Rust Lifetimes: A Complete Guide to Ownership and Borrowing
Table of Contents
This article is about Rust lifetimes. If you’re grappling with Rust’s compile-time guarantees, Earthly can streamline your build process. Check it out.
As a software developer, you’re probably familiar with common memory-related bugs, such as buffer overflows, use-after-free errors, and data races. These issues can cause a wide range of problems, including crashes, data corruption, and even security vulnerabilities.
Rust is a programming language that uses ownership and borrowing to address memory management issues while prioritizing both performance and memory safety. The approach is based on the concept of lifetimes, where the lifetime system tracks the lifespan of every value, ensuring that references do not outlive their intended lifetime and preventing issues, such as dangling pointers/references and memory leaks.
Unfortunately, Rust’s lifetimes can be difficult to understand, but they’re essential to Rust’s design, and they enable you to write secure and high-performing code while avoiding common memory-related problems found in other languages. In this article, you’ll learn all about lifetimes and the concepts of ownership, borrowing, and resource management in Rust.
Borrowing and References in Rust
Rust’s ownership system is a notable and distinctive feature that revolves around the concepts of ownership and borrowing, which allows developers to manage resources efficiently and safely. It’s designed to prevent memory leaks, data races, and other common problems that occur in other programming languages. Ownership refers to the idea that every value in Rust has a distinct owner. The owner is responsible for value deallocation when it goes out of scope. Rust enforces this ownership model to ensure that memory deallocation occurs automatically and reliably. By structuring ownership in this way, Rust eliminates explicit memory deallocation calls and prevents memory leaks at compile time. In comparison, borrowing refers to borrowing a reference to a resource from its owner. References are a way to access a resource without taking ownership of it, which makes it possible to share the resource between different parts of the program.
To demonstrate how borrowing works, take a look at the following example:
fn main() {
let a = 5;
let b = &a;
println!("{}", b);
}
This is what your output should look like:
5
In this example, b
borrows from the value of a
, the &
symbol acts as a way to create a reference of a
in memory that can be pointed to retrieve its value, and it makes use of an immutable reference to do so (more on this next).
Immutable and Mutable References
There are two types of references in Rust: immutable and mutable.
Immutable references allow read-only access to a resource. Immutable references are created using the &
symbol and can be created multiple times, which means that multiple parts of the program can access the same resource at the same time.
Say you have a vector of integers, and you want to print each element in the vector. You can create an immutable reference to the vector using the following code:
fn main()
let vec = vec![10, 11];
for i in &vec {
println!("{}", i);
}
}
Your output would look like this:
10
11
In comparison, mutable references are created using the &mut
symbol and allow read and write access to a resource. However, there can only be one mutable reference to a resource at any given time. This ensures that only one part of the program can modify the resource at a time, which prevents data races.
For example, suppose you have a mutable vector of integers, and you want to modify its first element. In that case, you can create a mutable reference to the vector using the following code:
fn main() {
let mut vec = vec![10, 11];
let first = &mut vec[0];
*first = 6;
println!("{:?}", vec);
}
Here’s the output:
[6, 11]
In this example, you create a mutable reference to the first element of the vector using the &mut
symbol. Then you modify the first element by dereferencing the reference using the *
operator and set its value to 6
.
Rules for Borrowing
While borrowing is a powerful feature in Rust, it comes with a set of rules that must be followed to ensure memory safety and avoid data races. These rules include the following:
- Each resource can only have one mutable reference or any number of immutable references at a time.
- References must always be valid, which means that the resource being referenced must remain in scope for the entire lifetime of the reference.
- A mutable reference cannot exist at the same time as any other reference, mutable or immutable.
The Rust compiler enforces these rules at compile time, ensuring that your code is safe from data races and other memory-related bugs.
When you follow these rules, you’ll be able to write safer and more efficient code that takes advantage of Rust’s ownership system.
Lifetimes
Lifetimes are a way of tracking the scope of a reference to an object in memory. In Rust, every value has one owner, and when the owner goes out of scope, the value is dropped, and its memory is freed. Lifetimes allow Rust to ensure that a reference to an object remains valid for as long as it’s needed.
In Rust, lifetimes are denoted using the 'a
syntax, where the 'a
is a placeholder for the actual lifetime. The lifetime can be defined as a generic parameter in a function, struct, or trait using angle brackets. The following is an example:
struct Path<'a> {
: &'a i32,
point_x: &'a i32,
point_y}
fn main() {
let p_x = 3200;
let p_y = (p_x / 2) as i32;
let maze = Path { point_x: &p_x, point_y: &p_y };
println!("x = {}, y = {}", maze.point_x, maze.point_y);
}
Your output would look like this:
x = 3200, y = 1600
Here, a struct Path
is defined with two fields: point_x
and point_y
, which references an i32
type. The i32
value (or type) represents a signed integer from the number -2147483648 to 2147483647. The lifetime 'a
specifies that the reference must live at least as long as the instance of the struct.
Now, take a look at this example which is similar to the code above but results in an error:
…fn main() {
let p_x = 3200;
let p_y = {
let temp = 42;
&temp
};
let maze = Path { point_x: &p_x, point_y: p_y };
println!("x = {}, y = {}", maze.point_x, maze.point_y);
}
Can you see the error? Your output will look like this:
|
| let p_y = {
| --- borrow later stored here
| let temp = 42;
| &temp
| ^^^^^ borrowed value does not live long enough
| };
| - `temp` dropped here while still borrowed
In this example, the compiler detects an error when the lifetime reference of temp
goes out of scope. This error prevents the further use of p_y
in the program because the value of temp
has already been dropped. The issue arises because p_y
is assigned a borrowed reference, &temp
, which cannot exist outside the scope of p_y
. This error occurs due to the mismatched lifetimes.
Take a look at a modified version of the previous code snippet:
…
fn main() {
let p_x = 3200;
let temp;
let p_y = {
= 42;
temp &temp
};
let maze = Path { point_x: &p_x, point_y: p_y };
println!("x = {}, y = {}", maze.point_x, maze.point_y);
}
As you can see, this approach works because temp
and p_y
have the same lifetime, allowing the temp
variable to exist for the duration of the program. This means that the reference to temp
, assigned to p_y
, remains valid and can be used throughout the program.
Lifetime Elision
Rust’s lifetime elision rules allow the compiler to infer lifetimes in specific situations, which can reduce the amount of boilerplate code that is needed. The rules are based on the following three-lifetime elision principles:
- Each parameter that is a reference gets its lifetime parameter. In other words, a function with one parameter of type
&T
would have a single lifetime parameter, such asfn foo<'a>(x: &'a T)
. - If there is exactly one input lifetime parameter (ie,
&self
,&mut self
, or&
), that lifetime is assigned to all output lifetime parameters. - If there are multiple input lifetime parameters but one of them is
&self
or&mut self
, the lifetime of&self
or&mut self
is assigned to all output lifetime parameters.
By following these rules, the Rust compiler can automatically infer the correct lifetimes in many cases, reducing the amount of lifetime annotation needed in the code.
It’s important to note that these rules are not exhaustive, and there may be cases where manual lifetime annotation is still required.
Here’s an example of how the lifetime elision rules work:
#[derive(Debug)]
struct Num {
: i32,
x}
impl Num {
fn compare<'a>(&'a self, other: &'a Self) -> &'a Self {
if self.x > other.x {
self
} else {
other}
}
}
fn main() {
let num = Num { x: 3 };
let other_num = #
println!("{:?}", num.compare(other_num));
}
Here’s the output:
Num { x: 3 }
In this example, there is a struct (, i.e., Num
) that contains a single field called x
. There’s also an implementation block for Num
that defines a method comparison. The compare method takes a reference to self
and a reference to another Num
instance (i.e., other
) and returns a reference to the Num
instance with the higher x
value.
The compare
method above uses the third rule of the lifetime elision rules stated earlier. Applying these rules to the compare method:
- The input lifetimes are
&'a self
and&'a Self
. Since one of them is &self, the lifetime of self (’a) is assigned to the output lifetime. - The return type &’a Self uses the same lifetime ’a. Therefore, the code benefits from lifetime elision by avoiding the need to explicitly annotate the lifetimes, making it more concise and readable.
Take a look at another similar example below:
impl Num {
fn compare(&self, other: &Self) -> &Self {
if self.x > other.x {
self
} else {
other}
}
}
fn main() {
let num = Num { x: 3 };
let other_num = #
println!("{:?}", num.compare(other_num));
}
Notice the lifetime elisions removed from your compare
function.
The code above results in the following error:
| fn compare(&self, other: &Self) -> &Self {
| ----- -----
| |
| this parameter and the return type are declared with different lifetimes
...
| other
| ^^^^^ ...but data from `other` is returned here
error: aborting due to previous error
This error occurs because Rust fails to guarantee the lifetimes of self
and other
references in the compare
method, which then results in a lifetime mismatch error from the compiler.
The key to understanding the lifetime elisions here is the lifetime parameter 'a
on the compare
method. This parameter indicates that both the self
and other
references need to have the same lifetime 'a
. This is important because it allows the Rust compiler to automatically infer the lifetimes of these references rather than require explicit annotations.
Lifetime Bounds and Constraints
Lifetime bounds and constraints are a way to limit the lifetime of a reference to a particular scope. This can be useful in situations where you need to ensure that a reference lives long enough to be used in a particular context but not longer than necessary.
Lifetime Bounds
A lifetime bound is a way to specify the minimum lifetime that a reference must have to be used in a particular context. For example, if you have a function that takes a reference to a value and returns a reference to that same value, you can use a lifetime bound to ensure that the returned reference is valid as long as the input reference.
Here’s an example:
use std::fmt::Display;
#[derive(Debug)]
struct Movie<'a, T> {
: &'a str,
title: T,
rating}
impl<'a, T: 'a + Display + PartialOrd> Movie<'a, T> {
fn new(title: &'a str, rating: T) -> Self {
{
Movie ,
title,
rating}
}
}
fn main() {
let movie = Movie::new("The Shawshank Redemption", 9.3);
println!("{:#?}", movie);
}
Your output would look like this:
Movie {
title: "The Shawshank Redemption",
rating: 9.3,
}
Here, the Movie
struct has two fields, and the new
function takes two parameters: title
and rating
. Both of these need to have the same lifetime. The type T
is a generic type that must implement both the Display
and PartialOrd
traits. Additionally, type T
must have a lifetime long enough to persist in the program. This ensures that any type used to represent a rating satisfies the requirements 'a + Display + PartialOrd
.
The 'a
represents a sufficiently long lifetime, the Display
trait ensures that the value being passed is printable, and the PartialOrd
trait ensures that the type should be able to be compared partially using operators like <
, >
, >=
, or<=
. Failing to meet these requirements would result in compile-time errors.
In the main
function, a new Movie
object is created with the title “The Shawshank Redemption” and a rating
of 9.3
. Finally, theprinting
macro is used to print the Movie
object using the {:?}
formatter.
Try changing the rating type value in the Movie
instance being created in your main function:
fn main() {
let movie = Movie::new("The Shawshank Redemption", [9.8]);
println!("{:#?}", movie);
}
Notice the errors:
| let movie = Movie::new("The Shawshank Redemption",
&String::from("9.8"));
| ^|
| creates a temporary which is freed while still in use
| println!("{:#?}", movie);
| ----- borrow later used here
This creates a dangling reference error; A dangling reference error occurs when a temporary value that is not assigned to any variable is borrowed as a reference to nothing and used. In this example, you are passing a borrowed String
reference into the new
function, which it’s lifetime only lives as long as the new
function and not the Movie
instance itself, so even though the String
type implements both bounds Display
and PartialOrd
it still fails due to this reason.
Now try the same example but assign the borrowed String
reference to a variable.
fn main() {
let string_ref = &String::from("9.8");
let movie = Movie::new("The Shawshank Redemption", string_ref);
println!("{:#?}", movie);
}
No errors! This is because the value now has a longer lifetime being assigned to a variable, so it can live as long as the current scope.
Lifetime Constraints
Lifetime constraints are similar to lifetime bounds, but they specify an upper bound on the lifetime of a reference instead of a lower bound. This can be useful in situations where you need to ensure that a reference does not live longer than necessary to avoid creating memory leaks.
Here’s an example:
// Declare the Movie struct with a title and a rating
#[derive(Debug)]
struct Movie<'a> {
: &'a str,
title: u8,
rating}
// Declare the Reviewer struct with reference to a Movie and a name
#[derive(Debug)]
struct Reviewer<'a, 'b: 'a> {
: &'a Movie<'b>,
movie: &'a str,
name}
impl<'a, 'b> Reviewer<'a, 'b> {
// Create a new review
fn new(name: &'a str, movie: &'a Movie) -> Self {
{ movie: movie, name: name }
Reviewer }
}
fn main() {
// Create a movie instance
let movie = Movie {
: "The Rust Movie",
title: 8,
rating};
// Print the review information
println!("{:?}", Reviewer::print_review("Alice", &movie));
}
Notice the new
function located in the implementation of the Reviewer
struct? it takes in two(2) parameters, name
and movie.
The name is a &str
string slice type, while the movie
is a Movie
struct. see the lifetime parameters defined on the Reviewer struct and its implementation. This defines and lets the compiler know the lifetimes it’s working with at compile time.
The Reviewer
struct new
function parameters specify those lifetimes parameters as part of its arguments, which says that the name
and movie
parameters should have the same lifetimes in the function.
Now run your code.
|
| fn new(name: &'a str, movie: &'a Movie) -> Self {
| --------- help: add explicit lifetime `'b` to the type of `movie`: \
`&'a Movie<'b>`
| Reviewer { movie: movie, name: name }
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ lifetime `'b` required
This code fails at compile time; why?
Notice the Reviewer
struct defined in the code:
#[derive(Debug)]
struct Reviewer<'a, 'b: 'a> {
: &'a Movie<'b>,
movie: &'a str,
name}
It defines two lifetime parameters of 'a
and the other 'b
, which is a subtype of 'a
; more on Rust lifetime subtyping will be discussed further later in the article. This means that 'b
should outlive 'a
or at least as long as 'a
is valid, 'b
will also be valid at compile time. This makes the lifetime parameter 'b
an upper bound lifetime and 'a
a lower bound lifetime.
This creates a lifetime constraint on the field movie
on the Reviewer
struct.
The Reviewer struct has two(2) fields, movie
and name
. The movie
field specifies the 'a
and 'b
lifetime parameters where 'a
specifies the lifetimes both field movie
and name
should have, while 'b
specifies that lifetimes the Movie
struct possess must live as long or longer than the Reviewer
struct.
This is useful to prevent the Movie
instance passed to the new function from being dropped and leading to a dangling reference error.
Now to modify the new
function a bit so the code stops throwing errors:
impl<'a, 'b> Reviewer<'a, 'b> {
// Create a new review
fn new(name: &'a str, movie: &'b Movie) -> Self {
{
Reviewer ,
movie
name}
}
}
Here is the output:
Reviewer { movie: Movie { title: "The Rust Movie",
rating: 8 }, name: "Alice" }
Great! Now it works.
In this modification, the name
reference has the lifetime 'a
, while the movie reference has the lifetime ’b. The lifetimes assigned to name
and movie
are independent and not necessarily the same and are not directly tied together in this case.
Static Lifetime
The 'static
lifetime is a special lifetime that represents the entire duration of the program. Any reference with a 'static
lifetime can be used anywhere without worrying about its scope. Here’s an example:
const SECRET_PHRASE: &'static str = "Hello, world!";
Here, a constant variable is created called SECRET_PHRASE
with a 'static
lifetime, which means that it can be used anywhere in the program.
Rust Lifetime Examples
To help solidify your understanding of the Rust lifetimes, explore practical examples where lifetimes are commonly used.
Function Signatures With Lifetimes
Function signatures in Rust are one of the most common places you’ll encounter lifetimes. In Rust, it’s important to specify the lifetime of references passed as function arguments because the compiler needs to know how long a reference remains valid in order to ensure memory safety.
Take a look at an example:
fn shortest_route<'a>(a: &'a i32, b: &'a i32) -> &'a i32 {
if a > b {
b} else {
a}
}
fn main () {
let a = 300000;
let b = 100000;
println!("{}km", shortest_route(&a, &b));
}
Here’s the output:
100000km
Here, a function called shortest_route
is created that takes two signed integers (ie &i32
) as arguments and returns the shortest route between the two distances. The <'a>
syntax specifies that the lifetime 'a
is a generic lifetime parameter. This means that the function can take any two signed integer values with the same lifetime, and the return value has the same lifetime as the input slices.
Structs With Lifetimes
Lifetimes are often used when defining structs in Rust, particularly when a struct contains references to other values. Consider the following example:
struct ImportantExcerpt<'a> {
: &'a str,
part}
impl<'a> ImportantExcerpt<'a> {
fn new(part: &'a str) -> ImportantExcerpt<'a> {
{ part }
ImportantExcerpt }
fn announce_and_return_part(&self, announcement: &str) -> \
&str {
println!("Attention please: {}", announcement);
self.part
}
}
fn main() {
let novel = String::from("Call me Ishmael. Some years ago...");
let first_sentence = \
.split('.').next().expect("Could not find a '.'");
novellet i = ImportantExcerpt::new(first_sentence);
println!("{}", i.announce_and_return_part("Hello, world!"));
}
Your output would be:
Attention please: Hello, world!
Call me Ishmael
In this example, a struct ImportantExcerpt
is created and contains a reference to a string slice. The lifetime 'a
is used to specify that the reference to the string slice must live at least as long as the struct instance. The impl
block defines methods for the struct, including a constructor new
and a method announce_and_return_part
that returns the referenced string slice.
Here’s another example using this same code above, but this code could lead to errors by just changing the values passed to the struct in the main
function:
fn main() {
let novel = String::from("Call me Ishmael. Some years ago...");
let first_sentence = \
.split('.').next().expect("Could not find a '.'");
novellet i = ImportantExcerpt::new(&first_sentence.to_string());
println!("{}", i.announce_and_return_part("Hello, world!"));
}
The key difference between the two main functions is how the ImportantExcerpt
struct references data. In the first function, it holds a reference to a string slice (&str)
, while in the second function, it holds a reference to a String created using the to_string()
method. This distinction has implications for the lifetime of the referenced data.
If the ImportantExcerpt
struct were to hold a reference to a String
with a shorter lifetime than the struct itself, it would result in a dangling reference error. However, Rust’s compile-time checks prevent this error from occurring, which is shown below:
| let i = ImportantExcerpt::new(&first_sentence.to_string());
| ^^^^^^^^^^^^^^^^^^^^^^^^^^ - temporary value is freed at the end of this statement
| |
| creates a temporary which is freed while still in use
| println!("{}", i.announce_and_return_part("Hello, world!"));
| - borrow later used here
Lifetimes in Trait Implementations
Trait implementations can also use lifetimes when dealing with references. Here’s an example:
trait Summary<'a> {
fn summarize(&'a self) -> String;
}
struct NewsArticle<'a> {
: &'a str,
headline: &'a str,
location: &'a str,
author: &'a str,
content}
impl<'a> Summary<'a> for NewsArticle<'a> {
fn summarize(&'a self) -> String {
format!("{} by {} ({})", self.headline, self.author, self.location)
}
}
fn main() {
let article = NewsArticle {
: "A brand new world",
headline: "New York",
location: "John Doe",
author: "This is the content of the article",
content};
println!("{}", article.summarize());
}
And here’s the output:
A brand new world by John Doe (New York)
In this code block, the Summary
trait has a lifetime parameter 'a
that is used to define the lifetime of the reference in the summarize
method. The NewsArticle
struct also has a lifetime parameter 'a
that is used to define the lifetime of its fields. The implementation of the summary trait for the NewsArticle
struct specifies that the &self
reference has the same lifetime as the lifetime of the fields in the NewsArticle
struct.
By using lifetimes in trait implementations, you can ensure that the lifetimes of references are properly managed and avoid any potential memory safety issues.
Advanced Topics
In addition to the core concepts of Rust lifetimes, there are some advanced topics worth exploring. They include Lifetime Subtyping and Higher-Ranked trait bounds.
Lifetime Subtyping
Lifetime subtyping is a concept that allows lifetimes to be compared and ordered based on their relationship to one another. This can be useful when working with functions or data structures that require multiple lifetimes or when defining traits with lifetime constraints.
For example, imagine a function that takes two references with different lifetimes and returns a reference with a lifetime that is a sub-lifetime of both inputs. To demonstrate lifetime subtyping, this relationship can be expressed using the 'a
and 'b
lifetime parameters:
fn lifetime_subtyping<'a, 'b: 'a>(x: &'a str, y: &'b str) -> &'a str {
if x.len() > y.len() { x } else { y }
}
fn main() {
let string1 = String::from("abcd");
let string2 = "xyz";
let result = lifetime_subtyping(string1.as_str(), string2);
println!("The longest string is {}", result);
}
This is what your output would look like:
The longest string is abcd
In this example, the returned reference has a lifetime of 'b
, which is a sub-lifetime of 'a
. This will ensure the 'b
lifetime reference outlives 'a
.
Higher-Ranked Trait Bounds
Higher-ranked trait bounds are a type of trait bound that allows a function or data structure to specify requirements on lifetimes that are not directly related to the input or output lifetimes. This can be useful when working with complex data structures or algorithms that require more fine-grained control over memory usage.
Consider the following example:
// Define a trait with a method that takes a closure with a \
// reference parameter.
trait RefProcessor {
fn process_refs<F>(&self, f: F)
where
: Fn(&i32);
F}
// Implement the RefProcessor trait for a Vec<i32>.
impl RefProcessor for Vec<i32> {
fn process_refs<F>(&self, f: F)
where
: Fn(&i32),
F{
for item in self {
;
f(item)}
}
}
// A function that takes a type implementing RefProcessor and a closure \
// with a generic lifetime.
fn process_all_items<T>(ref_processor: &T, \
: impl for<'a> Fn(&'a i32))
closurewhere
: RefProcessor,
T{
.process_refs(closure);
ref_processor}
fn main() {
let numbers = vec![1, 2, 3, 4, 5];
// Pass a closure that prints the square of each item.
&numbers, |x| println!("{}", x * x));
process_all_items(}
Your output would look like this:
1
4
9
16
25
Here, a trait called RefProcessor
is defined with a method called process_refs
that takes a closure with a reference parameter. The trait is implemented for the Vec<i32>
type, and a function called process_all_items
is created that accepts a type implementing RefProcessor
and a closure with a generic lifetime. The higher-ranked trait bound is specified as impl for<'a> Fn(&'a i32)
for the closure, indicating that the closure can work with references of any lifetime.
Associated Types and Lifetimes
Associated types and lifetimes can be used to enforce complex relationships between data structures and their associated lifetimes. Associated types are a way of associating a type with a trait, while lifetimes can be used to specify the scope in which a reference remains valid.
For example, imagine a trait that defines a method for iterating over a data structure. You might want to associate the type of the iterator with the trait while also specifying a lifetime for the reference to the data structure:
trait Iter<'a> {
type Item;
type Iter: Iterator<Item = Self::Item> + 'a;
fn iter(&'a self) -> Self::Iter;
}
// using the Iter<'a> trait
struct List<T> {
: Link<T>,
head}
type Link<T> = Option<Box<Node<T>>>;
struct Node<T> {
: T,
elem: Link<T>,
next}
impl<T> List<T> {
fn new() -> Self {
{ head: None }
List }
}
impl<'a, T: 'a> Iter<'a> for List<T> {
type Item = &'a T;
type Iter = ListIter<'a, T>;
// Returns an iterator over the list.
fn iter(&'a self) -> Self::Iter {
{ next: self.head.as_ref().map(|node| &**node) }
ListIter }
}
struct ListIter<'a, T> {
: Option<&'a Node<T>>,
next}
impl<'a, T> Iterator for ListIter<'a, T> {
type Item = &'a T;
// Returns a reference to the current item and moves the iterator \
// to the next item.
fn next(&mut self) -> Option<Self::Item> {
match self.next {
Some(node) => {
self.next = node.next.as_ref().map(|node| &**node);
Some(&node.elem)
}
None => None,
}
}
}
fn main() {
let mut list = List::new();
.head = Some(Box::new(Node { elem: 1, next: None }));
list.head = Some(Box::new(Node { elem: 2, next: list.head }));
list.head = Some(Box::new(Node { elem: 3, next: list.head }));
list.head = Some(Box::new(Node { elem: 4, next: list.head }));
list.head = Some(Box::new(Node { elem: 5, next: list.head }));
list
for i in list.iter() {
println!("{}", i);
}
}
Here’s the output:
5
4
3
2
1
In this example, the trait created is used to create an iterable list of nodes where the type
keyword is used to define two associated types: Item
, which represents the type of item returned by the iterator, and Iter
, which represents the type of iterator itself. A lifetime parameter 'a
is also specified, which is used to ensure that the reference to the data structure remains valid for as long as the iterator is being used.
Conclusion
Although Rust’s approach to memory management through ownership and borrowing and its use of lifetimes to track references and prevent memory leaks can be challenging to understand, with practice and experience, you can master these concepts and become a proficient Rust developer.
In this article, you learned all about the basics of Rust lifetimes, including borrowing and references, lifetime syntax, and annotations. You also explored advanced topics, such as lifetime subtyping, higher-ranked trait bounds, and associated types and lifetimes. All the code samples for this tutorial are available in this GitHub repo.
With the information and examples provided here, you should have a solid understanding of Rust’s lifetimes and their importance in writing safe and efficient code. If you want to keep learning, try exploring the official Rust documentation and community resources.
Earthly Cloud: Consistent, Fast Builds, Any CI
Consistent, repeatable builds across all environments. Advanced caching for faster builds. Easy integration with any CI. 6,000 build minutes per month included.