rust-analyzer seems to be getting confused between trait methods and inherent methods

[TDplai]

rust-analyzer version: rust-analyzer 1.78.0 (9b00956 2024-04-29)

Can also reproduce with rust-analyzer 1.80.0-nightly (ab14f94 2024-05-13)

rustc version: rustc 1.78.0 (9b00956e5 2024-04-29)

editor or extension: Neovim 0.9.5

relevant settings:

• CARGO_HOME is set to $HOME/.local/share/cargo
• RUSTUP_HOME is set to $HOME/.local/share/rustup

code snippet to reproduce:

pub trait A {
    // Removing this associated type makes everything work
    type X: B;
}
// Making `B` not a subtrait of `A` makes everything work
pub trait B: A {
    // Changing this to take `&self` makes everything work
    fn confused_name(self, _: i32);
}

pub struct Foo;
impl Foo {
    // Changing this to take `self` by value makes everything work
    pub fn confused_name(&self) {}
}

pub fn repro<T: A>() {
    // rust-analyzer reports "expected 1 argument, found 0"
    Foo.confused_name();
}

Note that the above code does compile.

[dfireBird]

This seems like a problem with chalk, I traced it to this trait_solve_query call:

rust-analyzer/crates/hir-ty/src/method_resolution.rs

Line 1180 in 6524922

if db.trait_solve(krate, block, goal.cast(Interner)).is_none() {

Which here should return None but returns a Some and it correctly returns None if any of the changes mentioned in the example code. I'm not sure what more info should I add here, if there is anything needed please ask.

EDIT: The goal that's sent to chalk: (UCanonical { canonical: Canonical { value: InEnvironment { environment: Env([for<> FromEnv(!0_1: TraitId(0))]), goal: Implemented(AdtId(StructId(StructId(0)))<[]>: TraitId(1)) }, binders: [] }, universes: 1 })

[flodiebold]

Probably rust-lang/chalk#750 or rust-lang/chalk#727.