Thai Dao

Posted on Aug 13

Value Semantics vs Reference Semantics in Swift: A Practical Guide

#swift #ios

When learning Swift, we often hear:

struct is a Value Type, while class is a Reference Type.

That statement is correct, but it doesn't explain what actually happens when we copy, mutate, or share data.

In real-world iOS development, understanding Value Semantics, Reference Semantics, Copy-on-Write (CoW), and storage is much more useful than simply remembering "struct = stack, class = heap".

Let's break it down.

1. Value Semantics
A type has Value Semantics when copying it creates an independent value.

var a = 10
var b = a

b = 20

print(a) // 10
print(b) // 20

Changing b doesn't affect a.

The same applies to a struct:

struct User {
    var name: String
}

var user1 = User(name: "Alice")
var user2 = user1

user2.name = "Bob"

print(user1.name) // Alice
print(user2.name) // Bob

Conceptually:

user1 → User("Alice")
user2 → User("Bob")

The important idea is:

A copy behaves as an independent value.

2. Reference Semantics
A class has Reference Semantics.

final class User {
    var name: String

    init(name: String) {
        self.name = name
    }
}

let user1 = User(name: "Alice")
let user2 = user1

user2.name = "Bob"

-------------------
Now:
print(user1.name) // Bob
print(user2.name) // Bob

-------------------
Why?
Because `user1` and `user2` reference the same object:

user1 ─────┐
           ▼
       User Object
       name = "Bob"
           ▲
           │
user2 ─────┘

Copying a reference type copies the reference, not the object.

This is useful when we actually want shared identity and shared state—for example, a ViewModel, service, coordinator, or resource owner.

3. Don't Think struct = Stack and class = Heap
A common explanation is:

struct → Stack
class  → Heap

This is too simplistic.

Stack and Heap describe storage implementation, while Value and Reference Semantics describe behavior.

For example:

struct Point {
    let x: Int
    let y: Int
}

A small, fixed-size value like this can be stored inline. If it's a local variable, that representation may be on the stack, but the compiler is free to optimize it differently.

On the other hand:

struct RingtoneStore {
    var ringtones: [Ringtone]
}

Array can contain 100,000 elements, so its data requires dynamic storage.

Conceptually:

RingtoneStore
┌─────────────────────┐
│ Array representation│
└──────────┬──────────┘
           │
           ▼
     Backing Storage
     [Ringtone ...]

The Array value is still a Value Type, even though its backing storage is dynamically allocated and commonly lives on the heap.

So the better question is not:

"Is this struct on the Stack or Heap?"
but:
"How is this value represented and where is its backing storage?"

4. Copy-on-Write
This becomes especially interesting with large collections.

Suppose we have 100,000 ringtones:

let ringtones = [Ringtone](...)

Now:

let a = ringtones
let b = a

Does Swift immediately copy all 100,000 elements?

No.

Array uses Copy-on-Write (CoW).

Conceptually:

a ─────┐
       ├──> Shared Storage
b ─────┘

There are two Array values, but they can temporarily share the same backing storage.

If we only read:

print(b[0])

there is no reason to copy the storage.

This makes copying a large Array relatively cheap.

5. What Happens When We Mutate?

Now:

var a = [1, 2, 3]
var b = a

b.append(4)

Before mutation:

a ─────┐
       ├──> Storage A
b ─────┘
       [1, 2, 3]

If b modified Storage A directly, a would also change. That would violate Value Semantics.

So Swift separates the storage:

a ─────────> Storage A
             [1, 2, 3]

b ─────────> Storage B
             [1, 2, 3, 4]

This is Copy-on-Write.

The important optimization is:

Don't copy until mutation actually requires it.

6. What Does "Uniquely Referenced" Mean?

CoW doesn't mean every mutation requires an O(n) copy.

Consider:

var a = [1, 2, 3]

a.append(4)

If a is the only value using its backing storage:

a ─────> Storage A

the storage is uniquely referenced.

Swift can mutate it directly.

But:

var a = [1, 2, 3]
var b = a

gives:

a ─────┐
       ├──> Storage A
b ─────┘

The storage is shared.

If b mutates, CoW may need to create a new storage.

So the simplified model is:

Mutation
   ↓
Is storage shared?
   │
   ├── No  → mutate in place
   │
   └── Yes → copy → mutate

This is the key to understanding the performance of Swift collections.

7. A Struct Can Still Contain a Class

Here's an important edge case:

final class AudioPlayer {
    var volume = 50
}

struct Ringtone {
    var name: String
    var player: AudioPlayer
}

Although Ringtone is a struct, player is a reference to a class instance.

After:

let ringtone2 = ringtone1

we can have:

ringtone1 ──┐
            │
            ▼
       AudioPlayer
       volume = 50
            ▲
            │
ringtone2 ──┘

Therefore:

ringtone2.player.volume = 100

can also be observed through:

ringtone1.player.volume

because both structs share the same `AudioPlayer`.

This gives us an important rule:

A struct provides Value Semantics at its own level, but it does not automatically make every property deeply independent.

This is why exposing mutable reference-type state directly from a value type should be done carefully.

8. Value Semantics and Concurrency

Value Semantics are also useful in concurrent code because they reduce shared mutable state.

For example:

struct UserState {
    var name: String
    var age: Int
}

If two tasks work with independent copies:

Task A → UserState A
Task B → UserState B

then mutation in Task A doesn't change Task B's value.

This significantly reduces opportunities for Data Races.

However, it's important to be precise:

struct does not automatically mean thread-safe.

A struct can still contain a mutable class reference or access external shared state.

So a better statement is:

Value Semantics reduce shared mutable state, which makes concurrent code easier to reason about and reduces potential data races.

Conclusion

The most useful mental model is not:

struct = Stack
class  = Heap

Instead:

Value Type
    ↓
Value Semantics
    ↓
Independent state
    ↓
CoW can optimize storage sharing

and:

Reference Type
    ↓
Reference Semantics
    ↓
Shared identity
    ↓
Shared mutable state
    ↓
ARC manages lifetime

And when performance is involved:

Array copy
    ↓
Share backing storage
    ↓
Read → no copy
    ↓
Mutation
    ↓
Is storage unique?
   ├── Yes → mutate in place
   └── No  → Copy-on-Write

Once you understand these relationships, concepts like Swift Concurrency, Sendable, ARC, data races, and collection performance become much easier to reason about.

The key takeaway is simple: