Regression in what path is chosen for error messages

[dtolnay]

#[macro_use]
extern crate serde_derive;

#[derive(Serialize)]
struct S;

fn main() {
    i32::from(""); // error
}

Error with rustc 1.21.0-beta.3 (ea1fd7d 2017-09-15):

error[E0277]: the trait bound `i32: std::convert::From<&str>` is not satisfied
 --> src/main.rs:8:5
  |
8 |     i32::from(""); // error
  |     ^^^^^^^^^ the trait `std::convert::From<&str>` is not implemented for `i32`

Error with rustc 1.22.0-nightly (417c738 2017-10-05):

error[E0277]: the trait bound `i32: _IMPL_SERIALIZE_FOR_S::_serde::export::From<&str>` is not satisfied
 --> src/main.rs:8:5
  |
8 |     i32::from(""); // error
  |     ^^^^^^^^^ the trait `_IMPL_SERIALIZE_FOR_S::_serde::export::From<&str>` is not implemented for `i32`

I would expect the error message to show std::convert::From, not Serde's re-export at least in code that has nothing to do with Serde.

[ExpHP]

d93036a...dead08c

Only two big merges in this timeframe. cc @alexcrichton?

[alexcrichton]

Sorry I don't know what there may have led to this :(

[najamelan]

@dtolnay I have the impression this is fixed on 1.23, can you confirm?

[dtolnay]

Seems like it. Thanks!

[dtolnay]

Reopening because while the original code now gives a reasonable error, I am still seeing serde::export in error messages seemingly at random. With rustc 1.23.0-nightly (e97ba83 2017-11-25):

extern crate serde;

fn main() {
    struct S;
    S.clone();
}

error[E0599]: no method named `clone` found for type `main::S` in the current scope
 --> src/main.rs:5:7
  |
4 |     struct S;
  |     --------- method `clone` not found for this
5 |     S.clone();
  |       ^^^^^
  |
  = help: items from traits can only be used if the trait is implemented and in scope
  = note: the following trait defines an item `clone`, perhaps you need to implement it:
          candidate #1: `serde::export::Clone`

Adding extern crate chrono before extern crate serde fixes the message, but adding it after and the message is still wrong.

[najamelan]

I also see error messages that change when the order of use statements is changed, which was unexpected for me. But maybe that's for a different issue? I agree that in the code example you show above it is undesirable to mention serde in the error, but maybe I miss something fundamental about how serde interacts with compilation.

[dtolnay]

#46112 has some more examples of what looks like the same bug.

[dtolnay]

Fixed by #46708 and #46838.