Tracking Issue for Vec::split_at_spare_mut

[KodrAus]

Feature gate: #![feature(vec_split_at_spare)]

This is a tracking issue for Vec::split_at_spare_mut, which is used as the basis for a few methods internally to give a slice pointing to the initialized portion of the Vec and a slice of MaybeUninit pointing to the uninitialized portion.

Public API

impl<T, A: Allocator> Vec<T, A> {
    pub fn split_at_spare_mut(&mut self) -> (&mut [T], &mut [MaybeUninit<T>]) {}
}

Steps / History

• Implementation: Make Vec::split_at_spare_mut public #81687
• Final commenting period (FCP)
• Stabilization PR

Unresolved Questions

• Should we consider changing the name? split_at_spare_mut doesn't seem like an intuitive name
• Should we deprecate Vec::spare_capacity_mut? Any usecase of Vec::spare_capacity_mut can be replaced with Vec::split_at_spare_mut (but not vise-versa):
  edit: spare_capacity_mut has been stabilised now in Stabilize vec_spare_capacity #93016 so deprecating it is probably not an option

[Amanieu]

Interesting! I personally use spare_capacity_mut in my code to read into uninitialized buffers. I feel that it may be worth keeping it as a convenience method, but then again it is quite rarely used so I don't mind having to write an extra .1.

[WaffleLapkin]

Hm, why there are no unresolved questions yet? It doesn't seem that the questions from the implementation PR were resolved:

• Should we consider changing the name? split_at_spare_mut doesn't seem like an intuitive name
• Should we deprecate Vec::spare_capacity_mut? Any usecase of Vec::spare_capacity_mut can be replaced with Vec::split_at_spare_mut (but not vise-versa)

[KodrAus]

@WaffleLapkin I just missed those questions that came up after the PR approval is all 🙂 They're in the tracking issue now.

[RalfJung]

Any usecase of Vec::spare_capacity_mut can be replaced with Vec::split_at_spare_mut (but not vise-versa)

That is not true: split_at_spare_mut creates mutable references to the entire initialized part of the vector and will thus invalidate any preexisting pointers. spare_capacity_mut, if it is implemented carefully (which I have not checked) could be made to not invalidate existing pointers into the initialized part of the vector.

[WaffleLapkin]

@RalfJung ohhh. I haven't thought of this. Then I guess the question "do we need both?" is resolved: yes.

if it is implemented carefully

Current implementation just calls self.split_at_spare_mut().1 😅

[RalfJung]

Current implementation just calls self.split_at_spare_mut().1

Well, that doesn't count as "careful" I guess. ;)

[WaffleLapkin]

I guess so 😅

I'll try to make the implementation more careful (revert it?) and make a PR :)

[saethlin]

Oops, what I commented here is far more relevant to Vec::spare_capacity_mut, comment moved: #75017 (comment)

[ast-ral]

Apologies if this is out of place, but it might be good to have a analogous leak_and_split_spare_capacity_mut method, since leak doesn't shrink the vector before returning and there's no (safe) way to get access to the unused capacity that I'm aware of.