Tracking Issue for maybe_uninit_array_assume_init

[joboet]

Feature gate: #![feature(maybe_uninit_array_assume_init)]

This is a tracking issue for MaybeUninit::array_assume_init.

Public API

// core::mem::maybe_uninit

impl<T> MaybeUninit<T> {
    pub unsafe fn array_assume_init<const N: usize>(array: [Self; N]) -> [T; N];
}

Steps / History

• Implementation: Add MaybeUninit method array_assume_init #80600
• Final commenting period (FCP)
• Stabilization PR

Unresolved Questions

• Should this be a const function?
• Should this be a method of array?
• Should this method be replaced by a conversion method from [MaybeUninit<T>; N] to MaybeUninit<[T; N]> (see comment).

[bstrie]

The documentation for this item suggests that it should be used with another unstable API, MaybeUninit::uninit_array. However that API is not contained under this feature flag, and furthermore I can't find a tracking issue for it. Should we use this issue to track that one as well?

[cole-miller]

Fwiw, this function is now used by hashbrown (nightly crate feature only, introduced in rust-lang/hashbrown#239). I'd love to see it stabilized soon! (Along with maybe_uninit_uninit_array.)

[the8472]

A possible alternative is to have a conversion method from [MaybeUninit<T>; N] to MaybeUninit<[T; N]> and then assume_init() the latter. Then a separate array_assume_init() function isn't needed.

[joboet]

A possible alternative is to have a conversion method from [MaybeUninit<T>; N] to MaybeUninit<[T; N]> and then assume_init() the latter. Then a separate array_assume_init() function isn't needed.

That's a good idea! MaybeUninit::from(_).assume_init() is about the same length as MaybeUninit::array_assume_init(_) and allows for better integration with the other methods.

[Kixunil]

Converting in the other direction could be useful too as well as between MaybeUninit<[MaybeUninit<T>; N]> and the rest.

[clarfonthey]

Would very much like some version of uninit_array and array_assume_init to be stable; right now there's not really a stable equivalent.

IMHO, we should at least collect the array-based issues into a single issue (likely this one) like what's already done with the slice methods in #63569.

[Raekye]

• Should this be a const function?

MaybeUninit::assume_init is const as of 1.59, and I'd like to see this be const as well

One issue with the conversion traits (From and Into) is that they can't be used in const contexts, so I'd like this function to exist as well

[clarfonthey]

I filed #96097 since MaybeUninit::uninit_array didn't have a tracking issue, and linked to this one. So, stuff that doesn't specifically apply to MaybeUninit::array_assume_init and concerns uninit arrays in general can probably go there instead of here.

Also created #96099 which adds constness to this method + links everything to the new issue so people are funnelled in the same place. If you think array_assume_init should still point here, I would voice so there.

[joboet]

Thank you! I'll close this issue then. For future readers: please go to #96097 for further discussion of this feature.