Revert `compute_const` changes in WF

[varkor]

See https://github.com/rust-lang/rust/pull/56723/files/03b892860da5aff7a10cf4ca1364dbcacb95bfcd#r376641987. Opening this issue so I don't forget about it.

cc @eddyb

This issue has been assigned to @lcnr via this comment.

[eddyb]

I was trying something out to show @ruabmbua their idea can't work yet and accidentally came up with something that compiles even if it shouldn't. I'm pretty sure it's because of this bug.

[ruabmbua]

Extended example: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2018&gist=aa9afdff6a2f23a8f0cc588fdafe4c17

It would be nice, if is possible to build something like that. Even, if it`s not exactly the same.

[varkor]

Assigning to myself so I don't forget, but I'm not sure I'll be able to get to this very soon, so if someone else wants to take it, you're welcome to.

[eddyb]

WF is also missing something like this for ty::ConstKind::Infer:

rust/src/librustc_trait_selection/traits/wf.rs

Lines 556 to 593 in 3c6f982

	// Inference variables are the complicated case, since we don't
	// know what type they are. We do two things:
	//
	// 1. Check if they have been resolved, and if so proceed with
	// THAT type.
	// 2. If not, check whether this is the type that we
	// started with (ty0). In that case, we've made no
	// progress at all, so return false. Otherwise,
	// we've at least simplified things (i.e., we went
	// from `Vec<$0>: WF` to `$0: WF`, so we can
	// register a pending obligation and keep
	// moving. (Goal is that an "inductive hypothesis"
	// is satisfied to ensure termination.)
	ty::Infer(_) => {
	let ty = self.infcx.shallow_resolve(ty);
	if let ty::Infer(_) = ty.kind {
	// not yet resolved...
	if ty == ty0 {
	// ...this is the type we started from! no progress.
	return false;
	}
	let cause = self.cause(traits::MiscObligation);
	self.out.push(
	// ...not the type we started from, so we made progress.
	traits::Obligation::new(
	cause,
	self.param_env,
	ty::Predicate::WellFormed(ty),
	),
	);
	} else {
	// Yes, resolved, proceed with the
	// result. Should never return false because
	// `ty` is not a Infer.
	assert!(self.compute(ty));
	}
	}

Fixing this might require Predicate::WellFormedConst, which I thought would replace Predicate::ConstEvaluatable but now I'm less sure it would.
It shouldn't hurt to have both at once and we can deduplicate if possible later.

The other problem is this needs to add Predicate::WellFormedConsts:

rust/src/librustc_typeck/check/mod.rs

Lines 3341 to 3348 in 3c6f982

	/// Registers obligations that all types appearing in `substs` are well-formed.
	pub fn add_wf_bounds(&self, substs: SubstsRef<'tcx>, expr: &hir::Expr<'_>) {
	for ty in substs.types() {
	if !ty.references_error() {
	self.register_wf_obligation(ty, expr.span, traits::MiscObligation);
	}
	}
	}

[lcnr]

@rustbot claim