Implement AnyBitPattern for MaybeUninit<T> where T: AnyBitPattern.

[zachs18]

(Partially) resolves #108 .

A feature flag is required because MaybeUninit was unstable in bytemuck's MSRV. I just used the #[cfg(feature = "zeroable_maybe_uninit")] flag since that's required anyway, and it didn't seem reasonable to add another feature flag for this.

---

As discussed in #152 , it is probably unsound to have MaybeUninit<T> implement AnyBitPattern when T contains interior mutability, so a blanket impl<T: Copy + 'static> AnyBitPattern for MaybeUninit<T> would be unsound under that reasoning. However, if T implements AnyBitPattern, then it is known to not have any interior mutability, so T: AnyBitPattern is a sufficient (but not necessary) condition for MaybeUninit<T>: AnyBitPattern to be sound.

Since T: AnyBitPattern sufficient-but-not-necessary requirement, it could be relaxed later, e.g. if APIs around Freeze become available.

[Lokathor]

Follow up question: Can't you cast from AnyBitPattern to AnyBitPattern? Like that's the idea right? Then, wouldn't this impl allow you to cast from MaybeUninit to something else? Because that could put uninit bytes into the something else where they're not supposed to be.

[zachs18]

No, I think you can only cast from NoUninit to AnyBitPattern. AnyBitPattern does not alone imply NoUninit.

AnyBitPattern
The requirements for this is very similar to Pod, except that the type can allow uninit (or padding) bytes.

(from docs)

[Lokathor]

#150 hm, but couldn't you use pod_read_unaligned?

[zachs18]

No, pod_read_unaligned (and the try_/checked:: versions) take in a &[u8], which cannot be uninitialized.

[Lokathor]

uhhh, yeah you're right, you're right. Sorry for the extra questions, MaybeUninit just makes me extra paranoid.