"No item found" diagnostic should suggest items that differ by case

[wesleywiser]

Given the following code: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=a494a5e7386a6c9e55f7f39695dece41

enum Foo {
    Bar(u32)
}

fn main() {
    matches!(todo!(), Foo::BAR(_));
}

The current output is:

error[E0599]: no variant or associated item named `BAR` found for enum `Foo` in the current scope
 --> src/main.rs:6:28
  |
1 | enum Foo {
  | -------- variant or associated item `BAR` not found here
...
6 |     matches!(todo!(), Foo::BAR(_));
  |                            ^^^ variant or associated item not found in `Foo`

Ideally the output should include a note suggesting the variant with the same name (but different case):

error[E0599]: no variant or associated item named `BAR` found for enum `Foo` in the current scope
 --> src/main.rs:6:28
  |
1 | enum Foo {
  | -------- variant or associated item `BAR` not found here
...
6 |     matches!(todo!(), Foo::BAR(_));
  |                            ^^^ variant or associated item not found in `Foo`
  | help: a similarly named variant exists: `Bar`

[inquisitivecrystal]

@rustbot label +C-enhancement +D-newcomer-roadblock

[ben0x539]

This only doesn't Just Work because the Levenshtein thing is case sensitive, right? There is some special-casing for case-insensitive suggestions but afaict it only applies if the case-sensitive Levenshtein distance is also already low, which seems wrong at first glance.

[ben0x539]

im working on this
Edit: I'm not working on this anymore because I ran out of time getting past incremental compilation / llvm errors after a rebase.

[ben0x539]

Okay, judging by

rust/compiler/rustc_span/src/lev_distance/tests.rs

Lines 40 to 42 in d192c80

	let input = vec![Symbol::intern("AAAA")];
	// Returns None because `lev_distance > max_dist / 3`
	assert_eq!(find_best_match_for_name(&input, Symbol::intern("aaaa"), None), None);

, this is explicitly not supposed to work. shrug emoji

Edit: Okay, maybe that test case didn't really care.

The use of case-insensitive match as tie breaker for equal levenshtein distance only was introduced with #46347 (comment) afaict.