Should Miri report MIR validation failures?

[cbeuw]

rustc contains a MIR validation pass which detects ill-formed MIR. Currently Miri does not report MIR validation failures, but should it?

For instance, in runtime MIR, Deref must the the first projection if one exists. This code (rust-lang/rust#110228) performs a dereference after a field projection, so it fails MIR validation, but Miri runs this without any complaint

#![feature(custom_mir, core_intrinsics)]
extern crate core;
use core::intrinsics::mir::*;
#[custom_mir(dialect = "runtime", phase = "optimized")]
pub fn fn0() {
    mir! {
        let x: i32;
        let tuple: (*mut i32,);
        {
            tuple.0 = core::ptr::addr_of_mut!(x);
            *(tuple.0) = 1;
            Return()
        }
    }
}
pub fn main() {
    fn0();
}

[RalfJung]

I am not entirely sure why that validation pass is skipped in Miri. @oli-obk is this some accident in MIR pass management?

[oli-obk]

That's really odd... I looked into it and it makes no sense. We fetch optimized_mir, which must run a validation pass. I'll need to actually compile a rustc and miri with debug asserts and look at some logs

[matthiaskrgr]

I had a look since I already have a rustc+miri with debug assertions laying around and I didn't see anything interesting 🤔

MIRIFLAGS="-Zvalidate-mir" RUSTFLAGS="-Zvalidate-mir" ~/.cargo/bin/cargo +local-debug-assertions miri run

Preparing a sysroot for Miri (target: x86_64-unknown-linux-gnu)... done
WARNING: Ignoring `RUSTC_WRAPPER` environment variable, Miri does not support wrapping.
    Finished dev [unoptimized + debuginfo] target(s) in 0.05s
     Running `/home/matthias/.rustup/toolchains/local-debug-assertions/bin/cargo-miri runner target/miri/x86_64-unknown-linux-gnu/debug/miritest`

[RalfJung]

If I try this with today's Miri, I do get the expected ICE.