Rustc says method doesn't exist instead of hinting at unsatisfied trait bounds

[Kixunil]

As described in #67642 (comment) when one attempts to call a triat method with unsatisfied bounds, the message says the method doesn't exist instead of explaining that the bounds weren't satisfied.

I tried this code:

use std::collections::HashMap;

struct Foo;

fn main() {
    let a: HashMap<Foo, String> = HashMap::default();
    let mut b: HashMap<Foo, String> = HashMap::default();
    b.extend(a);
}

I expected to see this happen: Rustc message saying that Eq and Hash aren't satisfied.

Instead, this happened: Rustc says that extend doesn't exist.

Meta

The bug is in nightly

Tested on the playground:

 1.44.0-nightly (2020-04-22 b2e36e6c2d229126b59e)

[estebank]

Current output:

error[E0599]: no method named `extend` found for struct `HashMap` in the current scope
 --> src/main.rs:8:7
  |
8 |     b.extend(a);
  |       ^^^^^^ method not found in `HashMap<Foo, String>`

[estebank]

This was an attempt at exploring what is needed to track this information correctly.

@lcnr / @compiler-errors (question about trait solver): I noticed that we currently don't track the elaborated obligations' failures beyond the signal EvaluatedToErr. Would it make sense to make that a signaling error that collects all the predicates that failed, in order to be able to present information on E0599 when only some of the trait bounds failed for a method that otherwise should be present?

[lcnr]

I am hesitant about this. How does your desired behavior differ from the behavior of using a fullfillment context/ObligationCtxt? I think we should not report errors for failures when using evaluation but instead get a FulfillmentError.

If you prove Vec<T>: Clone via a fulfillment context you already get an error talking about T: Clone.