfmt's focus on working with and formatting valid Rust code.

However, we could certainly tweak this to support maintaining the original qualifier, and it'd be up to users to remove their invalid usage of it (outside the wasm context) vs. rustfmt removing it for them.

The relevant code for anyone interested in working on this can be found here:

rustfmt/src/formatting/items.rs

Lines 3136 to 3144 in 144c9ec

	ast::ForeignItemKind::Fn(_, ref fn_sig, ref generics, _) => rewrite_fn_base(
	context,
	shape.indent,
	self.ident,
	&FnSig::new(&fn_sig.decl, generics, self.vis.clone()),
	span,
	FnBraceStyle::None,
	)
	.map(|(s, _)| format!("{};", s)),

rustfmt/src/formatting/items.rs

Lines 237 to 252 in 144c9ec

	pub(crate) fn new(
	decl: &'a ast::FnDecl,
	generics: &'a ast::Generics,
	vis: ast::Visibility,
	) -> FnSig<'a> {
	FnSig {
	decl,
	generics,
	ext: ast::Extern::None,
	is_async: Cow::Owned(ast::Async::No),
	constness: ast::Const::No,
	defaultness: ast::Defaultness::Final,
	unsafety: ast::Unsafe::No,
	visibility: vis,
	}
	}

IMO the values set via FnSig::new still feel like a reasonable/correct set of defaults but perhaps it could be augmented or another from_* function could be added for building a FnSig geared towards formatting foreign item fns and then leverage that (instead of new) accordingly.

[ayazhafiz]

Shoot, sorry @topecongiro, I had started working on this before you assigned yourself and failed to see your assignment. Feel free to close my PR if you'd like. Apologies for the duplication of work.