Ownership Borrowing

Rust Ownership and Borrowing

Understand and apply Rust's ownership system to write memory-safe code without a garbage collector. Master borrowing rules, lifetimes, and smart pointers.

When to Use

• Encountering borrow checker errors ("cannot borrow as mutable")
• Deciding between references, cloning, and moving
• Adding lifetime annotations to structs or functions
• Choosing between Box, Rc, Arc for heap allocation and sharing
• Understanding when to derive Clone vs Copy

Core Patterns

Pattern 1: Ownership Rules

Three rules govern all Rust memory:

1. Each value has exactly one owner.
2. When the owner goes out of scope, the value is dropped.
3. Ownership can be transferred (moved) but not duplicated (unless Copy).

fn main() {
    let name = String::from("Alice"); // name owns the String
    let greeting = greet(name);       // ownership moves to greet
    // println!("{name}");            // ERROR: name was moved
    println!("{greeting}");
}

fn greet(name: String) -> String {    // takes ownership
    format!("Hello, {name}!")         // returns a new owned String
}

Pattern 2: Borrowing -- Shared and Mutable References

Borrow data without taking ownership. Two rules:

• Any number of shared references (&T) OR exactly one mutable reference (&mut T).
• References must always be valid (no dangling pointers).

fn analyze(data: &[i32]) -> (i32, i32) {
    // Shared borrow: can read, cannot modify
    let sum: i32 = data.iter().sum();
    let count = data.len() as i32;
    (sum, count)
}

fn normalize(data: &mut Vec<f64>) {
    // Mutable borrow: can read and modify
    let max = data.iter().cloned().fold(f64::NEG_INFINITY, f64::max);
    if max != 0.0 {
        for val in data.iter_mut() {
            *val /= max;
        }
    }
}

fn main() {
    let mut values = vec![1.0, 2.0, 3.0];
    normalize(&mut values);     // mutable borrow
    let (sum, _) = analyze(&[1, 2, 3]); // shared borrow
    println!("{sum}");
}

Pattern 3: Lifetimes

Lifetimes tell the compiler how long references are valid. Most are inferred; annotate when the compiler cannot determine the relationship.

// The returned reference lives as long as the shortest input lifetime
fn longest<'a>(a: &'a str, b: &'a str) -> &'a str {
    if a.len() >= b.len() { a } else { b }
}

// Lifetime in structs: the struct cannot outlive the referenced data
struct Excerpt<'a> {
    text: &'a str,
}

impl<'a> Excerpt<'a> {
    fn first_word(&self) -> &'a str {
        self.text.split_whitespace().next().unwrap_or("")
    }
}

fn main() {
    let novel = String::from("Call me Ishmael. Some years ago...");
    let excerpt = Excerpt {
        text: novel.split('.').next().unwrap(),
    };
    println!("{}", excerpt.first_word());
}

Pattern 4: Smart Pointers

Use smart pointers when ownership rules need more flexibility.

// Box<T>: heap allocation, single owner
fn build_tree() -> Box<Node> {
    Box::new(Node {
        value: 1,
        left: Some(Box::new(Node { value: 2, left: None, right: None })),
        right: Some(Box::new(Node { value: 3, left: None, right: None })),
    })
}

// Rc<T>: shared ownership, single-threaded (reference counted)
use std::rc::Rc;

fn shared_config() {
    let config = Rc::new(AppConfig::default());
    let service_a = Service::new(Rc::clone(&config));
    let service_b = Service::new(Rc::clone(&config));
    // Both services share the same config; dropped when last Rc is dropped
}

// Arc<T>: shared ownership, thread-safe (atomic reference counted)
use std::sync::Arc;

fn concurrent_cache() {
    let cache = Arc::new(Mutex::new(HashMap::new()));

    let handles: Vec<_> = (0..4)
        .map(|i| {
            let cache = Arc::clone(&cache);
            std::thread::spawn(move || {
                cache.lock().unwrap().insert(i, i * 10);
            })
        })
        .collect();

    for h in handles {
        h.join().unwrap();
    }
}

Pattern 5: Clone vs Copy

Copy: bitwise copy, implicit, for small stack types (integers, bools, tuples of Copy types). Clone: explicit .clone(), for types that need deep duplication.

// Copy: automatically duplicated on assignment
let x: i32 = 42;
let y = x;      // x is copied, both x and y are valid
println!("{x} {y}");

// Clone: explicit duplication
let a = String::from("hello");
let b = a.clone(); // deep copy -- both a and b are valid
println!("{a} {b}");

// Without clone, a would be moved:
let c = String::from("world");
let d = c;        // c is MOVED to d
// println!("{c}"); // ERROR: c was moved

// Derive both when your struct contains only Copy types
#[derive(Debug, Clone, Copy)]
struct Point {
    x: f64,
    y: f64,
}

Pattern 6: Common Borrow Checker Patterns

Solutions to frequent borrow checker issues.

// Problem: cannot borrow as mutable because also borrowed as immutable
// Solution: limit the scope of the immutable borrow
fn update_map(map: &mut HashMap<String, Vec<i32>>, key: &str) {
    // Use entry API to avoid double-borrow
    map.entry(key.to_string())
        .or_insert_with(Vec::new)
        .push(42);
}

// Problem: returning a reference to a local variable
// Solution: return owned data
fn create_greeting(name: &str) -> String {
    // Return owned String, not &str
    format!("Hello, {name}!")
}

// Problem: self-referential struct
// Solution: use indices instead of references, or use Pin
struct Document {
    content: String,
    // Instead of &str references into content, store byte offsets
    highlights: Vec<(usize, usize)>,
}

impl Document {
    fn highlighted_text(&self, idx: usize) -> &str {
        let (start, end) = self.highlights[idx];
        &self.content[start..end]
    }
}

Anti-Patterns

• Cloning to silence the borrow checker -- .clone() everywhere is a sign of misunderstood ownership. Restructure code to use borrows instead.

• Unnecessary Arc<Mutex<T>> -- Only use Arc when data is shared across threads. Single-threaded code should use Rc or plain ownership.

• Lifetime annotations on everything -- Let the compiler infer lifetimes. Only annotate when the compiler asks or when the relationship is ambiguous.

• Returning references from functions that create data -- Functions that allocate must return owned types. References can only point to data that outlives them.

Quick Reference

Move vs Copy: types implementing Copy are duplicated on assignment; all others are moved.