There is a key difference not mentioned in the other answers.
To test this, drop the following code in Playground.
1st Attempt:
import Foundation
import PlaygroundSupport
PlaygroundPage.current.needsIndefiniteExecution = true
class Person{
var age = 0
lazy var timer: Timer? = {
let _timer = Timer.scheduledTimer(timeInterval: 1.0, target: self, selector: #selector(fireTimer), userInfo: nil, repeats: true)
return _timer
}()
init(age: Int) {
self.age = age
}
@objc func fireTimer(){
age += 1
print("age: \(age)")
}
deinit {
print("person was deallocated")
}
}
// attempt:
var person : Person? = Person(age: 0)
let _ = person?.timer
person = nil
So let me ask you a question. At the last line of the code, I just set person
to nil
. That means the person
object is deallocated and all its properties are set to nil
and removed from memory. Right?
An object is deallocated as long as no other object is holding a strong a reference to it. In our case the timer
is still holding a strong reference to person, because the run-loop has a strong reference to the timer§ hence the person
object will not get deallocated.
The result of the above code is that it still continues to execute!
Let's fix it.
2nd Attempt:
Let's set the timer to nil
. This should remove the strong reference of timer
pointing to person
.
var person : Person? = Person(age: 0)
let _ = person?.timer
person?.timer = nil
person = nil
WRONG! We only removed our pointer to the timer
. Yet the result of the above code is just like our initial attempt. It still continues to execute...because the run loop is still targeting/referencing self
.
So what do we need to do?
Glad you asked. We must invalidate
the timer!
3rd Attempt:
var person : Person? = Person(age: 0)
let _ = person?.timer
person?.timer = nil
person?.timer?.invalidate()
person = nil
This looks better, but it's still wrong. Can you guess why?
I'll give you a hint. See code below 👇.
4th Attempt (correct)
var person : Person? = Person(age: 0)
let _ = person?.timer
person?.timer?.invalidate()
person?.timer = nil
person = nil
// person was deallocated
Our 4th attempt was just like our 3rd attempt, just that the sequence of code was different.
person?.timer?.invalidate()
removes the run loop's strong reference
to its target, i.e. self
, and now if a pointer to person
is removed...our person object gets deallocated!
The attempt below is also correct:
5th Attempt (correct)
var person : Person? = Person(age: 0)
let _ = person?.timer
person?.timer?.invalidate()
person = nil
// person was deallocated
Notice that in our 5th attempt we didn't set the timer to nil
. But Apple recommends that we do such:
Once invalidated, timer objects cannot be reused.
See Task Management - Timer
Setting it to nil
is also an indicator that for other parts of code. It helps up so that we can check against it and if it wasn't nil
then we'd know the timer is still valid and also to not have a meaningless object around.
After invalidating the timer you should assign nil
to the variable
otherwise the variable is left pointing to a useless timer. Memory
management and ARC have nothing to do with why you should set it to
nil
. After invalidating the timer, self.timer
is now referencing a
useless timer. No further attempts should be made to use that value. Setting it to nil
ensures that any further attempts to access
self.timer will result in nil
from rmaddy's comment above
That being said I think isValid
is a more meaningful approach just as isEmpty
is more meaningful and efficient than doing array.count == 0
...
So why is 3rd attempt not correct?
Because we need a pointer to the timer so we can invalidate it. If we set that pointer to nil
then we loose our pointer to it. We lose it while the run-loop has still maintained its pointer to it! So if we ever wanted to turn off the timer we should invalidate
it BEFORE we lose our reference to it (ie before we set its pointer to nil
) otherwise it becomes an abandoned memory (not leak).
Conclusion:
- To get stop a timer correctly you must use
invalidate
. Do not nil
the timer
before you invalidate
it.
- After you've invalidated a
timer
, set it to nil
so it doesn't get reused.
- Calling
invalidate
will remove the run loop's pointer to self
. Only then the object containing the timer will be released.
So how does this apply when I'm actually building an application?
If your viewController has person
property and then your popped this viewController off your navigation stack then your viewController will get deallocated. In its deinit
method you must invalidate the person's timer. Otherwise your person instance is kept in memory because of the run loop and its timer action will still want to execute! This can lead to a crash!
Correction:
Thanks to Rob's answer
If you're dealing with repeating [NS]Timers, don't try to invalidate them in dealloc of the owner of the [NS]Timer because the dealloc obviously will not be called until the strong reference cycle is resolved. In the case of a UIViewController, for example, you might do it in viewDidDisappear
That being said viewDidDisappear
may not always be the correct place since viewDidDisappear
also gets called if you just push a new viewController on top of it. You should basically do it from a point that it's no longer needed. You get the idea...
§: Because the run loop maintains the timer, from the perspective of
object lifetimes there’s typically no need to keep a reference to a
timer after you’ve scheduled it. (Because the timer is passed as an
argument when you specify its method as a selector, you can invalidate
a repeating timer when appropriate within that method.) In many
situations, however, you also want the option of invalidating the
timer—perhaps even before it starts. In this case, you do need to
keep a reference to the timer, so that you can stop it whenever
appropriate.
With all the credit going to my colleague Brandon:
Pro Tip:
Even if you don't have a repeating timer, the Runloop [as mentioned within the docs] will hold a strong reference to your target if you use the selector function, until it fires, after that it will release it.
However if you use the block based function then as long as you point weakly to self inside your block then the runloop will not retain self
. However it will continue to execute, due to the lack of calling invalidate
If you don't use [weak self]
then the block based will act just like the selector kind, that it will deallocate self
after it has been fired.
Paste the following code in Playground and see the difference. The selector version will be deallocated after it fires. The block base will be deallocated upon deallocation. Basically the lifecycle of one is governed by the runloop while for the other it's governed by the object itself
@objc class MyClass: NSObject {
var timer: Timer?
func startSelectorTimer() {
timer = Timer.scheduledTimer(timeInterval: 3, target: self, selector: #selector(MyClass.doThing), userInfo: nil, repeats: false)
}
func startBlockTimer() {
timer = Timer.scheduledTimer(withTimeInterval: 3, repeats: false, block: { [weak self] _ in
self?.doThing()
})
}
@objc func doThing() {
print("Ran timer")
}
deinit {
print("My Class deinited")
}
}
var mySelectorClass: MyClass? = MyClass()
mySelectorClass?.startSelectorTimer()
mySelectorClass = nil // Notice that MyClass.deinit is not called until after Ran Timer happens
print("Should have deinited Selector timer here")
RunLoop.current.run(until: Date().addingTimeInterval(7))
print("---- NEW TEST ----")
var myBlockClass: MyClass? = MyClass()
myBlockClass?.startBlockTimer()
myBlockClass = nil // Notice that MyClass.deinit IS called before the timer finishes. No need for invalidation
print("Should have deinited Block timer here")
RunLoop.current.run(until: Date().addingTimeInterval(7))