This crate provides a derive macro for generating safe conversions from &T
to &U
where the struct U
contains a single field of type T
.
use ref_cast::RefCast;
#[derive(RefCast)]
#[repr(transparent)]
struct U(String);
fn main() {
let s = String::new();
// Safely cast from `&String` to `&U`.
let u = U::ref_cast(&s);
}
Note that #[repr(transparent)]
is required in order for the conversion to
be sound. The derive macro will refuse to compile if that is not present.
Suppose we have a multidimensional array represented in a flat buffer in row-major order for performance reasons, but we want to expose an indexing operation that works in column-major order because it is more intuitive in the context of our application.
const MAP_WIDTH: usize = 4;
struct Tile(u8);
struct TileMap {
storage: Vec<Tile>,
}
// `tilemap[x][y]` should give us `tilemap.storage[y * MAP_WIDTH + x]`.
The signature of the Index
trait in Rust is such that the output is
forced to be borrowed from the type being indexed. So something like the
following is not going to work.
struct Column<'a> {
tilemap: &'a TileMap,
x: usize,
}
// Does not work! The output of Index must be a reference that is
// borrowed from self. Here the type Column is not a reference.
impl Index<usize> for TileMap {
fn index(&self, x: usize) -> Column {
assert!(x < MAP_WIDTH);
Column { tilemap: self, x }
}
}
impl<'a> Index<usize> for Column<'a> {
fn index(&self, y: usize) -> &Tile {
&self.tilemap.storage[y * MAP_WIDTH + self.x]
}
}
Here is a working approach using RefCast
.
#[derive(RefCast)]
#[repr(transparent)]
struct Strided([Tile]);
// Implement `tilemap[x][y]` as `tilemap[x..][y * MAP_WIDTH]`.
impl Index<usize> for TileMap {
type Output = Strided;
fn index(&self, x: usize) -> &Self::Output {
assert!(x < MAP_WIDTH);
Strided::ref_cast(&self.storage[x..])
}
}
impl Index<usize> for Strided {
type Output = Tile;
fn index(&self, y: usize) -> &Self::Output {
&self.0[y * MAP_WIDTH]
}
}
&T
to &U
where the struct U
contains a single field of
type T
.