Re-exported types have confusingly different documentation

[dtolnay]

Re-exported types seem to have more confusing documentation than the original type. I noticed this while reviewing rayon::Configuration which is re-exported from rayon-core.

---

original/src/lib.rs

pub trait T {}
pub struct S;

impl S {
    pub fn f<F>(_: F, _: Box<T>) -> Self
        where F: Fn()
    {
        unimplemented!()
    }
}

This is documented basically like what I wrote, which is what I would expect:

---

src/lib.rs

extern crate original;
pub use original::S;

This is documented with a couple changes, all of which are technically correct but unexpected.

• The Box<T> is now Box<T + 'static>.
• The return type has changed from Self to S.
• The where clause gives an explicit -> (). fixed as of 1.50
• The parameter name placeholders are gone (which I prefer, but it is a difference).

I am using rustc 1.22.0-nightly (f861b6e 2017-09-01).

@GuillaumeGomez

[GuillaumeGomez]

Dark magic! 😆

I'll take a look later.

[chordowl]

cc #24305

[camelid]

I think this is a cross-crate reexports issue, so labeling as such.

[jyn514]

Triage as of rustdoc 1.50.0-nightly (0f6f2d6 2020-12-06): This bug has gone away:

The where clause gives an explicit -> ().

The rest are still changed from the inner crate.

Inner:
Outer:

[jyn514]

I expect this is an inconsistency between how clean treats HIR and rustc_middle::ty somewhere (the difference being that ty works off serialized metadata).

[fmease]

Temporarily assigning myself to this issue since I am working on fixing several issues related to the handling of cross-crate trait object types in rustdoc and other minor cross-crate things. I won't fix everything mentioned in this issue however, hence this is only a temporary assignment until the PR is open.

@rustbot claim
@rustbot label -C-enhancement +C-bug

[fmease]

Update: Thanks to #103885, #107637 & #112463, the output is finally very close to acceptable (source vs. rendered):

- pub fn f<F>(_: F, _: Box<dyn T>) -> Self
- where F: Fn()
+ pub fn f<F>(_: F, _: Box<dyn T>) -> S
+ where
+     F: Fn(),

The remaining issue concerns Self types in cross-crate scenarios. I'm not gonna work on a PR for that in the foreseeable future but at some point I will probably submit a PR for it. I predict it to be a bit nasty performance-wise since it requires performing type unification on every Ty inside an impl.