Fix ICEs on invalid vtable size/alignment const UB errors

[lqd]

The invalid vtable size/alignment errors from InterpCx::read_size_and_align_from_vtable were "freeform const UB errors", causing ICEs when reaching validation. This PR turns them into const UB hard errors to catch them during validation and avoid that.

Fixes #86193

r? @RalfJung

(It seemed cleaner to have 2 variants but they can be merged into one variant with a message payload if you prefer that ?)

[rust-highfive]

Some changes occured to the CTFE / Miri engine

cc @rust-lang/miri

Some changes occured to the CTFE / Miri engine

cc @rust-lang/miri

[RalfJung]

I feel like InvalidVtable{Size,Alignment,DropFn} would be clearer names for these variants. What do you think?

[lqd]

sure, I'll do that ! I initially did name those specifically for vtables but then InvalidDropFn "convinced me" not to do so ^^

[RalfJung]

Suggested change

	// ICEs as tracked by #86193. So we also use the transparent wrapper to verify actual const UB hard
	// errors are emitted now instead of ICEs.
	// ICEs as tracked by #86193. So we also use the transparent wrapper to verify proper validation
	// errors are emitted instead of ICEs.

[RalfJung]

@oli-obk reading this, I feel like we should reorder these error messages a bit, and make them more like

type validation failed at .0: encountered invalid vtable: size is bigger than largest supported object

What do you think? (@lqd certainly not for this PR though)

[lqd]

if you and oli want this, I can certainly do that in this PR

[RalfJung]

This will affect many tests, so it should be a separate PR. Reviewing is also much easier if we keep PRs small. But sure, if you want to help with this that'd be great. :)

[lqd]

Agreed, I'll do that in another PR.

Would something like this work for you ? Changing ValidationFailure's content from a String message to a path: Option<String>, mesg: String so that

rust/compiler/rustc_mir/src/interpret/validity.rs

Lines 27 to 47 in 3e827cc

	macro_rules! throw_validation_failure {
	($where:expr, { $( $what_fmt:expr ),+ } $( expected { $( $expected_fmt:expr ),+ } )?) => {{
	let msg = rustc_middle::ty::print::with_no_trimmed_paths(|| {
	let mut msg = String::new();
	msg.push_str("encountered ");
	write!(&mut msg, $($what_fmt),+).unwrap();
	let where_ = &$where;
	if !where_.is_empty() {
	msg.push_str(" at ");
	write_path(&mut msg, where_);
	}
	$(
	msg.push_str(", but expected ");
	write!(&mut msg, $($expected_fmt),+).unwrap();
	)?
	msg
	});
	throw_ub!(ValidationFailure(msg))
	}};
	}

can handle the empty "where", and the formatting you want can be implemented at

rust/compiler/rustc_middle/src/mir/interpret/error.rs

Line 326 in 2779fc1

ValidationFailure(ref err) => write!(f, "type validation failed: {}", err),

---

(I don't know when I'll get to this though: I'm currently having trouble executing some 32bits tests of the master branch, which fail with syntax errors when executed in the test runner but not by themselves, making blessing tests rather ... difficult. Thankfully it did not happen on the test in this PR 😓 )

[RalfJung]

Let's discuss the implementation details elsewhere (possibly on Zulip), if/when @oli-obk agrees that we even want to do this. :)

[oli-obk]

I agree that the proposed ordering is better than the current one

[RalfJung]

This looks great, thanks a lot. :-)
@bors r+

[bors]

📌 Commit 9c7133c95bdfaf3b14798f61d40edeb1564b2ed3 has been approved by RalfJung

[RalfJung]

@bors r-

[lqd]

CI is green after actually calling rustfmt...

@bors r=RalfJung

[bors]

📌 Commit e29f3e8 has been approved by RalfJung

[bors]

⌛ Testing commit e29f3e8 with merge fb3ea63...

[bors]

☀️ Test successful - checks-actions
Approved by: RalfJung
Pushing fb3ea63 to master...