Re-exported types have confusingly different documentation

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 -> ().
• The parameter name placeholders are gone (which I prefer, but it is a difference).

I am using rustc 1.22.0-nightly (f861b6ee4 2017-09-01).

@GuillaumeGomez

1. 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 (f861b6ee4 2017-09-01).

   @GuillaumeGomez

   <!-- TRIAGEBOT_START --> <!-- TRIAGEBOT_ASSIGN_START --> <!-- TRIAGEBOT_ASSIGN_DATA_START$${"user":"fmease"}$$TRIAGEBOT_ASSIGN_DATA_END --> <!-- TRIAGEBOT_ASSIGN_END --> <!-- TRIAGEBOT_END -->

   jyn at 2022-10-31 16:16:05

2. Dark magic! :laughing:

   I'll take a look later.

   Guillaume Gomez at 2017-09-04 08:27:08

3. cc #24305

   Lukas H. at 2017-09-04 21:31:18

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

   Noah Lev at 2020-10-12 04:36:51

5. Triage as of rustdoc 1.50.0-nightly (0f6f2d681 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:

   jyn at 2020-12-16 01:53:29

6. 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).

   jyn at 2020-12-16 01:54:33

7. 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

   León Orell Valerian Liehr at 2022-10-27 15:08:06

8. 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.

   León Orell Valerian Liehr at 2023-10-30 18:32:56