Removes boxed closures to compile w/ latest nightly.

[drbawb]

Boxed closures were removed in rust-lang/rust/pull/20578; their use is now a compile error.
Most of these changes are straightforward but I would like to draw special attention to commit 853c36e which fixes up src/sdl2/timer.rs

I discovered a safety violation: but I believe it was present in the previous revision with boxed closures as well. If the Timer is dropped before SDL fires the callback, and the remove_on_drop flag is not set:
then the C code is calling back into a stack-frame which just got freed.

Even putting the closure in a Box<Fn() -> uint> does not help here. The boxed closure would just be stored inside the Timer which was just dropped.

I believe the constructor should be split in two:

1. A safe constructor which forces remove_on_drop to true.
2. An unsafe constructor which allows the caller to set the remove_on_drop flags themselves.

remove_on_drop: false will only work in the case that the closure lives longer than the Timer<'a>.
Perhaps there is some other way we could express this w/ lifetime bounds, though?

---

Let me know your thoughts... but I believe this patch preserves the semantics of the previous revision.

[bitonic]

I've done the same but in a different way in #265. I think it's better to force the input to be a Box'ed unboxed closure, which doesn't require you to get a TraitObject (edit: I realise now that you might not need the TraitObject with your approach either -- not sure about it though). See https://github.com/bitonic/rust-sdl2/blob/912c644e370f62b62242ae71e51e81ad0d258f77/src/sdl2/timer.rs#L29 for my code, and rust-lang/rust@c7dd3c4 for a relevant commit.

I also think that stopping the timer when Timer is dropped is the only nice option. I don't think providing an unsafe interface is even worth it.

[drbawb]

I had tried using a Box there. It still doesn't live long enough; the Boxed closure will be dropped as soon as Timer goes out of scope.

Both Box<Fn() -> uint> and &Fn() -> uint segfault on Mac OS 10.7.4 if the Timer is dropped before SDL fires the callback. I opted for avoiding the extra allocation. (EDIT: unless you specify remove on drop, of course.)

Mine should be equivalently as safe as storing the closure inside the Timer: because I have bounded the callback to 'a which means the closure is on a stack frame which lives [at least as long as] the lifetime of the Timer struct itself.

[bitonic]

Wait, I agree that it's going to be trouble if the Timer is dropped before the callback is fired, which is why I stop the callback as soon as the Timer is dropped.

I see now that it's probably fine to not box the closure. I still don't see the need to transmute into a TraitObject though.

By the way, I wrote a little test, should be easy port to your code: https://github.com/bitonic/rust-sdl2/blob/912c644e370f62b62242ae71e51e81ad0d258f77/src/sdl2/timer.rs#L72