jiff range int trait impls introduce a bunch of ambiguity for Into::into

[khuey]

After upgrading a project from env_logger 0.11.6 (which uses humantime) to env_logger 0.11.7 (which uses jiff 0.2) I get a bunch of errors like this:

 error[E0283]: type annotations needed
    --> data-rewriter/src/rust_std.rs:1222:40
     |
1222 |                 assert!(index < length.into());
     |                               -        ^^^^
     |                               |
     |                               type must be known at this point
     |
     = note: multiple `impl`s satisfying `i128: PartialOrd<_>` found in the following crates: `core`, `jiff`:
             - impl PartialOrd for i128;
             - impl<MIN, MAX> PartialOrd<jiff::util::rangeint::ri128<MIN, MAX>> for i128
               where the constant `MIN` has type `i128`, the constant `MAX` has type `i128`;
help: try using a fully qualified path to specify the expected types
     |
1222 |                 assert!(index < <u64 as Into<T>>::into(length));
     |                                 +++++++++++++++++++++++      

(where index is i128 and length is u64).

[BurntSushi]

What is your desired outcome here? There are maybe some trait impls I could remove here, but I think the underlying issue here is that your code is specifically relying on the absence of traits impl for inference to work. There are some well known cases where we kind of rely on this (IIRC, it's why slice indexing on works with usize even though there are From<u8> for usize and From<u16> for usize trait impls), but for cases like this, any crate can add impls like Jiff is doing, so it would be better to write code that is robust with respect to these sorts of things.

Alas, I can see how if these errors are showing up in dependencies, it might not be feasible to do this. I'm in general happy to be pragmatic about this and see if there are some trait impls I can remove even if makes some of Jiff's internals a little less convenient. But I can't do that without an MRE. Can you please provide one?

[khuey]

The minimal code required to trigger this:

use env_logger; // Works with env_logger 0.11.6, fails with env_logger 0.11.7.

fn main() {
    let x: u64 = 1;
    let y: i128 = 0;
    assert!(y < x.into());
}

It fails with e.g. u32 and i64 as well.

What is your desired outcome here?

Well, ideally inference would just keep working. I recognize that that may not be possible and I may have to switch to e.g. i128::from(x) everywhere but having simple inferences polluted by foundational crates like this is certainly unergonomic.

[BurntSushi]

Yes, I'm definitely willing to give up some internal conveniences to avoid this. I'll take a look at this more closely this evening.

Even if Jiff fixes this, I'd still suggest switching your code to i128::from(..) or whatever else to avoid this. Otherwise it's likely you'll just hit this problem again with some other crate.

[BurntSushi]

This should be fixed in jiff 0.2.4 on crates.io.