"unexpected ambiguity: CanonicalQueryInput" with conflicting associated type specifiers

@theemathas

Code

pub trait Trait {
    type Assoc: Copy + 'static;
}

const fn conjure<T>() -> T {
    panic!()
}

const fn get_assoc<T: Trait>() -> impl Copy {
    conjure::<<T as Trait>::Assoc>()
}

pub fn foo<T: Trait<Assoc = i32> + Trait<Assoc = i64>>() {
    const {
        get_assoc::<T>();
    }
}

(slightly adapted from @theemathas's example here)

Error output

note: no errors encountered even though delayed bugs were created

note: those delayed bugs will now be shown as internal compiler errors

error: internal compiler error: unexpected ambiguity: CanonicalQueryInput { canonical: Canonical { value: ParamEnvAnd { param_env: ParamEnv { caller_bounds: [Binder { value: ProjectionPredicate(AliasTerm { args: [T/#0], def_id: DefId(0:4 ~ playground[81cc]::Trait::Assoc), .. }, Term::Ty(i64)), bound_vars: [] }, Binder { value: ProjectionPredicate(AliasTerm { args: [T/#0], def_id: DefId(0:4 ~ playground[81cc]::Trait::Assoc), .. }, Term::Ty(i32)), bound_vars: [] }, Binder { value: TraitPredicate(<T as Trait>, polarity:Positive), bound_vars: [] }, Binder { value: TraitPredicate(<T as std::marker::Sized>, polarity:Positive), bound_vars: [] }] }, value: AliasTy { args: [T/#0], def_id: DefId(0:4 ~ playground[81cc]::Trait::Assoc), .. } }, max_universe: U0, variables: [] }, typing_mode: PostAnalysis } Canonical { value: QueryResponse { var_values: CanonicalVarValues { var_values: [] }, region_constraints: QueryRegionConstraints { outlives: [] }, certainty: Ambiguous, opaque_types: [], value: NormalizationResult { normalized_ty: ^0 } }, max_universe: U0, variables: [Ty(General(U0))] }
  |
  = note: delayed at compiler/rustc_trait_selection/src/traits/query/normalize.rs:288:30 - disabled backtrace

note: we would appreciate a bug report: https://github.com/rust-lang/rust/issues/new?labels=C-bug%2C+I-ICE%2C+T-compiler&template=ice.md

note: rustc 1.89.0 (29483883e 2025-08-04) running on x86_64-unknown-linux-gnu

note: compiler flags: --crate-type lib -C embed-bitcode=no -C codegen-units=1 -C debuginfo=2

note: some of the compiler flags provided by cargo are hidden

query stack during panic:
end of query stack

@rustbot label A-impossible-bounds T-types regression-from-stable-to-stable

Regression in nightly-2024-11-27

This can't be. I've just tried nightly-2024-11-01 (to see what the output used to look like) and it also reproduces.
Maybe it's flip-flopping?

Edit: As a matter of fact, the regression didn't occur in nightly-2024-11-27; nightly-2024-11-26, nightly-2024-11-27, nightly-2024-11-28 all ICE for me (in check or build). I'm not sure what your bisect-rustc invocation looked like.

Smaller (no const blocks, etc.):

trait Trait { type Assoc; }

fn conjure<T>() -> T { loop {} }

fn get_assoc<T: Trait>() -> impl Sized { conjure::<T::Assoc>() }

fn foo<T: Trait<Assoc = i32> + Trait<Assoc = i64>>() { get_assoc::<T>(); }

Without conjure but with assoc fn:

trait Trait { type Assoc; fn make() -> Self::Assoc; }

fn get_assoc<T: Trait>() -> impl Sized { T::make() }

fn foo<T: Trait<Assoc = i32> + Trait<Assoc = i64>>() { get_assoc::<T>(); }

Since cargo-bisect-rustc is acting up for me for some reason (maybe my commit range is too large?), I'll be manually bisecting for a bit.

Local lower bound (PASS): 1.63 (2022-08-08)
Local upper bound (ICE): 1.64 (2022-09-19)

searched toolchains nightly-2022-06-24 through nightly-2022-08-05

Regression in nightly-2022-07-16

looking for regression commit between 2022-07-15 and 2022-07-16
fetching (via remote github) commits from max(c2f428d, 2022-07-13) to 23e21bd
ending github query because we found starting sha: c2f428d
get_commits_between returning commits, len: 9
commit[0] 2022-07-14: Auto merge of #99252 - lqd:win-dwarf5, r=eddyb
commit[1] 2022-07-14: Auto merge of #96544 - m-ysk:feature/issue-96358, r=cjgillot
commit[2] 2022-07-15: Auto merge of #98203 - kckeiks:gather-body-owners-in-hir-item-queries, r=cjgillot
commit[3] 2022-07-15: Auto merge of #97406 - aliemjay:verify-bounds-fix-master, r=oli-obk
commit[4] 2022-07-15: Auto merge of #99267 - Dylan-DPC:rollup-d8l7y1o, r=Dylan-DPC
commit[5] 2022-07-15: Auto merge of #99013 - RalfJung:dont-poison-my-places, r=oli-obk
commit[6] 2022-07-15: Auto merge of #99278 - Dylan-DPC:rollup-fcln6st, r=Dylan-DPC
commit[7] 2022-07-15: Auto merge of #99046 - nnethercote:final-derive-output-improvements, r=Mark-Simulacrum
commit[8] 2022-07-15: Auto merge of #99015 - lcnr:fn-ctxt-no-more, r=compiler-errors

Maybe #97406 or #99015?

definitely #99015 😁 see 7808f69

I would assume we've already got different issues where this ICEs? The underlying bug is older, we should either already error during HIR typeck or not normalize assuming there won't be errors afterwards

trait Trait { type Assoc; fn make() -> Self::Assoc; }

fn get_assoc<T: Trait>() -> impl Sized { T::make() }

fn foo<T: Trait<Assoc = i32> + Trait<Assoc = i64>>() { get_assoc::<T>(); }

this makes sense to me. proving T: Trait is easy while normalizing <T as Trait>::Assoc fails with ambiguity (as it should) 😁

The reveal all pass shouldn't expect things to succeed if we're in an impossible env. I think we've already got stuff there which replaces MIR bodies with nothing if the env has any impossible where-clauses?

It's just not possible to detect "is the given set of where-clauses inconsistent" in general 🤔

@rustbot

Assigning priority (discussion on Zulip).

@rustbot label -I-prioritize +P-medium

A version of this which has actually-possible bounds (it ICEs on the old solver, but compiles with next-solver):

trait Trait {
    type Assoc;
    fn make() -> Self::Assoc;
}

fn get_assoc<T: Trait>() -> impl Sized {
    T::make()
}

fn foo<T: Trait<Assoc = U> + Trait<Assoc = V>, U, V>() {
    get_assoc::<T>();
}

struct Thing;
impl Trait for Thing {
    type Assoc = i32;
    fn make() -> i32 { 0 }
}

fn main() {
    foo::<Thing, i32, i32>();
}

A version that has actually-possible bounds, and ICEs on both the old solver and next-solver:

trait Trait {
    type Assoc: Copy;
}

const fn get_assoc<T: Trait>() -> Option<impl Copy> {
    None::<T::Assoc>
}

fn foo<T: Trait<Assoc = U> + Trait<Assoc = V>, U, V>() {
    const { get_assoc::<T>(); }
}

struct Thing;
impl Trait for Thing {
    type Assoc = i32;
}

fn main() {
    foo::<Thing, i32, i32>();
}

Navigation Menu

Repository navigation

"unexpected ambiguity: `CanonicalQueryInput`" with conflicting associated type specifiers #146371

Code

Meta

Error output

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

"unexpected ambiguity: CanonicalQueryInput" with conflicting associated type specifiers #146371

Description

Code

Meta

Error output

Activity

fmease commented on Sep 9, 2025

fmease commented on Sep 9, 2025

fmease commented on Sep 9, 2025

fmease commented on Sep 10, 2025

fmease commented on Sep 10, 2025

lcnr commented on Sep 11, 2025

apiraino commented on Sep 19, 2025

theemathas commented on Mar 21, 2026

theemathas commented on Mar 21, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

"unexpected ambiguity: `CanonicalQueryInput`" with conflicting associated type specifiers #146371