Prototype for MaybeUninit formulation of arrayvec

[bluss]

• Testing ideas for arrayvec 1.0
• Looking at what is well defined and what's not
• Not adequate yet:
  • How to handle uninitialization?
  • How to handle ArrayString? It's Copy & uninitialized (always an array of u8)

[bluss]

cc @arielb1 and thanks a lot for the help in searching for the best expression of this in Rust

[tbu-]

Is the repr(C) supposed to inhibit the enum optimization?

[bluss]

repr(C) was put there only there so that we can cast the *const MaybeUninit<T> to *const T. There is a missing link right now, that needs to be rationalized as sound for the intermediate ManuallyDrop too.

[bluss]

A question I didn't fully understand if it was answered:

If we have a MaybeUninit<[T; 3]> which is partially initialized; only its first element is initialized.

1. Is it sound to extract a &[T; 3] from this?
2. Must arrayvec instead extract a *const T? (With the eventual goal of course, of creating &T to an element only if that element is initialized.)

I think the following would be problematic if answered with no:

3. Is it sound to call <[_]>::as_ptr() on a partially initialized [T; 3]?

[bluss]

Yesterday's discussion https://botbot.me/mozilla/rust-libs/2017-10-20/?msg=92551510&page=1

[tbu-]

Thanks for the link to the discussion. Sounds like having a &[T; 3] with uninitialized elements is UB or at least not clearly not UB. We could work with *mut [T; 3] and *mut [T].

[bluss]

What we were trying to do is to get docs amended for ManuallyDrop so that it can be "MaybeUnitialized", but that seems to not be such a good idea. I'm still not entirely convinced.

[bluss]

New version in #97