When to use weak self and why

We all want to write good, beautiful and stable code. This includes preventing memory leaks, which we can, using [weak self] when writing a closure that needs access to self. But what's the real reason for needing this weak capture? And do we need it all the time? In this week's Quick Tip, I want to help you find an answer to this question so you can make more informed decisions about your capture lists in the future.

This post contains the following topics:

• Understanding what a capture list is
• Understanding different kinds of captures
• Knowing when to rely on implicit or strong captures
• Deciding whether you need a weak or an unowned self capture

Enjoy!

Understanding what a capture list is

When you write a closure, it will implicitly capture all properties referenced inside of the closure. Let's look at a short code sample to illustrate this:

func multiply(by multiplier: Int) -> ((Int) -> Int) {
  return { (input: Int) -> Int in
    return input * multiplier
  }
}

var multiplier = 2
let multiplyTwo = multiply(by: multiplier) // you can now call multiplyTwo as if it's a function
multiplier = 4
let multiplyFour = multiply(by: multiplier) // you can now call multiplyFour as if it's a function

The preceding code sample shows a function that takes a multiplier and returns a closure that takes a different input and multiplies it with the multiplier that was passed to multiply(by:). The closure that is returned from this function captures the multiplier and uses it as the multiplier for the input you give it. In practice, this means that calling multiplyTwo(2) will return 4 and multiplyFour(2) returns 8. The multiplier variable that is defined does not affect the closure held by multiplyTwo or multiplyFour because of this capture behavior.

I know that this can be quite confusing when you've just started learning about closures, but bear with me as we go through some more examples.

Take a look at the following example:

var name = "Donny"
var appendToName = { (string: String) -> String in
  return name.appending(string)
}

let one = appendToName("Wals")
name = "D"
let two = appendToName("Wals")

What would you expect the values of one and two to be? Remember that I just explained how closures capture properties that are used inside of a closure.

If you expect one and two to both be "DonnyWals" I don't blame you, it seems to make sense! But unfortunately, this isn't correct. The value for one is "DonnyWals" and two is "DWals". How this closure is different from the closure you saw before is that everything, from the closure to the property it references is in the same context. The closure can read the current value of name because it's on the same level. We can, however, explicitly capture name using a capture list as follows:

var name = "Donny"
var appendToCapturedName = { [name] (string: String) -> String in
  return name.appending(string)
}

let one = appendToCapturedName("Wals")
name = "D"
let two = appendToCapturedName("Wals")

When you run this code in a Playground, both one and two will equal "DonnyWals" because you explicitly told the closure to capture the current value of name by putting it in a capture list. This can be called an explicit or strong capture. You can capture more than one property in a capture list by comma separating them: [property, anotherProperty].

You just learned about strong and implicit capturing of properties. Let's look at other kinds of captures.

Understanding different kinds of captures

When you strongly capture a property, the closure will own this property. For value types like structs and enums, this means that the closure copies the current value of an object over to its own area in memory where it owns the object. When you do the same for a reference type, like a class, the closure will maintain a strong pointer reference to the object. To understand the implications of this, you need to know one thing about reference types: they are never deallocated as long as at least one other object holds a reference to them. When an object holds an unintended reference to a reference type, it could be considered a memory leak. If you're not sure what this means, don't worry, it should be a little bit clearer after the next code sample. The following code demonstrates the memory leak that I described earlier:

class MyClass {}
var instance: MyClass? = MyClass()

var aClosure = { [instance] in
  print(instance)
}

aClosure() // MyClass
instance = nil
aClosure() // MyClass

The second time we call aClosure we still print the same instance of MyClass because aClosure holds a strong reference to the instance. Sometimes this is exactly what we want, but usually, we don't want our closures to keep objects alive after they've been deinitialized. An example that comes to mind is a closure that might capture a view controller while it's waiting for a network request to finish. If the view controller is dismissed before the network request is finished, we want the view controller to be removed from memory, or deallocated. If the network request's completion closure has a strong reference to the view controller, the closure keeps the view controller alive because it still holds a reference to the view controller.

So how would we make this strong capture a not so strong capture? Well, how about we make it weak instead?

Tip: You can run the code below in a Swift Playground to see the result.

class MyClass {}

var myInstance: MyClass? = MyClass()

var aClosure = { [weak myInstance] in
  print(myInstance)
}

aClosure() // MyClass
myInstance = nil
aClosure() // nil

Because the closure only holds a weak reference to the instance of MyClass, the system doesn't count our closure's reference to instance which means that as soon as the playground releases its reference to our instance of MyClass, it can be deallocated. The downside here is that this might lead to subtle bugs where you don't immediately notice that instance was deallocated before your closure was called. If you want to assert that instance is still around when the closure is called, and want to crash your app if it's not you can use an unowned reference:

Tip: You can run the code below in a Swift Playground to see the result.

class MyClass {}
var instance: MyClass? = MyClass()

var aClosure = { [unowned instance] in
  print(instance)
}

aClosure() // MyClass
instance = nil
aClosure() // crash

The impact on memory for an unowned capture is pretty much the same as weak. It's very similar to safely unwrapping an optional value with ? or doing so forcefully with ! which crashes your app if the value to unwrap is nil. In practice, you'll find that unowned is almost never what you're looking for.

Now that you have some understanding of what weak and unowned are, and how you can implicitly or strongly capture a value, let's have a look at when you should use these different capture methods.

Knowing when to use weak, unowned, strong or implicit capture

As mentioned at the start of this post, a lot of developers use [weak self] in their closures to prevent memory leaks. But are memory leaks really that common when working with closures? That depends on what they're used for exactly. In this section, we'll explore different kinds of captures in closures and what their implications are. We'll first look at when you might want to use an implicit capture. Then we'll look at strong capture and last we'll look at weak and unowned captures.

When to use implicit capture

Implicit capture is often used when you're dealing with closures that capture self, where self is a value type, like a struct. Since structs don't have pointers that reference them, closures won't accidentally keep a struct alive for longer than it should. In fact, trying to use weak or unowned on a non-class type isn't allowed in Swift.

You can use implicit capture on reference types as long as you're certain that the closure you're calling won't be retained by the object that will receive your closure. A good example of this is performing work on a DispatchQueue:

class MyClass {
  func dispatchSomething() {
    DispatchQueue.global().async {
      // it's okay to implicitly capture self here
    }
  }
}

Since the closure passed to async isn't retained, you can safely capture it without a capture list. You know it's not retained because the closure is executed shortly after calling async, and it's only called once. If your closure is retained, for example when it's used as an event handler for an object, you should make sure to capture self weakly to avoid keeping a reference to self that you don't want. Keep in mind that you're often relying on implementation details when you're doing this so even though you don't have to use weak self here, it might be a good idea to limit your usage of implicit captures to code you own and control.

The easiest way to know whether a closure that you pass to a function is retained, is to check whether it's marked as @escaping. Closures that are not marked with @escaping do not leave the scope of the function that you pass it too which means that it can't be retained for longer than the function's scope. If you want to learn more about @escaping, take a look at this post.

When to use an explicit strong capture

Strongly, or explicitly capturing references isn't done very often. It's most useful if you want to allow partial deallocation of objects. For example, if you perform a network request and want to store its result in a Core Data store without doing anything else, the object that initiated the request doesn't have to stick around; you only need the data store. An example of this might look as follows:

struct SomeDataSource {
  var storage: MyDataStore
  let networking: NetworkingLayer
  // more properties

  func refreshData() {
    networking.refresh { [storage] data in
      storage.persist(data)
    }
  }
}

Since the closure only requires the storage property, there's no need to capture self here since that would keep the entire SomeDataSource object around. Keep in mind though that the storage property is captured at the time the closure is created as I showed you earlier in this post. So in this case that means that the value of storage is captured when we call refreshData. If MyDataStore is a struct, that means that it's copied at capture-time in the closure and any changes that are made to storage after the closure is created are not visible inside of the closure.

If MyDataStore is a reference type, you would be able to see changes made to the instance that's captured in the closure, but if you change storage by assigning a new storage to it after storage is captured, you will have captured the old storage instead of the new one.

When to use weak and unowned captures

Weak capture is by far the most common capture and it's usually a good default choice. It should typically be used when you don't want an object to stick around until the closure is performed, or if the closure is retained for an unknown amount of time which is often the case for network requests. Keep in mind that a closure with a weak capture will treat the captured property as an optional. So if you have a [weak self] in your capture list, you'll likely want to unwrap self in your closure body using guard let self else { return } to make sure that self still exists by the time the closure is executed. Using weak or unowned is the only way to make 100% sure your reference type doesn't hang around in memory for longer than necessary. This makes it a good option if you're not sure which type of capture if most appropriate for your current use case.

When it comes to deciding between weak and unowned my personal opinion is to go with weak. There are little to no performance gains when using unowned (not to the point where it will matter in most apps anyway) and if any of your assumptions become incorrect later on, your unowned capture would crash you app which isn't great.

I always think of unowned as a force unwrap. It's probably fine in some cases, but you're essentially putting a little landmine in your code that in my experience is likely to go off when you least expect it.

In Summary

And that's another Quick Tip that became much larger than I initially intended. Closures and capture lists have all kinds of subtle implications that are important to keep in mind while programming and I hope this post has given you some insights into why you sometimes need a [weak self] for reference types but can reference self freely in value types (remember, it's all about the reference count). You saw that you can even capture specific properties of an object if you don't need to capture the entire self. You can even apply weak or unowned to individual properties of an object to prevent your closures from keeping individual objects around.

Now that you know pretty much everything there is to know about closures and capture lists, go ahead and look through your codebase to see if there are any improvements you can make based on the information you just read. As always, thanks for reading and don't hesitate to reach out on Twitter if you have any feedback or questions.

Special thanks to Bas Broek for proofreading this article!