Nightly-only cfg gating like `#[cfg(target_has_atomic)]` is confusing

EDIT: this is intended that cfg(target_has_atomic) is nightly-only (not feature-gated), but I find it still very confusing.

Note that #[cfg(target_has_atomic)] is not to be confused with #[cfg(target_has_atomic = "8")].

Given a snippet (just rustc --crate-type="lib", no additional --cfg flags)

#[cfg(target_has_atomic)]
fn foo() {}

• On stable 1.82.0 this produces no warnings.
• On nightly 2024-11-20 this produces a dead_code warning indicating target_has_atomic cfg is enabled.

warning: function `foo` is never used
 --> src/lib.rs:2:4
  |
2 | fn foo() {}
  |    ^^^
  |
  = note: `#[warn(dead_code)]` on by default

AFAICT there is no test coverage for the intended behavior of just #[cfg(target_has_atomic)] alone, other test coverage is mostly for check-cfg.

cc @Urgau since you worked on builtin-cfg rejection do you know anything about if this is intended, and if so, what is the intended behavior here?

Start by reproducing the two reported cases with rustc --crate-type="lib" and the bare #[cfg(target_has_atomic)] snippet on stable and nightly. Review the existing builtin-cfg and check-cfg tests, then establish what behavior is intended and add coverage for the bare form; done means the behavior is explicitly settled and tested.