Intro of `rustc_attr_parsing::attributes::prelude` regressed "canonical" paths of certain well-known compiler items in rustdoc search

[fmease]

Important

This is likely an instance of #124608 (solved by rust-lang/rfcs#3011). However it might be worth discussing this recent and concrete regression in isolation.

#145507 (cc @jdonszelmann) introduced rustc_attr_parsing::attributes::prelude which among other things re-exports several "important" / well-known items, namely rustc_span::{DUMMY_SP, Ident, Span, Symbol, sym} (check out https://github.com/rust-lang/rust/blob/d327d651e2583eb601978179f2ca9808f5e243bb/compiler/rustc_attr_parsing/src/attributes/prelude.rs for a complete list (NB: only focus on cross-crate re-exports, not local ones)).

rustdoc-search now considers this prelude to be the "canonical" location of these items. This means that searching for Symbol no longer renders rustc_span::span::Symbol as its first result:

but rustc_attr_parsing::attributes::prelude::Symbol:

---

To be clear, they still link to the same item, it's just no longer apparent where this item actually comes from. See also:

[fmease]

This isn't a super big deal (since most ppl use IDE auto-import instead of copy/pasting from API docs) but it might confuse new contributors who try to learn the internal compiler surface.

[fmease]

#[doc(no_inline)] on these re-exports doesn't help unfortunately, it would make it even worse actually and lead to said path / hyperlink pointing to a non-existing site.

[jdonszelmann]

Huh that's awful.... And it does that despite this module being inaccessible outside rustc_attr_parsing ? Because in most places you can't actually use this path to import these items at all

[fmease]

Yea, it's bad. The fact that it's private doesn't really matter here because we document the APIs with --document-private-items, soo

[bjorn3]

Maybe use #[doc(hidden)] for the prelude re-exports? That would also avoid rust-analyzer suggesting something like use rustc_attr_parsing::attributes::prelude::Symbol.

[fmease]

Could be worth a try 🤷

[jdonszelmann]

Well within those modules I'd quite like the prelude to work. But I guess it's better than the current situation