Inconsistent evaluation order for assignment operations

[eefriedman]

UPDATE: This is fixed by the MIR-based borrow checker and just needs a test. See the mentoring instructions below.

---

Currently, the borrow checker thinks that the RHS of += is evaluated before the LHS. trans thinks that the LHS is evaluated before the RHS. The disagreement leads to bad results.

fn main() {
    let x = Box::new(0);
    let mut y = 0;
    *{ drop(x); &mut y } += *x;
    assert_eq!(y, 0);
}

[nrc]

triage: P-medium

Also fixed by MIR (almost a dup of #28160, but not quite)/

[nikomatsakis]

See this internals thread: https://internals.rust-lang.org/t/settling-execution-order-for/4253

[Mark-Simulacrum]

So the internals thread says this: "before MIR, we were actually quite inconsistent here, in that borrowck and trans disagreed on some of the particulars. MIR eliminated that inconsistency, but there are still some matters that we should try to resolve -- and quick!" If that's the case, then this should be closed. @nikomatsakis Is this the case?

[nikomatsakis]

Well, we're not yet running the borrowck on MIR, so I'm not sure.

[nikomatsakis]

But certainly the example seems to work ok now.