is it UB or not if using cell's underlying data raw ptr after cell is moved

[gftea]

rust playground

I am a bit confused about the "Cell::as_ptr()", user can escape the compiler check, and may still use the raw pointer after the cell is moved.

#[derive(Debug)]
struct Inner(isize);

impl Drop for Inner {
    fn drop(&mut self) {
        println!("drop inner {:?}", self as *const _);
    }
}

let m = Cell::new(Inner(0));
let m_ptr = m.as_ptr();

println!("before move: {:?}", m.as_ptr());

let h = thread::spawn(move || {
    // the `m` cell underlying data has different address after move
    println!("after move: {:?}", m.as_ptr());
    // `m` drop
});

// we can still read/write the original address after moving `m`
// should it be UB? 
// if yes, miri does not report it
// If not, what does the address really means?
unsafe {println!("m_ptr: {:?}, value: {:?}",  m_ptr, *m_ptr)};

[oli-obk]

Yea. We don't invalidate memory on move. Though maybe we should invalidate borrows to moved out memory?

[saethlin]

This is rust-lang/unsafe-code-guidelines#188

[RalfJung]

So the answer to "is this UB" is - maybe. It hasn't been decided yet, and it is not clear how to best make Miri detect these csses. @oli-obk, that would be possible (thougj not easy) with stacked borrows but with the plans we have for raw pointers in tree borrows it becomes a lot harder at best.

[RalfJung]

This is a UCG / t-opsem question, not a Miri question, so closing in favor of rust-lang/unsafe-code-guidelines#188.