The .
operator in Rust comes with a lot of magic! When you use .
, the compiler will insert as many *
s (dereferencing operations) necessary to find the method down the deref "tree". As this happens at compile time, there is no runtime cost of finding the method.
let mut name: String = "hello world".to_string();
// no deref happens here because push is defined in String itself
name.push('!');
let name_ref: &String = &name;
// Auto deref happens here to get to the String. See below
let name_len = name_ref.len();
// You can think of this as syntactic sugar for the following line:
let name_len2 = (*name_ref).len();
// Because of how the deref rules work,
// you can have an arbitrary number of references.
// The . operator is clever enough to know what to do.
let name_len3 = (&&&&&&&&&&&&name).len();
assert_eq!(name_len3, name_len);
Auto dereferencing also works for any type implementing std::ops::Deref
trait.
let vec = vec![1, 2, 3];
let iterator = vec.iter();
Here, iter
is not a method of Vec<T>
, but a method of [T]
. It works because Vec<T>
implements Deref
with Target=[T]
which lets Vec<T>
turn into [T]
when dereferenced by the *
operator (which the compiler may insert during a .
).
Given two types T
and U
, &T
will coerce (implicitly convert) to &U
if and only if T
implements Deref<Target=U>
This allows us to do things like this:
fn foo(a: &[i32]) {
// code
}
fn bar(s: &str) {
// code
}
let v = vec![1, 2, 3];
foo(&v); // &Vec<i32> coerces into &[i32] because Vec<T> impls Deref<Target=[T]>
let s = "Hello world".to_string();
let rc = Rc::new(s);
// This works because Rc<T> impls Deref<Target=T> ∴ &Rc<String> coerces into
// &String which coerces into &str. This happens as much as needed at compile time.
bar(&rc);
For functions that need to take a collection of objects, slices are usually a good choice:
fn work_on_bytes(slice: &[u8]) {}
Because Vec<T>
and arrays [T; N]
implement Deref<Target=[T]>
, they can be easily coerced to a slice:
let vec = Vec::new();
work_on_bytes(&vec);
let arr = [0; 10];
work_on_bytes(&arr);
let slice = &[1,2,3];
work_on_bytes(slice); // Note lack of &, since it doesn't need coercing
However, instead of explicitly requiring a slice, the function can be made to accept any type that can be used as a slice:
fn work_on_bytes<T: AsRef<[u8]>>(input: T) {
let slice = input.as_ref();
}
In this example the function work_on_bytes
will take any type T
that implements as_ref()
, which returns a reference to [u8]
.
work_on_bytes(vec);
work_on_bytes(arr);
work_on_bytes(slice);
work_on_bytes("strings work too!");
use std::ops::Deref;
use std::fmt::Debug;
#[derive(Debug)]
struct RichOption<T>(Option<T>); // wrapper struct
impl<T> Deref for RichOption<T> {
type Target = Option<T>; // Our wrapper struct will coerce into Option
fn deref(&self) -> &Option<T> {
&self.0 // We just extract the inner element
}
}
impl<T: Debug> RichOption<T> {
fn print_inner(&self) {
println!("{:?}", self.0)
}
}
fn main() {
let x = RichOption(Some(1));
println!("{:?}",x.map(|x| x + 1)); // Now we can use Option's methods...
fn_that_takes_option(&x); // pass it to functions that take Option...
x.print_inner() // and use it's own methods to extend Option
}
fn fn_that_takes_option<T : std::fmt::Debug>(x: &Option<T>) {
println!("{:?}", x)
}
Deref
has a simple rule: if you have a type T
and it implements Deref<Target=F>
, then &T
coerces to &F
, compiler will repeat this as many times as needed to get F, for example:
fn f(x: &str) -> &str { x }
fn main() {
// Compiler will coerce &&&&&&&str to &str and then pass it to our function
f(&&&&&&&"It's a string");
}
Deref coercion is especially useful when working with pointer types, like Box
or Arc
, for example:
fn main() {
let val = Box::new(vec![1,2,3]);
// Now, thanks to Deref, we still
// can use our vector method as if there wasn't any Box
val.iter().fold(0, |acc, &x| acc + x ); // 6
// We pass our Box to the function that takes Vec,
// Box<Vec> coerces to Vec
f(&val)
}
fn f(x: &Vec<i32>) {
println!("{:?}", x) // [1,2,3]
}