Accessing &mut inside of @mut requires first borrowing the @mut

[nikomatsakis]

Right now, if you have an &mut pointer inside of an @mut data structure, you cannot access it without first borrowing the @mut to &mut:

struct Foo { x: &mut int }
fn foo(m: @mut Foo) {
    *m.x += 1; // ERROR

    let n = &mut *m; // borrow m as &mut
    *n.x += 1; // OK
}

The reason for this is that we require a unique access path to mutate or re-borrow &mut pointers, and @mut is aliasable. Once the @mut is borrowed to &mut, though, we know that no mutable aliases exist, because otherwise the borrow would have failed. We could just however accept the original code and insert a write guard check ourselves. I don't do this now because it would require various subtle bits of code to make sure that the automatically inserted borrow has the appropriate lifetime to guarantee soundness. Still, it's kind of annoying.

cc #5074

[graydon]

Nominating for well-defined. Though I think this is absolutely fine as-is. If anything I still sorta feel like & doesn't belong inside @ at all. But that's a different kettle of fish.

[nikomatsakis]

I thought we had decided on (and possibly even implemented?) a rule to prohibit & from @, actually? @pcwalton didn't I see such a commit go by, or just prep commits?

[catamorphism]

Obsolete, closing