Don't do post-method-probe error reporting steps if we're in a suggestion

[compiler-errors]

Currently in method probing, if we fail to pick a method, then we reset and try to collect relevant candidates for method errors:

rust/compiler/rustc_hir_typeck/src/method/probe.rs

Lines 953 to 993 in 3458211

	// things failed, so lets look at all traits, for diagnostic purposes now:
	self.reset();
	let span = self.span;
	let tcx = self.tcx;
	self.assemble_extension_candidates_for_all_traits();
	let out_of_scope_traits = match self.pick_core() {
	Some(Ok(p)) => vec![p.item.container_id(self.tcx)],
	Some(Err(MethodError::Ambiguity(v))) => v
	.into_iter()
	.map(|source| match source {
	CandidateSource::Trait(id) => id,
	CandidateSource::Impl(impl_id) => match tcx.trait_id_of_impl(impl_id) {
	Some(id) => id,
	None => span_bug!(span, "found inherent method when looking at traits"),
	},
	})
	.collect(),
	Some(Err(MethodError::NoMatch(NoMatchData {
	out_of_scope_traits: others, ..
	}))) => {
	assert!(others.is_empty());
	vec![]
	}
	_ => vec![],
	};
	if let Some((kind, def_id)) = private_candidate {
	return Err(MethodError::PrivateMatch(kind, def_id, out_of_scope_traits));
	}
	let similar_candidate = self.probe_for_similar_candidate()?;
	Err(MethodError::NoMatch(NoMatchData {
	static_candidates,
	unsatisfied_predicates,
	out_of_scope_traits,
	similar_candidate,
	mode: self.mode,
	}))

However, we do method lookups via lookup_method_for_diagnostic and only care about the result if the method probe was a success.

Namely, we don't need to do a bunch of other lookups on failure, since we throw away these results anyways, such as an expensive call to:

rust/compiler/rustc_hir_typeck/src/method/probe.rs

Line 959 in 3458211

self.assemble_extension_candidates_for_all_traits();

And:

rust/compiler/rustc_hir_typeck/src/method/probe.rs

Line 985 in 3458211

let similar_candidate = self.probe_for_similar_candidate()?;

---

This PR also renames some methods so it's clear that they're for diagnostics.

r? @nnethercote

[nnethercote]

I find the for_diagnostic suffix naming convention confusing. I think it means "method exists (as considered for diagnostic purposes)", rather than "method exists for this diagnostic"? It feels like there should be a contrasting method_exists_for_non_diagnostic method. I'm just not sure how to read it.

[compiler-errors]

Yes, it means "and you should only call this method if you're doing diagnostics".

[compiler-errors]

Kinda don't want to change it though now lol

[nnethercote]

Ok, I can live with it.

[nnethercote]

r=me once we consider the for_diagnostic suffix. Is it necessary? I'm having trouble thinking of a better alternative.

[compiler-errors]

Is it necessary?

Well I'd like to make it more difficult for people to call methods that aren't correct on the good path. It's not necessary per se, but we are really bad at modularizing diagnostics code in rustc_hir_typeck. If and once we finally do that, then it'll be self-explanatory due to (hopefully) mod privacy, but until then, I'd like to keep the suffix or give it a similar name that makes it sure the user knows that this obtusely named method is not meant for the good path.

[compiler-errors]

This regressed in #120730, which added new calls to assemble_extension_candidates_for_traits_in_scope which significant increases the amount of work we are doing when calls to (e.g.) lookup_method_for_diagnostic, etc fail. Since this is on the error path, we call this a lot more times than on the good path.

Before 1.78, this file still takes a few seconds. After 1.78, it essentially hangs. After this PR, it should take less than a second.

[compiler-errors]

Beta nominating as it can pretty negatively affect user experience when a programmer has many method errors in large projects.

For example: Slow edit_distance on Zulip

[nnethercote]

Excellent that you have a test case! Please add a brief comment explaining what this is testing and pointing to this PR.

[nnethercote]

r=me with the comment on the test. Thanks for fixing this!

[compiler-errors]

@bors r=nnethercote

[bors]

📌 Commit 8f97a25 has been approved by nnethercote

It is now in the queue for this repository.

[apiraino]

Beta backport accepted as per compiler team on Zulip. A backport PR will be authored by the release team at the end of the current development cycle.

@rustbot label +beta-accepted