Yank 2024-02-12 (possibly unsound impl valrow::Borrowable for Box<_>)

[MaulingMonkey]

Concerns were raised about Box's Unique / noalias requirements. Miri looks like it should be able to catch such bugs, but I haven't been able to convince my copy to catch any bugs in this test.

Testing

rustup toolchain install nightly -c miri

set RUSTFLAGS=--cfg xxx_borrowable_box
cargo +nightly miri test --all-features

set MIRIFLAGS=-Zmiri-unique-is-unique -Zmiri-tree-borrows
cargo +nightly miri test --all-features

References

• https://stdrs.dev/nightly/x86_64-unknown-linux-gnu/core/ptr/unique/struct.Unique.html
• What are the uniqueness guarantees of Box and Vec? rust-lang/unsafe-code-guidelines#326
• https://github.com/rust-lang/miri/

valrow/src/borrowable.rs

Lines 143 to 167 in b38c862

	#[cfg(xxx_borrowable_box)]
	#[cfg(feature = "alloc")] #[test] fn try_to_break_box_valrows() {
	let a = alloc::boxed::Box::new(core::cell::Cell::new(42));
	let b = crate::Valrow::new(&a);
	let c = &a;
	a.set(1);
	b.set(2); // possibly a problem?
	c.set(3);
	a.set(4);
	b.set(5); // possibly a problem?
	c.set(6);
	let fmt = alloc::format!("{:?}", (&a, b, c));
	#[cfg(feature = "std")] std::println!("{fmt}");
	// Maybe the temp-Deref s aren't a problem, but would having a persistent pair of different-address `&Box<Cell<_>>`s trigger miri?
	let b : &alloc::boxed::Box<_> = &*b;
	a.set( 7);
	b.set( 8); // possibly a problem?
	c.set( 9);
	a.set(10);
	b.set(11); // possibly a problem?
	c.set(12);
	let fmt = alloc::format!("{:?}", (&a, b, c));
	#[cfg(feature = "std")] std::println!("{fmt}");
	}

[MaulingMonkey]

• a3fb304
• b38c862
• ffa5340