Identifiers with combining characters do not work as captured identifiers in `println!` format strings

[interactive-toast]

Identifiers with combining characters do not work as captured identifiers in println! format strings.

Minimal reproducible example

Identifier using combining characters([0x61, 0xCC, 0x80]) as a captured identifier

fn main() {
    let à = 100;
    println!("à = {à}");
}

I expected to see this happen:

Successful compilation

Instead, this happened:

error[E0425]: cannot find value `à` in this scope
 --> not_works.rs:3:21
  |
3 |     println!("à = {à}");
  |                    ^ not found in this scope

error: aborting due to 1 previous error

Other tests

	Combining Characters	Single Codepoint
Captured Identifier	🚫	✅
Positional Argument	✅	✅

1. Identifier using a single unicode codepoint([0xC3, 0xA0]) as a positional argument

fn main() {
    let à = 100;
    println!("à = {}", à);
}

This compiles and works as expected

2. Identifier using combining characters([0x61, 0xCC, 0x80]) as a positional argument

fn main() {
    let à = 100;
    println!("à = {}", à);
}

This compiles and works as expected

3. Identifier using a single unicode codepoint([0xC3, 0xA0]) as a captured identifier

fn main() {
    let à = 100;
    println!("à = {à}");
}

This compiles and works as expected

Meta

Using rust playground, I tested all these examples with stable, beta, and nightly.

1.84.1
1.85.0-beta.9 (2025-02-13 461de7492e5354419cf2)
1.86.0-nightly (2025-02-14 d8810e3e2dab96778d20)

Not sure if this is a known limitation but I couldn't find any info on it.

Seems like a unicode normalization issue.

Also, mixing a variable that is declared as combining characters and then used
as a single codepoint works, it's just combining characters as a captured
identifier that doesn't work.

[workingjubilee]

Yes, identifiers are NFC normalized but it seems strings are not, which leads to this amusing consequence.

I'm not sure if we should fix this.

[crlf0710]

I do think this is a bug. Glancing through the code, it seems the fundamental issue is caused that this code:

rust/compiler/rustc_builtin_macros/src/format.rs

Line 376 in 1c3b035

let name = Symbol::intern(name);

This string didn't go through the nfc normalize process here:

rust/compiler/rustc_parse/src/lexer/mod.rs

Lines 996 to 1005 in 1c3b035

	pub fn nfc_normalize(string: &str) -> Symbol {
	use unicode_normalization::{IsNormalized, UnicodeNormalization, is_nfc_quick};
	match is_nfc_quick(string.chars()) {
	IsNormalized::Yes => Symbol::intern(string),
	_ => {
	let normalized_str: String = string.chars().nfc().collect();
	Symbol::intern(&normalized_str)
	}
	}
	}

If possible, i'd suggest this refactoring:

1. Split Symbol::intern into Symbol::intern_for_ident and Symbol::intern_for_nonident. The former goes through the nfc normalize process, while the later does not.
2. Go through all calls to the old api, and replace them with one of the above two.

I believe this can fix all similar issues once for all.