`unnecessary_safety_comment` inconsistency with wrong-style safety section

[ojeda]

Summary

unnecessary_safety_comment appears to detect /// SAFETY: ... as a safety comment (i.e. not # Safety section) in a normal function, but it doesn't inside an impl block.

Lint Name

unnecessary_safety_comment

Reproducer

I tried this code:

/// SAFETY: ...
fn f() {}

and it triggers unnecessary_safety_comment. However, this:

struct S;
impl S {
    /// SAFETY: ...
    fn f() {}
}

does not trigger it.

I assume unnecessary_safety_comment is intended to only lint on actual safety comments (i.e. not safety docs/sections), even if they use /// instead of //. If so, then the former is expected, and the latter is a false negative.

Please also see this other suggested lint for consistency on safety comments/docs: #15033. For instance, if the former case above were to be a false positive, one could use that new/suggested lint to catch a case like this where the function was supposed to be unsafe fn. And if the former is indeed expected, then perhaps the message of the lint could be clarified if the user used /// instead of just //.

More combinations: https://godbolt.org/z/43onqxYrh

Version

rustc 1.89.0-nightly (d13a431a6 2025-06-09)
binary: rustc
commit-hash: d13a431a6cc69cd65efe7c3eb7808251d6fd7a46
commit-date: 2025-06-09
host: x86_64-unknown-linux-gnu
release: 1.89.0-nightly
LLVM version: 20.1.5