apply query response: actually define opaque types

[lcnr]

not sure whether this fixes any code considering that #107891 doesn't break anything, but this is currently wrong as the eq there should just always fail right now.

We can definitely hit this code if we remove the replace_opaque_types_with_inference_vars hack. Doing so without this PR causes a few tests to ICE, e.g.

rust/tests/ui/impl-trait/issue-99642.rs

Lines 1 to 7 in bd4a96a

	// check-pass
	fn test() -> impl Iterator<Item = impl Sized> {
	Box::new(0..) as Box<dyn Iterator<Item = _>>
	}
	fn main() {}

r? @oli-obk

[oli-obk]

@bors r+

Ah that explains it. We used to not run the infer var replacement for TAITs, so it worked before. But once I flipped the default to "not infer opaque types", this "regressed". I do wonder why we use the evaluate queries so little in typeck, but apparently do so for the return type

@bors r+ rollup

[bors]

📌 Commit cff59f8 has been approved by oli-obk

It is now in the queue for this repository.

[bors]

💡 This pull request was already approved, no need to approve it again.

• There's another pull request that is currently being tested, blocking this pull request: Rollup of 8 pull requests #108339

[bors]

📌 Commit cff59f8 has been approved by oli-obk

It is now in the queue for this repository.

[lcnr]

I do wonder why we use the evaluate queries so little in typeck, but apparently do so for the return type

we only do it during mir typeck, hir typeck never uses canonical queries atm. I expect that by using the inference var replacement for opaques in the return type and so on, we tend to not have opaque types in the types of locals or sth 🤔