Add option to flag uninitialized integers as UB

[RalfJung]

The reference currently states that this code is UB, but Miri deliberately accepts it:

fn main() {
    let _val = unsafe { std::mem::MaybeUninit::<usize>::uninit().assume_init() };
}

The reason we accept it is that the lang-team is not sure if we really want this to be UB (also see rust-lang/unsafe-code-guidelines#71), so the conservative choice for the reference is to make it UB for now, but in Miri I fear this might lead to too many errors that people could consider false positives (even though by the letter of the reference they are true positives).

Still it could be interesting to see how much code (that Miri can run) actually fails when considering uninitialized integers UB, so having a flag in Miri to enable stricter checking would be interesting.

[oli-obk]

Maybe this is something where we could just emit a warning (silenceable of course) by default?

[RalfJung]

Yeah that also sounds good... but then we have three levels, allow/warn/deny, so it really seems like we should use the lint system.^^

That will also be much harder though, I imagined just having a machine hook that affects whether validation checks integers for being initialized (instead of doing that based on whether we are in "const mode").

[oli-obk]

seems good, too. If we make it configurable in miri via the command line, users can turn it off if it bothers them (or we start with opt-in)

[beepster4096]

Looks like this is the place where the check would go

[RalfJung]

Yes that is correct.

[RalfJung]

While at it we should probably also mark ptr-to-int transmutes as UB (Cc rust-lang/unsafe-code-guidelines#286)

[camelid]

I started working on this, but when I try to use my locally-built rustc with Miri, I get a strange error:

error[E0463]: can't find crate for `rustc_apfloat`
  --> src/lib.rs:10:1
   |
10 | extern crate rustc_apfloat;
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^ can't find crate

For more information about this error, try `rustc --explain E0463`.

When I built rustc, it said that it built rustc_apfloat, so I'm not sure what's going on. What should I do to fix this?

[oli-obk]

I don't know how to work with a locally built miri except by doing ./x.py test src/tools/miri

It may be that we just don't properly support going the rustup toolchain link route

[RalfJung]

@camelid have you followed the instructions at https://github.com/rust-lang/miri/blob/master/CONTRIBUTING.md#building-miri-with-a-locally-built-rustc ? Those are what I used for locally built rustc and they always worked for me (but it's been a few weeks sine I last tried this).

[RalfJung]

Hm, actually I am always using stage 2, not stage 1. But you are the one who changed the instructions to recommend stage 1 instead of stage 2... are you saying that does not work any more? Should we revert bb59980?

[camelid]

Hm, actually I am always using stage 2, not stage 1. But you are the one who changed the instructions to recommend stage 1 instead of stage 2... are you saying that does not work any more? Should we revert bb59980?

Yes, I'm almost done with a stage2 build to see if that fixes it. If that works, I was planning to also suggest we revert that commit :)