One surprising feature of type inference in languages like Rust is defining functions with generic return types. The idea is that by specifying at some later point in the code which type you want your function to return, the compiler can go back and fill in the blanks.

For example, let’s have a look at this function:

fn new<T: Default>() -> T {
  T::default()
}

You pick the output

It has no value parameters, but one type parameter, T. That T is its return type and also used in the function body. You can call it like so:

let x: u32 = new();

Or, being explicit about the type parameter, like this:

let x = new::<i32>();

This is quite neat!

More generic: collect

A promising way to be more generic in Rust is to use more traits! Have a look at how the Iterator::collect method is defined:

fn collect<B: FromIterator<Self::Item>>(self) -> B // ...

You can read this type signature as

Consume self and return something of a type that implements can be made From [an] Iterator for the type of items we are iterating over.

Like above, we call this by specifying what kind of output type want. [Looking][FromIterator implementors] at some of the types FromIterator is implemented for is pretty revealing of the use cases. You can get:

  • a Vec by collecting any items,
  • a BTreeMap or HashMap by collecting tuples,
  • but also PathBuf by collecting Paths,
  • and String for strings and string slices.

Note: All these types are what you might call “container” types.

One more for the road

More generic? More traits.

There is one more gem hidden in FromIterator:

impl<A, E, V> FromIterator<Result<A, E>> for Result<V, E> where
    V: FromIterator<A>, // ...

This means: You can construct a Result containing any type of container of items A by collecting items that are Results of type A. (The first Err will make the outer Result be an Err.) Here’s an example, see the docs for another one:

let input: Vec<Result<i32, ()>> = vec![Ok(1), Ok(2)];
let output: Result<Vec<i32>, ()> = input.into_iter().collect();

Note: If you like type theory: What we’re building is a Result<<T<A>, E>> by collecting Result<A, E>s and specifying T.