Evaluate Unevaluated const inside Const::Ty in valtree conversion

[cijiugechu]

Handle ConstKind::Unevaluated inside Const::Ty.

Closes #156410

[rustbot]

r? @folkertdev

rustbot has assigned @folkertdev.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: compiler
• compiler expanded to 73 candidates
• Random selection from 18 candidates

[folkertdev]

This looks OK to me given that it's only used for better diagnostics right now, but probably someone from wg-const-eval should look at it.

r? wg-const-eval

[rust-bors]

☔ The latest upstream changes (presumably #157094) made this pull request unmergeable. Please resolve the merge conflicts.

[lcnr]

r? @BoxyUwU

I feel like we vaguely shouldn't have unevaluated type system consts in the MIR body, but maybe Boxy has a better idea of what's going on

[rustbot]

BoxyUwU is currently at their maximum review capacity.
They may take a while to respond.

[BoxyUwU]

we should in theory for have unevaluateds in MIR bodies but they should all be rigid (i.e. it should not be possible to further normalize/evaluate them).

So I think it definitely makes sense to not be ICEing in rustc_mir_build here, but instead of calling const_eval_resolve we should use normalize erasing regions or something to that affect so we get actual type system normalization of our type system consts. I would expect this PR causes us to ICE on paths to free type consts as they have no body so const_eval_resolve will complain.

The rustc_codegen_ssa case I don't understand. If we're in codegen then shouldn't the MIR body be fully concrete (i.e. no more generic parameters) and as such all Unevaluateds have already been normalized to Values 🤔 Can you look into what's going on in the rustc_codegen_ssa ICE case and then explain what's going on there :3

If it does turn out that it's expected to get unevaluated consts there then the same thing about const_eval_resolve will need doing and we should be using normalize erasing regions or sth instead for the Const::Ty(ConstKind::Unevaluated case

[BoxyUwU]

@rustbot author