Binding subpattern with @ allows for illegal mutation

[Marwes]

fn main() {
    match &mut Some(1) {
        ref mut z @ &Some(ref a) => {
            **z = None;
            println!("{}", *a);
        }
        _ => ()
    }
}

This code compiles even though it should be rejected since writing to z invalidates the a variable.

[alexcrichton]

That sounds quite bad, nominating.

[pnkfelix]

P-backcompat-lang, 1.0.

[nikomatsakis]

cc me

[nikomatsakis]

Ah, I think I know why this happens. Fascinating. It's no doubt because the matched expression is an rvalue. I think we'll have to upgrade how rvalues are handled in borrowck to fix this.

[alexcrichton]

cc #12534

[nikomatsakis]

Basically the problem is that we treat rvalues as though they are only ever once -- but with a devious match expression like this one you can arrange to consume them / borrow them multiple times. I think we ought to modify the memcategorization to store the expr-id of the rvalue and not just the temporary scope -- the borrow checker can then treat rvalues just as it treats lvalues. Basically we want to have the borrow checker reason about the temporary that will be created.