Changing order of trait bounds changes available impls

[Pratyush]

I tried this code (playground link):

use core::ops::AddAssign;

trait TraitA<B>: Copy + AddAssign<B> + AddAssign<Self> {}

trait TraitC<B> { type Thing: TraitA<B>; }

fn use_c<B, C: TraitC<B>>(mut a: C::Thing, b: B) {
	a += b; // errors
	a += a;
}

I expected this to compile, because the trait bounds on TraitA are sufficient. Instead, rustc fails to find the AddAssign<B> bound for C::Thing, and errors out with the following error message:

  |
7 | fn use_c<B, C: TraitC<B>>(mut a: C::Thing, b: B) {
  |          - this type parameter
8 |     a += b; // errors
  |          ^ expected associated type, found type parameter `B`
  |
  = note: expected associated type `<C as TraitC<B>>::Thing`
              found type parameter `B`
help: consider further restricting this bound
  |
7 | fn use_c<B, C: TraitC<B> + TraitC<B, Thing = B>>(mut a: C::Thing, b: B) {
  |                          ^^^^^^^^^^^^^^^^^^^^^^

(Note that the help message, while technically correct, is actually unhelpful here.)

Meta

This bug occurs in the latest stable, beta, and nightly (this can be verified on the playground link).

This might be related to #41756.

[Pratyush]

Update: this compiles successfully with chalk; the output of

rustc +nightly --crate-name test_chalk --edition=2018 src/lib.rs --crate-type lib --emit=dep-info,metadata -C embed-bitcode=no -C debuginfo=2 -C metadata=1ff6265fd5bdefac -C extra-filename=-1ff6265fd5bdefac --out-dir /tmp/test_chalk/target/debug/deps -C incremental=/tmp/test_chalk/target/debug/incremental -L dependency=/tmp/test_chalk/target/debug/deps -C target-cpu=native -Z chalk=yes

is

warning: function is never used: `use_c`
 --> src/lib.rs:7:4
  |
7 | fn use_c<B, C: TraitC<B>>(mut a: C::Thing, b: B) {
  |    ^^^^^
  |
  = note: `#[warn(dead_code)]` on by default

warning: 1 warning emitted

[tesuji]

@rustbot modify labels: A-traits T-compiler A-associated-items

[Pratyush]

Update: this compiles on the latest nightly. Is there a way to figure out what changed?

[tesuji]

You could try out bisect-rustc to find which nightly and possibly which PR fix the problem: https://github.com/rust-lang/cargo-bisect-rustc/blob/master/TUTORIAL.md

[Pratyush]

Ahh I think it was this (see “ambiguous projection bounds”) #73905 (comment)

[Pratyush]

Can confirm that this is fixed due to #73905; closing.