Diagnostics regressed with 1.92.0

[weiznich]

Code

I tried this code:

use diesel::*; // 2.3.4, features = ["postgres", "sqlite"] 

table! {
    posts {
        id -> Integer,
        user_id -> Integer,
        hair_color -> Text,
    }
}

fn main() {
    let mut conn = PgConnection::establish("").unwrap();

    let _ = posts::table
        .group_by(posts::user_id)
        .distinct_on(posts::id)
        .select(posts::user_id)
        .get_results::<i32>(&mut conn);

    println!("Hello, world!");
}

I expected to see this happen: Code emits a clean helpful error message like on 1.91.0:

error[E0277]: the trait bound `SelectStatement<..., ..., ..., ..., ..., ..., ...>: DistinctOnDsl<_>` is not satisfied
  --> src/main.rs:18:10
   |
18 |         .distinct_on(posts::id)
   |          ^^^^^^^^^^^ the trait `DistinctOnDsl<_>` is not implemented for `SelectStatement<FromClause<table>, ..., ..., ..., ..., ..., ...>`
   |
   = note: a `DISTINCT ON` clause is not compatible with various other clauses like `LOCKING` clauses
   = note: a `DISTINCT ON` clause also disallows mixing aggregate and non-aggregate expressions with the `SELECT` clause
   = help: the trait `DistinctOnDsl<Selection>` is implemented for `SelectStatement<FromClause<F>, S, D, W, O, LOf, G, H>`
   = note: the full name for the type has been written to '/tmp/diesel_testing/target-bisector-nightly-2025-09-13-x86_64-unknown-linux-gnu/debug/deps/diesel_testing-a6733609d42d5652.long-type-11794557765803509610.txt'
   = note: consider using `--verbose` to print the full type name to the console

Instead, this happened: Code emits a second non-helpful could not infer type error:

error[E0277]: the trait bound `SelectStatement<..., ..., ..., ..., ..., ..., ...>: DistinctOnDsl<_>` is not satisfied
  --> src/main.rs:18:10
   |
18 |         .distinct_on(posts::id)
   |          ^^^^^^^^^^^ the trait `DistinctOnDsl<_>` is not implemented for `SelectStatement<FromClause<table>, ..., ..., ..., ..., ..., ...>`
   |
   = note: a `DISTINCT ON` clause is not compatible with various other clauses like `LOCKING` clauses
   = note: a `DISTINCT ON` clause also disallows mixing aggregate and non-aggregate expressions with the `SELECT` clause
   = help: the trait `DistinctOnDsl<Selection>` is implemented for `SelectStatement<FromClause<F>, S, D, W, O, LOf, G, H>`
   = note: the full name for the type has been written to '/tmp/diesel_testing/target-bisector-nightly-2025-10-24-x86_64-unknown-linux-gnu/debug/deps/diesel_testing-a6733609d42d5652.long-type-11312178195105605609.txt'
   = note: consider using `--verbose` to print the full type name to the console

error[E0282]: type annotations needed
  --> src/main.rs:16:13
   |
16 |       let _ = posts::table
   |  _____________^
17 | |         .group_by(posts::user_id)
18 | |         .distinct_on(posts::id)
   | |_______________________________^ cannot infer type

To be clear here: It's a regression in what users see, but it's not critically bad. It would be nicer to not have the second error, but if that's not possible for whatever reasons it might be OK to live with it.

Version it worked on

It most recently worked on: 1.91.0

Version with regression

rustc --version --verbose:

rustc 1.92.0 (ded5c06cf 2025-12-08)
binary: rustc
commit-hash: ded5c06cf21d2b93bffd5d884aa6e96934ee4234
commit-date: 2025-12-08
host: x86_64-unknown-linux-gnu
release: 1.92.0
LLVM version: 21.1.3

cargo bisect-rustc points to 54a8a1d, so this was introduced in #146885. (cc @lcnr)

@rustbot modify labels: +regression-from-stable-to-stable -regression-untriaged

[lcnr]

the commit which caused this is 6b379b5, see commit description

Unlike structurally_resolve_type, probe_op does not check whether
the infcx is already tainted, so this results in 2 previously not emitted
errors.

imo we should and can start to more strongly rely on checking whether the infcx is already tainted before emitting the second ambiguity error

[apiraino]

Assigning priority (discussion on Zulip).

@rustbot label -I-prioritize +P-medium