Generics
Generics allow you to use the same functions with multiple different concrete data types. You can read more about the concept of generics in the Rust documentation here.
Here is a trivial example showing the identity function that supports any type. In Rust, it is
common to refer to the most general type as T. We follow the same convention in Noir.
fn id<T>(x: T) -> T {
x
}
Numeric Generics
If we want to be generic over array lengths (which are type-level integers), we can use numeric
generics. Using these looks similar to using regular generics, but introducing them into scope
requires declaring them with let MyGenericName: IntegerType. This can be done anywhere a normal
generic is declared. Instead of types, these generics resolve to integers at compile-time.
Here's an example of a struct that is generic over the size of the array it contains internally:
struct BigInt<let N: u32> {
limbs: [u32; N],
}
impl<let N: u32> BigInt<N> {
// `N` is in scope of all methods in the impl
fn first(first: BigInt<N>, second: BigInt<N>) -> Self {
assert(first.limbs != second.limbs);
first
fn second(first: BigInt<N>, second: Self) -> Self {
assert(first.limbs != second.limbs);
second
}
}
In Structs
Generics are useful for specifying types in structs. For example, we can specify that a field in a
struct will be of a certain generic type. In this case value is of type T.
struct RepeatedValue<T> {
value: T,
count: Field,
}
impl<T> RepeatedValue<T> {
fn print(self) {
for _i in 0 .. self.count {
println(self.value);
}
}
}
fn main() {
let repeated = RepeatedValue { value: "Hello!", count: 2 };
repeated.print();
}
The print function will print Hello! an arbitrary number of times, twice in this case.
Calling functions on generic parameters
Since a generic type T can represent any type, how can we call functions on the underlying type?
In other words, how can we go from "any type T" to "any type T that has certain methods available?"
This is what traits are for in Noir. Here's an example of a function generic over
any type T that implements the Eq trait for equality:
fn first_element_is_equal<T, let N: u32>(array1: [T; N], array2: [T; N]) -> bool
where T: Eq
{
if (array1.len() == 0) | (array2.len() == 0) {
true
} else {
array1[0] == array2[0]
}
}
fn main() {
assert(first_element_is_equal([1, 2, 3], [1, 5, 6]));
// We can use first_element_is_equal for arrays of any type
// as long as we have an Eq impl for the types we pass in
let array = [MyStruct::new(), MyStruct::new()];
assert(array_eq(array, array, MyStruct::eq));
}
impl Eq for MyStruct {
fn eq(self, other: MyStruct) -> bool {
self.foo == other.foo
}
}
You can find more details on traits and trait implementations on the traits page.
Manually Specifying Generics with the Turbofish Operator
There are times when the compiler cannot reasonably infer what type should be used for a generic, or when the developer themselves may want to manually distinguish generic type parameters. This is where the ::<> turbofish operator comes into play.
The ::<> operator can follow a variable or path and can be used to manually specify generic arguments within the angle brackets.
The name "turbofish" comes from that ::<> looks like a little fish.
Examples:
fn main() {
let mut slice = [];
slice = slice.push_back(1);
slice = slice.push_back(2);
// Without turbofish a type annotation would be needed on the left hand side
let array = slice.as_array::<2>();
}
trait MyTrait {
fn ten() -> Self;
}
impl MyTrait for Field {
fn ten() -> Self { 10 }
}
struct Foo<T> {
inner: T
}
impl<T> Foo<T> {
fn generic_method<U>(_self: Self) -> U where U: MyTrait {
U::ten()
}
}
fn example() {
let foo: Foo<Field> = Foo { inner: 1 };
// Using a type other than `Field` here (e.g. u32) would fail as
// there is no matching impl for `u32: MyTrait`.
//
// Substituting the `10` on the left hand side of this assert
// with `10 as u32` would also fail with a type mismatch as we
// are expecting a `Field` from the right hand side.
assert(10 as u32 == foo.generic_method::<Field>());
}
Arithmetic Generics
In addition to numeric generics, Noir also allows a limited form of arithmetic on generics.
When you have a numeric generic such as N, you can use the following operators on it in a
type position: +, -, *, /, and %.
Note that type checking arithmetic generics is a best effort guess from the compiler and there
are many cases of types that are equal that the compiler may not see as such. For example,
we know that T * (N + M) should be equal to T*N + T*M but the compiler does not currently
apply the distributive law and thus sees these as different types.
Even with this limitation though, the compiler can handle common cases decently well:
trait Serialize<let N: u32> {
fn serialize(self) -> [Field; N];
}
impl Serialize<1> for Field {
fn serialize(self) -> [Field; 1] {
[self]
}
}
impl<T, let N: u32, let M: u32> Serialize<N * M> for [T; N]
where T: Serialize<M> { .. }
impl<T, U, let N: u32, let M: u32> Serialize<N + M> for (T, U)
where T: Serialize<N>, U: Serialize<M> { .. }
fn main() {
let data = (1, [2, 3, 4]);
assert_eq(data.serialize().len(), 4);
}
Note that if there is any over or underflow the types will fail to unify:
fn pop<let N: u32>(array: [Field; N]) -> [Field; N - 1] {
let mut result: [Field; N - 1] = std::mem::zeroed();
for i in 0..N - 1 {
result[i] = array[i];
}
result
}
fn main() {
// error: Could not determine array length `(0 - 1)`
pop([]);
}
Source code: test_programs/compile_failure/arithmetic_generics_underflow/src/main.nr#L1-L14
This also applies if there is underflow in an intermediate calculation:
fn main() {
// From main it looks like there's nothing sketchy going on
seems_fine([]);
}
// Since `seems_fine` says it can receive and return any length N
fn seems_fine<let N: u32>(array: [Field; N]) -> [Field; N] {
// But inside `seems_fine` we pop from the array which
// requires the length to be greater than zero.
// error: Could not determine array length `(0 - 1)`
push_zero(pop(array))
}
fn pop<let N: u32>(array: [Field; N]) -> [Field; N - 1] {
let mut result: [Field; N - 1] = std::mem::zeroed();
for i in 0..N - 1 {
result[i] = array[i];
}
result
}
fn push_zero<let N: u32>(array: [Field; N]) -> [Field; N + 1] {
let mut result: [Field; N + 1] = std::mem::zeroed();
for i in 0..N {
result[i] = array[i];
}
// index N is already zeroed
result
}