Cannot move upvars through multiple closures

[brson]

    let q = ~10;
    let addr = ptr::addr_of(*q);
    let f = fn@() -> *int {
        fn@() -> *int {
            ptr::addr_of(*q)
        }()
    };
    assert addr == f();

This fails because q ends up being a copy. Even adding capture clauses to both functions does not work because q cannot be moved out of the first closure.

This results in some subtle behavior with functional control structures, particularly since the copy happens implicitly. For example, in my rewrite of core::comm I have a requirement that a particular variable not be copied. This has to carefully be verified through tests to prevent the compiler from inserting copies and prevents me from using nice-looking functional code.

I don't know how to fix this. Probably the easiest thing is to disallow this pattern by requiring the copies to be explicit.

[brson]

Of course, requiring an explicit copy capture clause makes using higher-order functions for flow control a lot less appealing.

[pnkfelix]

I'm having trouble reproducing this bug. (There have been a number syntactic changes to the language since it was filed.)

[pnkfelix]

In particular, the interface to addr_of has changed so that it now takes a &T instead of a T. I tried porting the example by replacing the addr_of(*q) with addr_of(q), but I do not see the assertion fail. So I am not sure

1. if this problem has been resolved by some other means in the meantime, or
2. if I am misunderstanding the scenario that is being tested above, or
3. if the change to the addr_of API somehow defines away the problem being considered in the above scenario.

[nikomatsakis]

If there was ever a bug here (which is unclear to me), I think that this no longer applies due to the moves-based-on-type rules. I'm closing. @brson let me know if you disagree.