small uninitialized types are unsound

[RustyYato]

Similar to #3, but it can't be fixed by zeroing beforehand

struct Thing(MaybeUninit<usize>);
fn main() {
    assert_eq!(size_of::<Thing>(), size_of::<*mut Thing>());
    let thing = Thing(MaybeUninit::uninit());
    let stowed = Stowaway::new(thing); // miri fails here, encountered uninitialized bytes
}

[Lucretiel]

Can you create a PR with a test case that fails in miri so that I can investigate further and protect against future regressions?

[RustyYato]

Sure

[Lucretiel]

FWIW, it appears that the issue isn't that MIRI doesn't copy the uninitialized bytes, but that it doesn't consider filling a uninitialized int with them as UB until the int is "used" in some way. However, this is considered a bug in MIRI (rust-lang/miri#1340), so the unsoundness is still unsound. I'll be publishing a PR later today and I'd love if you could review it.

[RustyYato]

Cool, I wasn't aware of that! Thanks for the link.

[Lucretiel]

There's also a lot of good discussion in the answer & comments of the stackoverflow question I posted on this topic, especially related to the current progress in defining the soundness rules for this sort of thing.

[Lucretiel]

Published 2.0.0