Regression from 1.44 to 1.45

[mcarton]

The following MVE prints 42 with rustc 1.45.0 but 13 with rustc 1.44.0

struct Foo {
    x: i32,
}

fn main() {
    let mut foo = Foo { x: 42 };
    let x = &mut foo.x;
    *x = 13;
    let y = foo;
    println!("{}", y.x); // -> 42; expected result: 13
}

Initially reported on Stack Overflow.

[mcarton]

With beta or nightly this again prints 13 as expected.

struct Foo {
    x: i32,
}

fn main() {
    let mut foo = Foo { x: 42 };

    let x = &mut foo.x;
    *x = 13;
    let y = foo;
    println!("{:?}", (&y).x);  //only added this line
    println!("{:?}", y.x); //13
}

If there is any benefit, it works as expected with this code.

[lcnr]

This is a duplicate of the already fixed #73609

It seems like this bug somehow slipped into beta :/

[SimonSapin]

@rust-lang/release Should there be a 1.45.1 with #73609?

[peterjoel]

@SimonSapin Can 1.45.0 be yanked?

[jonas-schievink]

Yanking a release is not a thing. Presumably it wouldn't do anything once the stable channel is updated anyways.

[peterjoel]

@jonas-schievink It would do something for anyone who typed rustup install 1.45.0.

[benschulz]

I'm confused. Based on the examples this bug appears to be very impactful, i.e. it should break a lot of code. Yet it made it to stable which, I assume, involves a crater run and independent testing on beta. Is there a subtle condition to this bug that isn't met in most productive code?

[oli-obk]

The bug is almost impossible to trigger on real world code. You need all values that are going into the bug to be constant values and there can't be any control flow or function calls in between.