rustdoc: Downstream implementors are not linked to

[camelid]

Rustdoc doesn't link to implementors of a trait where the implementors are in a downstream crate; e.g., rustc_middle::mir::visit::Visitor:

I would think this is a limitation related to #74481, but the implementors are listed, they just aren't linked to, so I'm not sure what the cause is.

[jyn514]

I think this is a duplicate of #76736.

[jyn514]

If you can reproduce outside of rustc, that's a bug though.

[camelid]

I can reproduce it in a private project, but I'd rather not share the sources. I'll see if I can get an MCVE.

[camelid]

MCVE: https://github.com/camelid/rust-issue-78615

[jyn514]

The code from that repo:

// crate a
pub trait MyTrait {}

pub struct MyLocalStruct;

impl MyTrait for MyLocalStruct {}

// crate b
use crate_a::MyTrait;

pub struct MyStruct;

impl MyTrait for MyStruct {}

[camelid]

Yeah, I just wanted to provide a repo that you can just clone and build to reproduce :)

[camelid]

Weirdly, the issue seems to be spurious: Sometimes I see impl MyTrait for MyStruct in the "Implementors" list, sometimes I see nothing.

EDIT: Never mind, I was sometimes running with --document-private-items. I'll make MyStruct public to fix this.

[camelid]

The output for camelid/rust-issue-78615@3dfc5ea:

[camelid]

This partially worked as of nightly-2020-04-08:

Note that MyTrait is linked to for the MyStruct impl, but not for the MyLocalStruct impl. Also note that there is only a source link for MyLocalStruct.

[camelid]

searched nightlies: from nightly-2019-01-01 to nightly-2020-11-01
regressed nightly: nightly-2020-07-31
searched commits: from db0492a to cfc572c
regressed commit: 6b269e4

bisected with cargo-bisect-rustc v0.6.0

Host triple: x86_64-apple-darwin
Reproduce with:

cargo bisect-rustc --preserve --prompt --start=2019-01-01 -- doc 

---

This was caused by #73767.

[camelid]

I think this is unrelated, but I want to note that running cargo doc after making MyStruct private still shows it as an implementor. You have to run cargo clean to get it to go away.

I'm guessing this is because rustdoc isn't re-documenting crate_a when crate_b changes and thus it has stale data?

Interestingly, though, the reverse is not true: making MyStruct public and then running cargo doc will show it in the output... Some weird state bug I guess 🤷

[camelid]

cc @rust-lang/rustdoc

[camelid]

This appears to be fixed now, both in the rustc API docs and my MCVE! 🎉