Description
The Rust standard library contains these impls:
(click to expand search results for From<[...; N]>)
$ rg -B1 'From<\[.*; N\]' library/
library/std/src/collections/hash/set.rs
1010-// See also the equivalent impl on HashMap.
1011:impl<T, const N: usize> From<[T; N]> for HashSet<T, RandomState>
library/std/src/collections/hash/map.rs
1174-// See also the equivalent impl on HashSet.
1175:impl<K, V, const N: usize> From<[(K, V); N]> for HashMap<K, V, RandomState>
library/alloc/src/collections/linked_list.rs
1954-#[stable(feature = "std_collections_from_array", since = "1.56.0")]
1955:impl<T, const N: usize> From<[T; N]> for LinkedList<T> {
library/alloc/src/collections/btree/set.rs
1093-#[stable(feature = "std_collections_from_array", since = "1.56.0")]
1094:impl<T: Ord, const N: usize> From<[T; N]> for BTreeSet<T> {
library/alloc/src/collections/binary_heap.rs
1491-#[stable(feature = "std_collections_from_array", since = "1.56.0")]
1492:impl<T: Ord, const N: usize> From<[T; N]> for BinaryHeap<T> {
library/alloc/src/collections/btree/map.rs
2048-#[stable(feature = "std_collections_from_array", since = "1.56.0")]
2049:impl<K: Ord, V, const N: usize> From<[(K, V); N]> for BTreeMap<K, V> {
library/alloc/src/vec/mod.rs
2909-#[stable(feature = "vec_from_array", since = "1.44.0")]
2910:impl<T, const N: usize> From<[T; N]> for Vec<T> {
library/alloc/src/collections/vec_deque/mod.rs
3050-#[stable(feature = "std_collections_from_array", since = "1.56.0")]
3051:impl<T, const N: usize> From<[T; N]> for VecDeque<T> {
library/alloc/src/boxed.rs
1481-#[stable(feature = "box_from_array", since = "1.45.0")]
1482:impl<T, const N: usize> From<[T; N]> for Box<[T]> {
While most are pretty recent (1.56.0), the Vec one is quite older (1.44.0), but either way, std has been clearly moving to more by-value array functionality, and I think SmallVec should follow this trend.
As a concrete usecase, I'm working on a codebase that uses SmallVec throughout, and a lot of them are created with [...].into_iter().collect(), which would be nicer to spell as just [...].into().
I already am locked into requiring a Rust version new enough to have stable const generics, and AFAICT the const_generics feature of smallvec could be used to opt into a new From impl, but this existing one may be a small issue:
Line 1689 in 218e0bb
That is, that pre-existing impl requires an array as long as the inline capacity of the SmallVec, and while it could be swapped out with the more general one with #[cfg(feature = "const_generics")], the extra flexibility could make inference ambiguous (if anyone has relied on it somehow).
Then again, I don't think syntax like [T; _] (literally an underscore for the length, to allow it to be inferred) is stable yet, and something like SmallVec::from(Default::default()) is too ambiguous anyway, so it would take SmallVec::from(make_array_of_const_generic_len(...)) to be able to cause an issue.
Of course, with a smallvec 2.0 (#240), such compatibility issues (however implausible) wouldn't matter, but I'm not sure if that rewrite is coming any time soon.