Inconsistent failure in byte slice &[u8] comparison caused by unrelated code

[aidanhs]

I tried this code:

[package]
name = "rbup"
version = "0.1.0"
authors = ["Aidan Hobson Sayers <aidanhs@cantab.net>"]
edition = "2018"

[dependencies]
assert_fs = "1.0"

use std::os::unix::ffi::OsStrExt;
use std::path::PathBuf;

fn main() {}

pub fn foo() {
    //let temp = assert_fs::TempDir::new().unwrap(); // XXX
}

pub fn x() {
    let prefix: &[u8] = &[];
    let name: &Vec<u8> = &vec![];
    let _ = prefix > name.as_slice(); // works
    let _ = prefix > name; // broken iff XXX is uncommented
}

pub fn y() {
    let lastfile: &PathBuf = &PathBuf::new();
    let path: PathBuf = PathBuf::new();
    assert!(lastfile.as_os_str().as_bytes() > path.as_os_str().as_bytes()); // works
    assert!(lastfile.as_os_str().as_bytes() > &path.as_os_str().as_bytes()); // broken iff XXX is uncommented
}

I expected to see this happen: the code either consistently compiles or fails to compile.

Instead, this happened: depending on whether the line marked XXX is commented or not, the unrelated lines marked broken cause compilation failures (in turn, commenting those out allows compilation to succeed). i.e. broken and XXX lines are mutually exclusive.

Meta

rustc --version --verbose:

rustc 1.48.0-nightly (623fb90b5 2020-09-26)
binary: rustc
commit-hash: 623fb90b5a1f324e0ec44085116bf858cef19a00
commit-date: 2020-09-26
host: x86_64-unknown-linux-gnu
release: 1.48.0-nightly
LLVM version: 11.0

Backtrace

+ cargo --color=always build --release
   Compiling rbup v0.1.0 (/tmp/x)
error[E0277]: can't compare `[u8]` with `Vec<u8>`
  --> src/main.rs:14:20
   |
14 |     let _ = prefix > name; // broken iff XXX is uncommented
   |                    ^ no implementation for `[u8] < Vec<u8>` and `[u8] > Vec<u8>`
   |
   = help: the trait `PartialOrd<Vec<u8>>` is not implemented for `[u8]`
   = note: required because of the requirements on the impl of `PartialOrd<&Vec<u8>>` for `&[u8]`

error[E0277]: can't compare `[u8]` with `&[u8]`
  --> src/main.rs:21:45
   |
21 |     assert!(lastfile.as_os_str().as_bytes() > &path.as_os_str().as_bytes()); // broken iff XXX is uncommented
   |                                             ^ no implementation for `[u8] < &[u8]` and `[u8] > &[u8]`
   |
   = help: the trait `PartialOrd<&[u8]>` is not implemented for `[u8]`
   = note: required because of the requirements on the impl of `PartialOrd<&&[u8]>` for `&[u8]`

error: aborting due to 2 previous errors

For more information about this error, try `rustc --explain E0277`.
error: could not compile `rbup`

To learn more, run the command again with --verbose.

[tamuhey]

Smaller example (playground):

use bstr; // Comment out here
fn main() {
    let a: &[u8] = &[];
    let b: &Vec<u8> = &vec![];
    a > b;
}

And MVCE (playground):

fn main() {
    let a: &[u8] = &[];
    let b: &Vec<u8> = &vec![];
    a > b;
}

// ========== Comment out from this line ==========
use std::cmp::{Ordering, PartialOrd};

struct Foo();
impl PartialEq<[u8]> for Foo {
    fn eq(&self, other: &[u8]) -> bool {
        true
    }
}
impl PartialEq<Foo> for [u8] {
    fn eq(&self, other: &Foo) -> bool {
        true
    }
}
impl PartialOrd<[u8]> for Foo {
    fn partial_cmp(&self, other: &[u8]) -> Option<Ordering> {
        None
    }
}
impl PartialOrd<Foo> for [u8] {
    fn partial_cmp(&self, other: &Foo) -> Option<Ordering> {
        None
    }
}

[camelid]

Interestingly, changing this:

fn main() {
    let a: &[u8] = &[];
    let b: &Vec<u8> = &vec![];
    a > b;
}

To this:

fn main() {
    let a: &[u8] = &[];
    let b: &Vec<u8> = &vec![];
    a == b;
//    ^^ I changed the `>` to `==`
}

Does cause an error:

error[E0277]: can't compare `[u8]` with `std::vec::Vec<u8>`
 --> src/main.rs:5:22
  |
5 |     println!("{}", a == b);
  |                      ^^ no implementation for `[u8] == std::vec::Vec<u8>`
  |
  = help: the trait `std::cmp::PartialEq<std::vec::Vec<u8>>` is not implemented for `[u8]`
  = note: required because of the requirements on the impl of `std::cmp::PartialEq<&std::vec::Vec<u8>>` for `&[u8]`

[camelid]

Reposting from Zulip:

@tamuhey: I've confirmed that the same issue occurs with 1.13.0

[Dylan-DPC-zz]

Relevant discussion