12 iOS Swift and Types Interview Questions and Answers

Short answer: A struct is a value type: it is copied on assignment and when passed to a function. A class is a reference type: variables share one instance. Inheritance, deinit and identity (===) exist only for classes; mutating a struct requires var and mutating methods.

In depth:

1. Default to struct — predictable copying, no races on shared state, cheaper for ARC.
2. Class — when you need shared identity (a cache, a service), inheritance from the NSObject world, or deinit to release a resource.

⚠️ Common mistake: "structs always live on the stack." No: a struct stored in a class property, captured by an escaping closure, or boxed in an existential with a large payload lives on the heap. Value type defines copy semantics, not memory placement.

Short answer: Assigning an array of structs yields an independent copy — mutating the copy doesn't touch the original. Assigning an array of class instances copies only the references — the objects are shared, so the mutation is visible through both variables.

In depth:

struct Point { var x = 0 }
final class Box { var x = 0 }

var a = [Point(), Point()]
var b = a
b[0].x = 99
print(a[0].x)      // 0  — b has its own copy of the struct array

let c = [Box(), Box()]
let d = c
d[0].x = 99
print(c[0].x)      // 99 — only references were copied, Box is shared

1. Array is itself a struct — b = a is semantically a copy; physically the buffer is shared until the first mutation (CoW), but the observable behavior is a full copy.
2. Class elements — pointers get copied; d[0] and c[0] are the same heap object.
3. let d doesn't protect you — let pins the reference, not the object's state: class fields can still be mutated.

⚠️ Common mistake: answering "99 and 99", forgetting that the array's value semantics extends to its elements: b[0].x = 99 mutates an element inside the copy, not a shared object.

Short answer: CoW is an optimization of value semantics: assignment copies only the reference to a shared buffer, and the real copy happens on mutation, and only if the buffer is shared (isKnownUniquelyReferenced). Array/String/Dictionary implement it by hand. Your own struct does not get CoW automatically.

In depth:

1. Assignment — O(1): both variables point at one heap buffer, reference count = 2.
2. Mutation — a uniqueness check; if the buffer is shared, copy first, then write.
3. Your own struct — if it holds a reference to class-backed storage, only the pointer is copied, and a mutation "leaks" into every copy. You write the check yourself:

final class Storage { var values: [Int] = [] }

struct Buffer {
    private var storage = Storage()
    mutating func set(_ v: Int, at i: Int) {
        if !isKnownUniquelyReferenced(&storage) {
            storage = storage.copy()   // copy only when the buffer is shared
        }
        storage.values[i] = v
    }
}

⚠️ Common mistake: "all value types in Swift copy via CoW." No: a plain struct is copied bitwise immediately; CoW is a manual pattern — out of the box only the standard collections have it.

Short answer: An Optional is a plain generic enum with two cases: .some(Wrapped) and .none. T? is syntactic sugar for Optional<T>, and nil is a literal for .none. Force unwrap (!) is tolerable for invariants the programmer guarantees, but in business-logic flow it's a code smell.

In depth:

// Literally in the standard library:
enum Optional<Wrapped> {
    case none
    case some(Wrapped)
}

guard let user = findUser(id) else { return }  // early exit
if let name = user.nickname { greet(name) }    // local unwrap
let count = user.nickname?.count ?? 0          // chaining + nil-coalescing

The unwrapping ladder — from preferred to last resort:

1. guard let — an invariant for the rest of the function, early exit.
2. if let / ?. / ?? — local work with the value and defaults.
3. ! — only for programmer-guaranteed invariants: an IBOutlet after the view loads, a bundled resource, a fixture in tests. A crash here is a build-time signal, not a user scenario.

⚠️ Common mistake: "never use !." The mature answer is about invariants: URL(string: "https://api.example.com")! with a constant literal is more honest than a silent ?? with a meaningless default that masks a bug.

Short answer: A protocol with an associatedtype is not a concrete type but a template of requirements: the compiler doesn't know what T is. Before Swift 5.7 such a protocol couldn't be used as a type at all; now you can write any P — an existential box with dynamic dispatch, but the associated type comes out erased. The alternatives are generics or type erasure.

In depth:

protocol Repository {
    associatedtype Entity
    func all() -> [Entity]
}

// let repos: [Repository] = []   // ✗ before 5.7: 'Repository' can only be used
//                                //   as a generic constraint
let repos: [any Repository] = []  // ✓ boxed, but Entity is erased outside

func sync<R: Repository>(_ repo: R) -> [R.Entity] {
    repo.all()                    // ✓ generic: Entity is known to the compiler
}

1. any P — a heterogeneous collection becomes possible, but the result of all() loses its concrete Entity and goes through witness-table indirection.
2. Generic <R: Repository> — full access to R.Entity, static dispatch; but one call means one concrete type.
3. Type erasure — wrappers like AnyPublisher, AnySequence: they hide the concrete type while keeping the associated one (AnyPublisher<Int, Never>).

⚠️ Common mistake: fixing the compiler error by slapping on any without thinking. If the collection is homogeneous, a generic is both faster and type-preserving.

Short answer: some P is an opaque type: behind it stands exactly one concrete type known to the compiler ("reverse generics") — static dispatch, no boxing. any P is an existential box: the value is stored with witness-table indirection, and a large payload moves to the heap. <T: P> is a generic, monomorphized per call site.

In depth:

1. some — the implementation picks the type, not the caller; the type is stable across calls.
2. any — honest runtime polymorphism: [any Shape] with circles and squares mixed together.
3. Generic — the caller picks the type; the compiler clones and optimizes the code for each T.

⚠️ Common mistake: writing any everywhere "for convenience." That's a silent cost: boxing, indirect calls, loss of associated types. Rule of thumb: start with a generic or some; reach for any when you genuinely need heterogeneity.

Short answer: @escaping marks a closure that outlives the function call — it gets stored in a property or handed to asynchronous work. This changes two things: captures move to the heap (the closure lives longer than the stack frame), and the compiler requires explicit self, making capture semantics visible.

In depth:

final class Loader {
    var completion: (() -> Void)?

    func load(_ done: @escaping () -> Void) {
        completion = done                        // stored → escaping is mandatory
        DispatchQueue.global().async { done() }  // goes async → escaping too
    }

    func forEachItem(_ body: () -> Void) {
        body()                                   // non-escaping: dies before return
    }
}

1. Non-escaping (the default) — the closure is guaranteed to run before the function returns: captures can stay on the stack, self is implicit.
2. Escaping — lifetime is unknown: the capture context is heap-allocated, and the explicit self. is a signal saying "a retain cycle is possible here, consider [weak self]".
3. Where you meet it — network completion handlers, stored callbacks, DispatchQueue.async.

⚠️ Common mistake: saying @escaping "makes the closure asynchronous." No: it's only about lifetime. An escaping closure may run synchronously — or never at all.

Short answer: Each state is a case and its data are associated values: .loaded([Item]) physically cannot exist without the array, .error(Error) — without an error. Impossible states become unrepresentable, and switch forces you to handle every case.

In depth:

enum ScreenState {
    case loading
    case loaded([Item])
    case error(Error)
}

switch state {
case .loading:              showSpinner()
case .loaded(let items):    render(items)
case .error(let e):         showRetry(e)
}   // add a case — the compiler flags every site

1. Versus bool flags — isLoading + items: [Item]? + error: Error? gives 8 combinations of which 3 are valid: "loading and error at the same time" has to be banned by discipline. The enum bans them with types.
2. Data is bound to the case — you can't reach items until a switch proves the state is .loaded.
3. indirect — for recursive enums (a tree, JSON): indirect case node(Tree, Tree) — Swift boxes the recursion behind a pointer.

⚠️ Common mistake: putting default: in a switch over your own enum. It turns off the main benefit — exhaustiveness checking: a new case silently falls into default and the compiler stays quiet.

Short answer: Three mechanisms: direct (static) — the address is known at compile time; table-based — a vtable for classes and a witness table for protocols; message dispatch — objc_msgSend for @objc dynamic. The famous trap: a method from a protocol extension that is NOT declared as a protocol requirement dispatches statically — by the variable's type.

In depth:

1. Direct — struct methods, final methods, methods in extensions. Fastest, can be inlined.
2. Vtable / witness table — regular class methods; protocol requirements when called through any P.
3. Message — @objc dynamic: objc_msgSend, needed for KVO and swizzling.

protocol Greeter { func hello() }
extension Greeter {
    func hello() { print("protocol hello") }
    func bye()   { print("protocol bye") }   // NOT a protocol requirement
}
struct Person: Greeter {
    func hello() { print("Person hello") }
    func bye()   { print("Person bye") }
}

let g: any Greeter = Person()
g.hello()   // "Person hello"  — requirement → witness table → the type's impl
g.bye()     // "protocol bye"  — not a requirement → static, by the any Greeter type

let p = Person()
p.bye()     // "Person bye"    — concrete type is known

⚠️ Common mistake: expecting "Person bye" from g.bye(). The method isn't in the requirements list — the witness table doesn't know about it, so the compiler hardwires the extension version. Second trap: a class method defined in an extension can't be overridden in a subclass without @objc dynamic — extension methods dispatch directly.

Short answer: Both are members of the type, not of an instance. static is available everywhere (struct, enum, class) and forbids overriding — it is literally final class. class exists only in classes and allows override in subclasses. Stored class properties don't exist — computed only.

In depth:

class Animal {
    static func kind() -> String { "animal" }   // = final class func
    class  func sound() -> String { "..." }     // can be overridden

    // class var name = "x"      // ✗ class stored properties not supported
    class var legs: Int { 4 }    // ✓ computed — allowed
    static let planet = "Earth"  // ✓ stored — static only
}

class Dog: Animal {
    override class func sound() -> String { "woof" }
    // override static func kind() ...  // ✗ cannot override static method
}

Dog.sound()   // "woof" — dispatched on the metatype via the vtable

1. static — static dispatch, works in every type; stored type properties come only this way (and they initialize lazily and thread-safely).
2. class — participates in dynamic dispatch on the metatype: type(of: obj).sound() respects overrides.
3. Choosing — default to static; use class when type-level polymorphism is genuinely needed (factories, reuseIdentifier).

⚠️ Common mistake: "static is for structs, class is for classes." No: static lives happily in classes and means final class there; the difference is overridability, not where you can use it.

Short answer: On first access: it computes once, gets stored, and every later read returns the stored value. And no, it is not atomic: two threads reading an uninitialized lazy var simultaneously can run the initializer twice. Unlike global let and static let, which have dispatch_once semantics.

In depth:

final class ViewModel {
    lazy var formatter: DateFormatter = {
        let f = DateFormatter()      // runs on first access
        f.dateFormat = "dd.MM.yyyy"
        return f
    }()
}

// static let shared = ViewModel()  // this one IS lazy AND thread-safe

1. Why — defer expensive initialization (formatters, values computed from other properties); self is already available inside the closure.
2. Always var — the value is written after init, so lazy let doesn't exist.
3. Threads — the compiler adds no lock: a race between two threads can double-initialize. Need laziness plus thread safety — use static let or explicit synchronization.
4. In a struct — the first read mutates storage: you need a var instance, and reading it on a let struct is a compile error.

⚠️ Common mistake: assuming lazy var is thread-safe "like a singleton." The dispatch_once semantics belongs to globals and static constants — not to instance lazy properties.

Short answer: For @Wrapper var x the compiler synthesizes hidden storage private var _x: Wrapper and rewrites accesses to x as _x.wrappedValue. projectedValue is the wrapper's second, optional value, reachable via $x: in SwiftUI, $text on a @State is a Binding.

In depth:

@propertyWrapper
struct Clamped {
    private var value: Int
    let range: ClosedRange<Int>

    var wrappedValue: Int {
        get { value }
        set { value = min(max(newValue, range.lowerBound), range.upperBound) }
    }
    var projectedValue: Bool { value == range.upperBound }  // this is what $x is

    init(wrappedValue: Int, _ range: ClosedRange<Int>) {
        self.range = range
        self.value = min(max(wrappedValue, range.lowerBound), range.upperBound)
    }
}

struct Player {
    @Clamped(0...100) var health = 100
    // the compiler generated: private var _health: Clamped
}
var p = Player()
p.health = 250
print(p.health, p.$health)   // 100 true

1. Why — reusable access logic: validation, UserDefaults, thread safety — without copy-pasting it into every property.
2. Well-known wrappers — @Published ($ is a Publisher), @State/@Binding ($ is a Binding), @AppStorage.

⚠️ Common mistake: confusing $x with _x. _x is the wrapper storage itself (visible inside the type), $x is its projectedValue; if projectedValue isn't declared, the $x syntax simply doesn't exist.