Exhaustiveness checking: work around type normalization issues

[Nadrieril]

This should resolve #72476 and probably #72467.
This is a bit hacky but that's actually what the code was doing before #71930. I'm essentially reverting e5a2cd5. So despite being hacky, it's been tried and tested (so much so that code relies on it now x)).
Only the third commit does anything interesting.

[rust-highfive]

r? @davidtwco

(rust_highfive has picked a reviewer for you, use r? to override)

[LeSeulArtichaut]

cc @matthewjasper I think

[Nadrieril]

And cc @varkor who reviewed #71930

[davidtwco]

LGTM, probably should have someone more experienced look over it though.

[davidtwco]

r? @matthewjasper

[mark-i-m]

@Nadrieril Would it make sense to open an issue about properly fixing this?

[Nadrieril]

I guess yeah. I'm not super motivated to do it though, it would take some work to explain properly what the issue is

[matthewjasper]

The issue is that field.ty(..) doesn't normalize it's output after substituting. For example, in the test the field has type <T as A>::Projection, which after substituting becomes <() as A>::Projection. To get the normalized type you either have to explicitly normalize, or use the result from the type of the subpattern, which typeck has already normalized.

I think that I would prefer normalization to be used here, because it's more consistent with how things have to be done in general.

[Nadrieril]

I made this PR as an emergency fixup because the related issue was high priority. I'd still like to fix this properly, but I don't think adding normalization after field.ty would be enough. There's a bunch of different places that inspect types, and I don't understand the rationale and invariants related to type normalization. It's already the second uncaught problem surfaced by this change so I'd prefer to be safe and merge this first

[Nadrieril]

Since #72476 is P-critical, may I insist a bit to get this merged quickly?

[matthewjasper]

@bors r+

[bors]

📌 Commit 3e8ba3a has been approved by matthewjasper

[Nadrieril]

Thank you 🙏

[wesleywiser]

Per discussion in the compiler team triage meeting today, I'm going to bump the priority on this since it fixes a P-critical issue.

@bors p=1

[bors]

⌛ Testing commit 3e8ba3a with merge 7c0b6f8f94538a4818d099aaae778813a4d3303e...

[Dylan-DPC-zz]

@bors retry yield

[bors]

⌛ Testing commit 3e8ba3a with merge 052c781a6ae34fbd658d72226577c5cfb6be84b3...

[Dylan-DPC-zz]

@bors retry yield

[bors]

⌛ Testing commit 3e8ba3a with merge 9cd3f1c...

[bors]

☀️ Test successful - checks-azure
Approved by: matthewjasper
Pushing 9cd3f1c to master...

[Nadrieril]

The issue is that field.ty(..) doesn't normalize it's output after substituting.

Oh, I just understood this comment x) The error popped up again (in #89393) because I hadn't fixed it properly. I understand how to now.