by-val and by-move can be dangerous when combined

[nikomatsakis]

Currently, the compiler does not account for the fact that "by value" passing of ref-counted boxes is in fact a borrow. This can be fixed easily enough in borrowck (or, I imagine, in alias). In any case, I added an xfail'd test (by-val-and-by-move.rs) that demonstrates the problem. This test fails (note the environment variable RUST_POISON_ON_FREE---without that, the test may accidentally pass even though it is accessing a freed pointer).

// xfail-test
// exec-env:RUST_POISON_ON_FREE

fn it_takes_two(x: @int, -y: @int) -> int {
    free(y);
    #debug["about to deref"];
    *x
}

fn free<T>(-t: T) {
}

fn main() {
    let z = @3;
    assert 3 == it_takes_two(z, z);
}

[nikomatsakis]

just realized---this should be easy to fix with borrowck, but we don't handle it now. we don't seem to get the scope right when checking that there is no move from a borrow ptr.

[graydon]

Obsolete given scheduled removal of modes (or at least WONTFIX).