rustdoc: Various `#[doc]` attributes are not fully validated (e.g., `#[doc(hidden)]`, `#[doc(no_inline)]`)

[nik-rev]

The docs don't mention this. Is it supposed to represent the reason why the item is hidden, or was it an oversight?

12 instances of this on GitHub: https://github.com/search?q=%2F%23%5C%5Bdoc%5C(hidden%20%3D%20%22%2F&type=code

[zachs18]

Note that this string does not show up in the rendered documentation when using the (unstable) --document-hidden-items:

cargo new --lib doc-priv && cd doc-priv

// lib.rs
#[doc(hidden = "for some reason")]
pub fn __foo() {}

RUSTDOCFLAGS="-Zunstable-options --document-hidden-items" cargo +nightly doc --open

[fmease]

This is most definitely a bug (it's not tested or documented anywhere). In the early days, rustc used to perform none or only minimal validation for attributes (where they're placed, what shape they have) and over the years we've slowly gotten more principled in this area.

Due to the somewhat "widespread" usage of this specific form and just to be safe, this needs to start out as a lint warning of invalid_doc_attributes (hopefully we'll turn this lint into a future-incompatibility one at some point and in the far future make it a hard error).

---

#[doc(hidden = "…")] isn't the only form we don't warn / error on, there's a whole class of them:

• #[doc(hidden = 0)]
• #[doc(hidden(something))]
• #[doc(no_inline(var = 1))]
• …

[fmease]

However, I want to block anyone from blindly rushing in and addressing this; this should be done as part of the attribute rewrite (#131229) and Jana already claimed all #[doc] attrs (we already have a very early draft PR here: #147097, but that might not end up as the actual one that we'll merge).