Introduce stricter checks for might_permit_raw_init under a debug flag

[5225225]

This is intended to be a version of the strict checks tried out in #79296, but also checking number validity (under the assumption that let _ = std::mem::uninitialized::<u32>() is UB, which seems to be what rust-lang/unsafe-code-guidelines#71 is leaning towards.)

[rust-highfive]

Some changes occured to rustc_codegen_cranelift

cc @bjorn3

Some changes occured to the CTFE / Miri engine

cc @rust-lang/miri

[rust-highfive]

r? @davidtwco

(rust-highfive has picked a reviewer for you, use r? to override)

[5225225]

I'll leave the variant checks for future work.

Was initially going to try and get them in now (so it's not as much breakage to add them later), but I figure adding more panics on UB when you're using a debugging flag specifically to panic on UB is pretty justifiable.

That, and I want to get the infrastructure in here sooner than later. Will clean up this PR to just do the array and primitive checks.

[5225225]

@rustbot ready

[oli-obk]

somewhat annoying to pass this everywhere, when self is also passed. Though HasDataLayout doesn't have a useful way to get at this.

maybe add a field to TargetDataLayout for that bool?

Idk, doesn't seem great either. I guess let's roll with the PR as it is.

[oli-obk]

@bors r+

[bors]

📌 Commit d59cc92d7a30f0b4b7d9e192ca0e165a306f61ef has been approved by oli-obk

[oli-obk]

davidtwco approved these changes now

oh lol, apologies, that's funny timing

[RalfJung]

Cc #66151

[RalfJung]

@bors r-

We're not in a rush for this feature I think, so let's fix that nit before landing. :)

[RalfJung]

And also thanks a lot @5225225 for doing this, I quite like the idea of opting-in to the more strict checks. :D

FWIW, as @oli-obk proposed once, there is an elegant way to get the "fully strict" check but it requires more work and no code sharing with the non-strict case: fire up an InterpCx and create an actual zeroed/uninit allocation, and then ask the validity checker in the interpreter whether this is valid for the given type. That will handle all types properly; it is the same logic that Miri uses to detect incorrect values. So this might be better than trying something ad-hoc for multi-variant types.
If that peeks your interest, feel free to ask about details on Zulip. :D

[5225225]

Okay, I've fixed the formatting in the cranelift code, moved the zero: bool to a valid_as: ValidAs enum (hope you don't mind, but I was touching this code anyways :) ), and removed the redundant parameter on is_uninit_valid.

And yeah, the "actually make an allocation and then ask the interpreter if it's valid" seems like the cleanest solution.

[bjorn3]

cg_clif changes LGTM

[oli-obk]

@bors r+

[bors]

📌 Commit dd9f31d has been approved by oli-obk