Arc

You have an Arc<T>, you want a &mut T. Arc only hands out &T, so the usual workaround is clone-the-inner, mutate, rewrap. Arc::make_mut does that for you — and skips the clone when no one else is watching.

The manual version everyone writes once and then copies forever:

1
2
3
4
5
6
7
8
9

use std::sync::Arc;

let mut shared = Arc::new(vec![1, 2, 3]);

let mut owned: Vec<i32> = (*shared).clone(); // always clones
owned.push(4);
shared = Arc::new(owned);                    // always reallocates the Arc

assert_eq!(*shared, vec![1, 2, 3, 4]);

It works, but it clones the Vec and reallocates the Arc every single time — even when this Arc is the only one pointing at the data.

Arc::make_mut takes &mut Arc<T> and hands you &mut T:

1
2
3
4
5
6
7

use std::sync::Arc;

let mut shared = Arc::new(vec![1, 2, 3]);

Arc::make_mut(&mut shared).push(4);

assert_eq!(*shared, vec![1, 2, 3, 4]);

One call, one borrow, and — crucially — no clone when this Arc is unique:

1
2
3
4
5

use std::sync::Arc;

let mut solo = Arc::new(vec![1, 2, 3]);
Arc::make_mut(&mut solo).push(99); // strong_count == 1, mutates in place
assert_eq!(*solo, vec![1, 2, 3, 99]);

When the Arc is shared, make_mut quietly clones the inner value into a fresh allocation and detaches your handle from the rest. The other handles keep seeing the old data — clone-on-write, exactly like you’d want:

1
2
3
4
5
6
7
8
9

use std::sync::Arc;

let mut a = Arc::new(vec![1, 2, 3]);
let b = Arc::clone(&a);            // strong_count == 2

Arc::make_mut(&mut a).push(99);    // clones, then mutates the clone

assert_eq!(*a, vec![1, 2, 3, 99]); // a moved to its own allocation
assert_eq!(*b, vec![1, 2, 3]);     // b still sees the original

The same method exists on Rc for single-threaded code, with identical semantics. Reach for make_mut whenever you find yourself cloning the inside of an Arc just to change one field — you’ll skip the allocation in the common case and get an honest &mut T in return.

You need to own a T but all you have is an Arc<T>. The old pattern is a six-line fumble with try_unwrap. Arc::unwrap_or_clone collapses it into one call — and skips the clone entirely when it can.

The old dance

Arc::try_unwrap hands you the inner value — but only if you’re the last reference. Otherwise it gives your Arc back, and you have to clone.

1
2
3
4
5
6
7
8

use std::sync::Arc;

let arc = Arc::new(String::from("hello"));
let owned: String = match Arc::try_unwrap(arc) {
    Ok(inner) => inner,
    Err(still_shared) => (*still_shared).clone(),
};
assert_eq!(owned, "hello");

Every place that wanted an owned T from an Arc<T> wrote this same pattern, often subtly wrong.

The fix: unwrap_or_clone

Stabilized in Rust 1.76, Arc::unwrap_or_clone does exactly the right thing: move the inner value out if we’re the last owner, clone it otherwise.

1
2
3
4
5

use std::sync::Arc;

let arc = Arc::new(String::from("hello"));
let owned: String = Arc::unwrap_or_clone(arc);
assert_eq!(owned, "hello");

One call. No match. No deref gymnastics.

It actually skips the clone

The key win isn’t just ergonomics — it’s performance. When the refcount is 1, no clone happens; the T is moved out of the allocation directly.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

use std::sync::Arc;

let solo = Arc::new(vec![1, 2, 3, 4, 5]);
let v: Vec<i32> = Arc::unwrap_or_clone(solo); // no allocation, just a move
assert_eq!(v, [1, 2, 3, 4, 5]);

let shared = Arc::new(vec![1, 2, 3]);
let _other = Arc::clone(&shared);
let v2: Vec<i32> = Arc::unwrap_or_clone(shared); // clones, because _other still holds a ref
assert_eq!(v2, [1, 2, 3]);

Also on Rc

The same method exists on Rc for single-threaded code — identical semantics, identical ergonomics:

1
2
3
4
5

use std::rc::Rc;

let rc = Rc::new(42);
let n: i32 = Rc::unwrap_or_clone(rc);
assert_eq!(n, 42);

Anywhere you were reaching for try_unwrap().unwrap_or_else(|a| (*a).clone()), reach for unwrap_or_clone instead. Shorter, clearer, and it avoids the clone when it can.