Description
Feature gate: #![feature(slice_to_boxed)]
This is a tracking issue for copying/cloning slices into boxed slices.
The current recommended approach for turning a &[T] where T: Clone into a Box<[T]> seems to be to use .to_vec().into_boxed_slice(). This misses some opportunities for optimization, since the Vec is allocated with a known length. A Godbolt comparison shows that Vec::into_boxed_slice() emits unnecessary calls to realloc() and free() and uses some extra registers for local variables: https://godbolt.org/z/eq7Pc7.
There is currently an impl<T: Copy> From<&[T]> for Box<[T]>, but not for T: Clone like there is for Vec<T>.
We propose adding a function on slices to clone/copy them into boxed slices and relaxing the From<&[T]> bound to T: Clone for Box<[T]>. Both of these should be specialized to a memcpy() for the common case of T: Copy. This impl bound change would be insta-stable, so it should be delayed until we are ready to stabilize the rest of this feature.
Public API
impl<T: Clone> [T] {
pub fn to_boxed_slice(&self) -> Box<Self>;
pub fn to_boxed_slice_in<A: Allocator>(&self, alloc: A) -> Box<Self, A>;
}
// Bounds relaxed from T: Copy
impl<T: Clone> From<&[T]> for Box<[T]> { /* ... */ }
// Bounds relaxed from T: Copy
impl<T: Clone> From<Cow<'_, [T]>> for Box<[T]> { /* ... */ }Steps / History
- Implementation of new functions to clone/copy slices into boxed slices: Add
to_boxed_slice()to clone slice into boxed slice #80460 - Replacing
T: Copybounds withT: CopyonBox<[T]>'sFrom<&[T]>andFrom<Cow<'_, [T]>>implementations - Final commenting period (FCP)
- Stabilization PR
Unresolved Questions
- None yet.