Tracking Issue for `Vec::recycle`

[llogiq]

Feature gate: #![feature(vec_recycle)]

This is a tracking issue for the Vec::recycle method, which allows reusing a Vec in a loop even if the lifetime unification would otherwise disallow it.

Public API

// in alloc::vec
#![feature(transmutability)]
#![feature(trait_alias)]

use std::mem::{Assume, MaybeUninit, TransmuteFrom};

/// Implemented if the size and alignment `T` and `Self` are both equal.
trait Recyclable<T> =
    for<'a> TransmuteFrom<&'a MaybeUninit<T>, { Assume::SAFETY }> 
    where 
        T: for<'a> TransmuteFrom<&'a MaybeUninit<Self>, { Assume::SAFETY }>;

impl<T> Vec<T> {
    #[feature(vec_recycle)]
    pub fn recycle<U>(self) -> Vec<U> where U: Recyclable<T>;
}

Steps / History

• ACP: Vec::recycle libs-team#674
• Implementation: #...
• Final comment period (FCP)¹
• Stabilization PR

Unresolved Questions

• None yet.

Footnotes

1. https://std-dev-guide.rust-lang.org/feature-lifecycle/stabilization.html ↩

[GrigorenkoPV]

@rustbot claim
(implementation)

[GrigorenkoPV]

Unfortunately, there are some issues with the TransmuteFrom approach.

#![feature(transmutability)]

use core::mem::{Assume, TransmuteFrom};

const fn check<From, To>()
where
    To: TransmuteFrom<From, { Assume::SAFETY }>,
{
}

fn main() {
    const {
        assert!(align_of::<&[u8]>() == align_of::<&[&[u8]]>());
    }
    check::<&[&[u8]], &[u8]>();
    check::<&[u8], &[&[u8]]>();
}

When uncommenting the two lines, the latest nightly (2025-10-30) gives the following errors:

error[E0277]: `[&[u8]]` cannot be safely transmuted into `[u8]`
  --> src/main.rs:15:23
   |
15 |     check::<&[&[u8]], &[u8]>();
   |                       ^^^^^ analyzing the transmutability of `[&[u8]]` is not yet supported
   |

error[E0277]: `&[u8]` cannot be safely transmuted into `&[&[u8]]`
  --> src/main.rs:16:20
   |
16 |     check::<&[u8], &[&[u8]]>();
   |                    ^^^^^^^^ the minimum alignment of `&[u8]` (1) should be greater than that of `&[&[u8]]` (8)
   |

So in one direction it's unimplemented and in the other direction it's seemingly wrong? Maybe i don't understand what it means by "minimum alignment", but the align_of gives the same values for both types.