Treatment of bound / free regions is wrong

[nikomatsakis]

The current code for handling bound regions is not right. We need to track node-ids uniformly and not only for free regions. Here is an example of a test case that fails. I have a partial fix for this, but not yet committed.

// Ensure that you cannot use generic types to return a region outside
// of its bound.  Here, in the `return_it()` fn, we call with() but
// with R bound to &int from the return_it.  Meanwhile, with()
// provides a value that is only good within its own stack frame. This
// used to successfully compile because we failed to account for the
// fact that fn(x: &int) rebound the region &.

fn with<R>(f: fn(x: &int) -> R) -> R {
    f(&3)
}

fn return_it() -> &int {
    with(|o| o) //~ ERROR mismatched types
        //~^ ERROR reference is not valid outside of its lifetime
}

fn main() {
    let x = return_it();
    #debug["foo=%d", *x];
}

[nikomatsakis]

Another test case:

// Similar to regions-ret-borrowed.rs, but using a named lifetime.  At
// some point regions-ret-borrowed reported an error but this file did
// not, due to special hardcoding around the anonymous region.

fn with<R>(f: fn(x: &a/int) -> R) -> R {
    f(&3)
}

fn return_it() -> &a/int {
    with(|o| o) //~ ERROR mismatched types
        //~^ ERROR reference is not valid outside of its lifetime
}

fn main() {
    let x = return_it();
    #debug["foo=%d", *x];
}

[nikomatsakis]

This is fixed, although I'd still to convert the codebase to using debruijn indices or something similar.