If you find this helpful, please like, bookmark, and follow. To keep learning along, follow this series.
By the way, writing this article took even longer than writing the ownership chapter. Traits are truly a concept that is hard to understand.
10.4.1 Using Traits as Parameters
Let’s continue using the content from the previous article as the example:
pub trait Summary {
fn summarize(&self) -> String;
}
pub struct NewsArticle {
pub headline: String,
pub location: String,
pub author: String,
pub content: String,
}
impl Summary for NewsArticle {
fn summarize(&self) -> String {
format!("{}, by {} ({})", self.headline, self.author, self.location)
}
}
pub struct Tweet {
pub username: String,
pub content: String,
pub reply: bool,
pub retweet: bool,
}
impl Summary for Tweet {
fn summarize(&self) -> String {
format!("{}: {}", self.username, self.content)
}
}
If we define a new function notify, which takes NewsArticle and Tweet as the two types and prints Breaking news!, followed by the return value of calling the summarize method from Summary on the parameter, there is a problem:
the function accepts two different struct types. How can we make the parameter work for two types?
Let’s think about it: what do these two structs have in common? Exactly—they both implement the Summary trait. Rust provides a solution for this situation:
pub fn notify(item: &impl Summary) {
println!("Breaking news! {}", item.summarize());
}
Just write the parameter type as impl some_trait. Since both of these structs implement the Summary trait, we write impl Summary. And because this function does not need ownership of the data, we write it as a reference: &impl Summary. If some other data type also implements Summary, it can be passed in as well.
The impl trait syntax is suitable for simple cases. For more complex cases, trait bound syntax is usually used.
Using the same code, but written with trait bounds:
pub fn notify<T: Summary>(item: &T) {
println!("Breaking news! {}", item.summarize());
}
These two forms are equivalent.
However, this simple example does not show the advantages of trait bounds very well. Let’s look at another example. Suppose I want to design a new notify1 function. It takes two parameters, and the content after Breaking news! is the return value of calling summarize on each parameter.
Trait-bound version:
pub fn notify1<T: Summary>(item1: &T, item2: &T) {
println!("Breaking news! {} {}", item1.summarize(), item2.summarize());
}
impl trait version:
pub fn notify1(item1: &impl Summary, item2: &impl Summary) {
println!("Breaking news! {} {}", item1.summarize(), item2.summarize());
}
Clearly, the former function signature is easier to write and more intuitive than the latter.
In fact, impl trait is just syntax sugar for trait bounds, so it is understandable that it is not suitable for complex cases.
So what if the notify function needs its parameter to implement both the Display trait and the Summary trait? In other words, how do you write two or more trait bounds?
Example:
pub fn notify_with_display<T: Summary + std::fmt::Display>(item: &T) {
println!("Breaking news! {}", item);
}
Use + to connect each trait bound.
Another point: because Display is not in the prelude, when writing it you need to spell out its path. You can also import Display at the top of the code first, like this: use std::fmt::Display. Then you can write Display directly in the trait bounds:
use std::fmt::Display;
pub fn notify_with_display<T: Summary + Display>(item: &T) {
println!("Breaking news! {}", item);
}
Don’t forget that impl trait is also syntax sugar, and in that syntax sugar you also connect trait bounds with +:
use std::fmt::Display;
pub fn notify_with_display(item: &(impl Summary + Display)) {
println!("Breaking news! {}", item);
}
This form has one drawback: if there are too many trait bounds, the large amount of constraint information will reduce the readability of the function signature. To solve this, Rust provides an alternative syntax: write the trait bounds after the function signature using a where clause.
Here is the ordinary syntax for multiple trait bounds:
use std::fmt::Display;
use std::fmt::Debug;
pub fn special_notify<T: Summary + Display, U: Summary + Debug>(item1: &T, item2: &U) {
println!("Breaking news! {} and {}", item1.summarize(), item2.summarize());
}
The same code rewritten with a where clause:
use std::fmt::Display;
use std::fmt::Debug;
pub fn special_notify<T, U>(item1: &T, item2: &U)
where
T: Summary + Display,
U: Summary + Debug,
{
println!("Breaking news! {} and {}", item1.summarize(), item2.summarize());
}
This syntax is very similar to C#.
10.4.2 Using Traits as Return Types
Just like using traits as parameters, using traits as return values can also use impl trait. For example:
fn returns_summarizable() -> impl Summary {
Tweet {
username: String::from("horse_ebooks"),
content: String::from(
"of course, as you probably already know, people",
),
reply: false,
retweet: false,
}
}
This syntax has a drawback: if the return type implements a certain trait, then you must ensure that all possible return values of this function/method are only one type. That is because the impl form has some limitations in how it works, which is why Rust does not support it in every case. But Rust does support dynamic dispatch, which will be covered later.
For example:
fn returns_summarizable(flag:bool) -> impl Summary {
if flag {
Tweet {
username: String::from("horse_ebooks"),
content: String::from(
"of course, as you probably already know, people",
),
reply: false,
retweet: false,
}
} else {
NewsArticle {
headline: String::from("Penguins win the Stanley Cup Championship!"),
location: String::from("Pittsburgh, PA, USA"),
author: String::from("Iceburgh, Scotland"),
content: String::from(
"The Pittsburgh Penguins once again are the best \
hockey team in the NHL.",
),
}
}
}
There are two possible return types depending on the value of flag: Tweet and NewsArticle. At that point, the compiler will report an error:
error[E0308]: `if` and `else` have incompatible types
--> src/lib.rs:42:9
|
32 | / if flag {
33 | | / Tweet {
34 | | | username: String::from("horse_ebooks"),
35 | | | content: String::from(
36 | | | "of course, as you probably already know, people",
... | |
39 | | | retweet: false,
40 | | | }
| | |_________- expected because of this
41 | | } else {
42 | | / NewsArticle {
43 | | | headline: String::from("Penguins win the Stanley Cup Championship!"),
44 | | | location: String::from("Pittsburgh, PA, USA"),
45 | | | author: String::from("Iceburgh, Scotland"),
... | |
49 | | | ),
50 | | | }
| | |_________^ expected `Tweet`, found `NewsArticle`
51 | | }
| |_______- `if` and `else` have incompatible types
|
help: you could change the return type to be a boxed trait object
|
31 | fn returns_summarizable(flag:bool) -> Box<dyn Summary> {
| ~~~~~~~ +
help: if you change the return type to expect trait objects, box the returned expressions
|
33 ~ Box::new(Tweet {
34 | username: String::from("horse_ebooks"),
...
39 | retweet: false,
40 ~ })
41 | } else {
42 ~ Box::new(NewsArticle {
43 | headline: String::from("Penguins win the Stanley Cup Championship!"),
...
49 | ),
50 ~ })
|
The error message says that the return types of if and else are incompatible, meaning they are not the same type.
Trait Bound Example
Do you still remember the code for comparing numbers that was mentioned in 10.2. Generics? I’ll paste it here:
fn largest<T>(list: &[T]) -> T{
let mut largest = list[0];
for &item in list{
if item > largest{
largest = item;
}
}
largest
}
I’ll also paste the error that occurred at that time:
error[E0369]: binary operation `>` cannot be applied to type `T`
--> src/main.rs:4:17
|
4 | if item > largest{
| ---- ^ ------- T
| |
| T
|
help: consider restricting type parameter `T`
|
1 | fn largest<T: std::cmp::PartialOrd>(list: &[T]) -> T{
| ++++++++++++++++++++++
Now that we have learned traits, does your understanding of this code and its error message feel different?
Let’s start by analyzing the error message. The error says that the comparison operator > cannot be applied to type T. The help line below says to consider restricting type parameter T, and further down it gives the concrete approach: add std::cmp::PartialOrd after T (in the trait bound, you only need to write PartialOrd because it is in the prelude, so the full path is not needed). This is actually the trait used for comparisons. Try modifying it according to the hint:
fn largest<T: PartialOrd>(list: &[T]) -> T{
let mut largest = list[0];
for &item in list{
if item > largest{
largest = item;
}
}
largest
}
It still reports an error:
error[E0508]: cannot move out of type `[T]`, a non-copy slice
--> src/main.rs:2:23
|
2 | let mut largest = list[0];
| ^^^^^^^
| |
| cannot move out of here
| move occurs because `list[_]` has type `T`, which does not implement the `Copy` trait
|
help: if `T` implemented `Clone`, you could clone the value
--> src/main.rs:1:12
|
1 | fn largest<T: std::cmp::PartialOrd>(list: &[T]) -> T{
| ^ consider constraining this type parameter with `Clone`
2 | let mut largest = list[0];
| ------- you could clone this value
help: consider borrowing here
|
2 | let mut largest = &list[0];
| +
But the error is different this time: the element cannot be moved out of list, because T in list does not implement the Copy trait. The help below says that if T implements the Clone trait, consider cloning the value. There is also another help below that suggests borrowing.
Based on the above information, there are three solutions:
- Add the
Copytrait to the generic type - Use cloning, which means adding the
Clonetrait to the generic type - Use borrowing
Which solution should we choose? It depends on your needs. I want this function to handle collections of numbers and characters. Since numbers and characters are stored on the stack, they both implement the Copy trait, so it is enough to add Copy to the generic type:
fn largest<T: PartialOrd + Copy>(list: &[T]) -> T{
let mut largest = list[0];
for &item in list{
if item > largest{
largest = item;
}
}
largest
}
fn main() {
let number_list = vec![34, 50, 25, 100, 65];
let result = largest(&number_list);
println!("The largest number is {}", result);
let char_list = vec!['y', 'm', 'a', 'q'];
let result = largest(&char_list);
println!("The largest char is {}", result);
}
Output:
The largest number is 100
The largest char is y
What if I want this function to compare a String collection? Since String is stored on the heap, it does not implement the Copy trait, so the idea of adding Copy to the generic type does not work.
Then try cloning, which means adding the Clone trait to the generic type:
fn largest<T: PartialOrd + Clone>(list: &[T]) -> T{
let mut largest = list[0].clone();
for &item in list.iter() {
if item > largest{
largest = item;
}
}
largest
}
fn main() {
let string_list = vec![String::from("dev1ce"), String::from("Zywoo")];
let result = largest(&string_list);
println!("The largest string is {}", result);
}
Output:
error[E0507]: cannot move out of a shared reference
--> src/main.rs:3:18
|
3 | for &item in list.iter() {
| ---- ^^^^^^^^^^^
| |
| data moved here
| move occurs because `item` has type `T`, which does not implement the `Copy` trait
|
help: consider removing the borrow
|
3 - for &item in list.iter() {
3 + for item in list.iter() {
|
The error says that data cannot be moved because this form requires Copy, which String does not provide. What should we do?
Then do not move the data; do not use pattern matching. Remove the & in front of item, so item changes from T to an immutable reference &T. Then use the dereference operator * during comparison to dereference &T back to T and compare it with largest (the code below uses this approach), or add & in front of largest to make it &T. In short, the two values being compared must have the same type:
fn largest<T: PartialOrd + Clone>(list: &[T]) -> T{
let mut largest = list[0].clone();
for item in list.iter() {
if *item > largest{
largest = item.clone();
}
}
largest
}
fn main() {
let string_list = vec![String::from("dev1ce"), String::from("Zywoo")];
let result = largest(&string_list);
println!("The largest string is {}", result);
}
Remember that T does not implement the Copy trait, so when assigning to largest, you need to use the clone method.
Output:
The largest string is dev1ce
This form is written this way because the return value is T. If you change the return value to &T, then cloning is no longer needed:
fn largest<T: PartialOrd>(list: &[T]) -> &T{
let mut largest = &list[0];
for item in list.iter() {
if item > largest{
largest = item;
}
}
largest
}
fn main() {
let string_list = vec![String::from("dev1ce"), String::from("Zywoo")];
let result = largest(&string_list);
println!("The largest string is {}", result);
}
But remember that when initializing largest, you must set it to &T, so you need to add & in front of list[0] to make it a reference. Also, when comparing, you cannot use the method of dereferencing item; instead, you need to add & in front of largest.
10.4.3 Conditionally Implementing Methods with Trait Bounds
If you use trait bounds on an impl block with generic type parameters, you can conditionally implement methods for types that implement specific traits.
For example:
use std::fmt::Display;
struct Pair<T> {
x: T,
y: T,
}
impl<T> Pair<T> {
fn new(x: T, y: T) -> Self {
Self { x, y }
}
}
impl<T: Display + PartialOrd> Pair<T> {
fn cmp_display(&self) {
if self.x >= self.y {
println!("The largest member is x = {}", self.x);
} else {
println!("The largest member is y = {}", self.y);
}
}
}
No matter what the concrete type of T is, the new function will always exist on Pair. But the cmp_display method exists only when T implements both Display and PartialOrd.
You can also conditionally implement one trait for any type that implements another trait. Implementing a trait for all types that satisfy a trait bound is called a blanket implementation.
Take the standard library’s to_string function as an example:
impl<T: Display> ToString for T {
// ......
}
This means that ToString is implemented for all types that satisfy the Display trait, which is what a blanket implementation is: any type that implements Display can call methods on ToString.
Using an integer as an example:
let s = 3.to_string();
This works because i32 implements the Display trait, so it can call the to_string method from ToString.