rustdoc: support for precise capturing

[fmease]

Source:

#![feature(precise_capturing)]

pub fn demo(_: &str) -> impl Sized + use<> {}

Current rendered output (both HIR-based and middle-based) below (issue: dropping the precise capturing arg):

pub fn demo(_: &str) -> impl Sized

Tracking:

• Tracking Issue for precise_capturing #123432

[fmease]

I consider this issue to be partially blocked: Please don't try to add use<> support for inlined cross-crate reexports — more generally: middle-based cleaning — before I fix #113015 (which should happen very soon!).

Note that use<> doesn't get preserved post-HIR IINM, so it'd be non-trivial to impl anyway. To clarify, while it's possible to recover the list of captured regions from rustc_middle::ty data (via variance data: bivariant means not captured IIRC — maybe that's only true for TAITs not RPITs, haven't double-checked), it won't necessarily coincide with what the user wrote literally.

[fmease]

Furthermore, there's some design wiggle room which we should probably explore in a T-rustdoc meeting (I might add it to the agenda later).

Namely, with the edition differences around opaque type captures it might make sense for the rendered output to deviate from the actual source code as written by the user and to provide a 'normalized view' that's edition-agnostic (e.g., by always listing captures explicitly, etc. etc.).

[compiler-errors]

Note that use<> doesn't get preserved post-HIR IINM, so it'd be non-trivial to impl anyway. To clarify, while it's possible to recover the list of captured regions from rustc_middle::ty data (via variance data: bivariant means not captured IIRC — maybe that's only true for TAITs not RPITs, haven't double-checked), it won't necessarily coincide with what the user wrote literally.

I think we could/should probably just expose it as a query which is encoded in the crate.