safe ref checker permits vectors to be freed as they are being iterated over

[nikomatsakis]

Running this program:

use std;
import std::io;

fn iter(&&v: [uint], f: fn(uint)) {
    for i in v { f(i) }
}

fn main(args: [str]) {
    let v = vec::enum_uints(0u, 10u);
    iter(v) { |i|
         v = [];
         io::print(#fmt["i=%?\n", i]);
    }
}

yields this output

i=0
i=2
i=171048973
i=1979117988610105
i=4122818071364057393
i=3978138846657458487
i=3979273541751617588
i=51
i=17293822612877934602
i=17293822569102704640
i=9

I expected a compile error.

[nikomatsakis]

It occurs to me that I may have misunderstood the guarantee that the "Safe ref" checker is providing (I am perpetually doing that, it seems). Perhaps it does not try to guarantee that no mutation occurs of the referenced value. In that case, I would expect a bounds check error.

[catamorphism]

This looks like an issue with the mutability checker, which appears to be mostly @marijnh 's work, so assigning to him.

[nikomatsakis]

On IRC, Marijn and I discussed not fixing this but rather taking an
alternate approach, possibly based on regions.

[nikomatsakis]

The solution I think would be best is (a modified version of) this proposal:

https://github.com/mozilla/rust/wiki/Proposal-to-address-Dan%27s-bug

In particular, we do not need to introduce the "assignable" class but can instead rely on the mutability information found in region ptrs.