iOS8 Day-by-Day :: Day 1 :: Blagger’s Guide to Swift

Written by Sam Davies

Updated 9 Apr 2015: This post has been updated to Swift 1.2

This post is part of a daily series of posts introducing the most exciting new parts of iOS8 for developers – #iOS8DayByDay. To see the posts you’ve missed check out the introduction page, but have a read through the rest of this post first!

To enjoy each of the 39 posts all in one place, you can now also download the iOS8 Day-by-Day eBook free from our website!


Swift Logo

It won’t have gone unnoticed that at WWDC this year, in addition to announcing iOS8, they also introduced a new programming language in the form of Swift. This is quite a different language from objective-C in that it is strongly-typed and includes some features common to more modern languages.

In the interests of embracing anything and everything that’s new and shiny, this blog series will exclusively use Swift. There is a wealth of information out there about how to learn Swift, and how to interact with the Cocoa libraries – in fact you can’t go wrong with starting out by reading through the official books:

You should also check out the official swift blog, and some of the other resources made available by Apple.

Since there is so much good info out there about how to use Swift, this post is not going to attempt to cover any of that. Instead, it’s going to run through some of the important gotchas and potential pain points when using Swift for the first time – especially when relating to the system frameworks.

There is an Xcode 6 playground which accompanies this post – including short samples for each of the sections. You can get hold of it on the ShinobiControls Github page – at

If you have any questions or suggestions of other things to add to this post then do let me know – I’ll try to keep it up to date throughout the blog series. Drop a comment below, or gimme a shout on twitter – @iwantmyrealname.


Swift formalises the concepts surround initialisation of objects somewhat – including designated -vs- convenience initialisers, and sets a very specific order of the operations to be called within the initialisation phases of an object. In the coming weeks, there will be an article as part of this series which will go into detail about how initialisation works in Swift, and how this affects any objective-C that you write – so look out for this.

There is one other fairly major difference in initialisation between Swift and objective-C, and that is return values and initialisation failure. In objective-C an initialiser looks a lot like this:

- (instancetype)init {
  self = [super init];
  if (self) {
    // Do some stuff
  return self;

Whereas in Swift:

init {
  variableA = 10

Notice that in objective-C the initialiser is responsible for ‘creating’ and then returning self, but there is no return statement in the Swift equivalent. This means that there is actually no way in which you can return a nil object, which is a pattern commonly used to indicate an initialisation failure in objC.

Note: This has indeed changed, and Swift 1.1 introduced failable initialisers. Refer to Apple’s Swift book for more information.

This is apparently likely to change in an upcoming release of the language, but for now the only workaround is to use class methods which return optional types:

class MyClass {
  class func myFactoryMethod() -> MyClass? {

Interestingly, factory methods on objective-C APIs are converted into initialisers in Swift, so this approach is not preferred. However, until language support arrives, it’s the only option for initialisers which have the potential to fail.


The concept of (im)mutability is not new to Cocoa developers – we’ve been used to using NSArray and its mutable counterpart NSMutableArray where appropriate, and even understand that we should always prefer the immutable version wherever possible. Swift takes this concept to the next level, and bakes immutability into the language as a fundamental concept.

The let keyword defines an immutable variable, which means that you can’t change what it represents. For example:

let a = MyClass()
a = MySecondClass() // Not allowed

This means that you can’t redefine something specified with the let keyword. Depending on the type of the object referred to, it might itself be immutable too. If it is a value type (such as a struct) then it will also be immutable. If it is a reference type, such as a class then it will be mutable.

To see this in action, consider the following struct:

struct MyStruct {
  let t: Int
  var u: String

If you define a variable struct1 with the var keyword then you get the following behaviour:

var struct1 = MyStruct(t: 15, u: "Hello")
struct1.t = 13 // Error: t is an immutable property
struct1.u = "GoodBye"
struct1 = MyStruct(t: 10, u: "You")

You can mutate the u property, since this is defined with var, and you can redefine the struct1 variable itself, again because this is defined with var. You can’t mutate the t property, since this is defined with let. Now take a look what happens when you define an instance of a struct using let:

let struct2 = MyStruct(t: 12, u: "World")
struct2.u = "Planet" // Error: struct2 is immutable
struct2 = MyStruct(t: 10, u: "Defeat") // Error: struct2 is an immutable ref

Here, not only are you unable to mutate the struct2 reference itself, but you are also unable to mutate the struct itself (i.e. the u property). This is because a struct is a value type.

The behaviour is subtly different with a class:

class MyClass {
  let t: Int
  var u: String

  init(t: Int, u: String) {
    self.t = t
    self.u = u

Defining a variable using var gives behaviour you might be used to from objective-C:

var class1 = MyClass(t: 15, u: "Hello")
class1.t = 13 // Error: t is an immutable property
class1.u = "GoodBye"
class1 = MyClass(t: 10, u: "You")

You can mutate both the reference itself, and any properties defined using var, but you are unable to mutate any properties defined with let. Compare this to the behaviour when the instance is defined with let:

let class2 = MyClass(t: 12, u: "World")
class2.u = "Planet" // No error
class2 = MyClass(t: 11, u: "Geoid") Error: class2 is an immutable reference

Here you are unable to mutate the reference itself, but you can still mutate any properties defined with var within the class. This is because a class is a reference type.

This behaviour is fairly easy to understand, and is well-explained in the language reference books. There is potential for confusion when looking at Swift collection types though.

An NSArray is a reference type. That is to say that when you create an instance of NSArray, you create an object and your variable is a pointer to the location of the array itself in memory – hence the asterisk in the objective-C definition. If you take a look back over what you’ve learnt about the semantics of reference and value types with respect to let and var then you can probably work out how they would behave. In fact, if you want a mutable version of an NSArray you have to use a different class – in the shape of NSMutableArray.

Swift arrays aren’t like this – they are value types instead of reference types. This means that they behave like a struct, not a class. Therefore, the let or var keyword not only specifies whether or not the variable can be redefined, but also whether or not the created array is mutable.

An array defined with var can both be reassigned, and mutated:

var array1 = [1,2,3,4]
array1.append(5)  // [1,2,3,4,5]
array1[0] = 27 // [27,2,3,4,5] array1 = [3,2] // [3,2]

But an array defined with let can be neither:

let array2 = [4,3,2,1]
array2.append(0) // Error: array2 is immutable
array2[2] = 36 // Error: array2 is immutable array2 = [5,6] // Error: cannot reassign an immutable reference

This is an area with a huge potential for confusion. Not only does it completely change the way we think about mutability for collections, but it also mixes up two previously distinct concepts. There is potential that this might be changed in a future release of the language – so keep an eye on the language definition.

A corollary of this is that since arrays are value types, they are passed by copy. NSArray instances are always passed by reference – so a method which takes an NSArray pointer will point to exactly the same chunk of memory. If you pass a Swift array into a method, it will receive a copy of that array. Depending on the type of the objects stored in that array this could either be a deep, or a shallow copy. Be aware of this whilst writing your code!

Strong Typing and AnyObject

Strong typing is seen as a great feature of Swift – it can allow for safer code, since what in objective-C would have been runtime exceptions can now be caught at compile time.

This is great, but as you’re working with the objective-C system frameworks you’ll notice a lot of this AnyObject type. This is the Swift equivalent of objective-C’s id. In many respects, AnyObject feels rather un-Swift-like. It allows you to call any methods it can find on it, but these will result in a run-time exception. In fact, it behaves almost exactly the same as id in objective-C. The difference is that properties and methods which take no arguments will return nil if that method/property doesn’t exist on the AnyObject:

let myString: AnyObject = "hello"
myString.cornerRadius // Returns nil

In order to work in a more Swift-like way with the Cocoa APIs, you’ll see the following pattern a lot:

func someFunc(parameter: AnyObject!) -> AnyObject! {
  if let castedParameter = parameter as? NSString {
    // Now I know I have a string :)

If you know that you’ve definitely been passed a string, you don’t necessarily need to guard around the cast:

let castedParameter = parameter as! NSString

A top-tip is to realise that casting arrays is really easy too. All arrays that you’ll receive from a Cocoa framework will be of the type [AnyObject], since NSArray doesn’t support generics. However, in nearly every case not only are all the elements of the same type, but they are of a known type. You can cast an entire array in both the conditional and unconditional ways expressed above, with the following syntax:

func someArrayFunc(parameter: [AnyObject]!) {
  let newArray = parameter as! [String]
  // Do something with your strings :)

Protocol Conformance

Protocols are well-understood in Swift – defined as follows:

protocol MyProtocol {
  func myProtocolMethod() -> Bool

One of the things you often want to do is test whether an object conforms to a specified protocol, which you could do as follows:

if let class1AsMyProtocol = class1 as? MyProtocol {
  // We're in

Note: Swift 1.2 allows checking conformance of Swift-only protocols, so the following is no longer accurate.

However, this will have an error, because in order to check conformance of a protocol that protocol must be an objective-C protocol – and annotated with @objc:

@objc protocol MyNewProtocol {
  func myProtocolMethod() -> Bool

if let class1AsMyNewProtocol = class1 as? MyNewProtocol {
  // We're in

This can actually be more effort than you’d expect, since in order that a protocol be labelled as @objc, all of its properties and method return types must also be understood in the objective-C world. This means that you might end up annotating loads of classes you thought you only cared about in Swift with @objc.


Enums in Swift have become super-charged. Not only can an enum now have associated values (which needn’t be of the same type), but also contain functions too.

enum MyEnum {
  case FirstType
  case IntType (Int)
  case StringType (String)
  case TupleType (Int, String)

  func prettyFormat() -> String {
    switch self {
    case .FirstType:
      return "No params"
    case .IntType(let value):
return "One param: \(value)"
case .StringType(let value): return "One param: \(value)" case .TupleType(let v1, let v2): return "Some params: \(v1), \(v2)" default: return "Nothing to see here" } } }

This is really powerful – use it as follows:

var enum1 = MyEnum.FirstType
enum1.prettyFormat() // "No params"
enum1 = .TupleType(12, "Hello")
enum1.prettyFormat() // "Some params: 12, Hello"

It’ll take a little practice to see where you can get some benefit out of the power of these, but just as an indication of what you can achieve – the optionals system within Swift is built out of enumerations.


Swift is really powerful – and it’s going to take us a while to establish best practice and work out what idioms and patterns we can now use that weren’t possible within the constraints of objective-C. This post has outlined some of the common areas of confusion when moving from objective-C to Swift, but don’t let this put you off. All of the projects associated with this blog series are written using Swift, and on the most-part are really simple to understand.

There is a playground which contains some of the samples mentioned in this post – it’s part of the Github repo which accompanies this series. You can get it at

If you have any questions or suggestions for additions / updates on this page then please do get in contact. You can leave a comment below, or tweet me – I’m @iwantmyrealname.

The series starts properly on Monday – with a look at one of the significant new APIs within iOS8. Join us then – or use the subscribe button below to get an email reminder that a new post has been published.