There seems to be a (false) memory leak bug in Xcode 8 when using with iOS 10 & Swift 3.
The following code reports a memory leak in Instruments and the Xcode 8 memory debugger:
class SomeClass: NSObject {
var view: SomeView!
deinit {
print("SomeClass deinit")
}
}
class SomeView: UIView {
weak var reference: SomeClass?
deinit {
print("SomeView deinit")
}
}
class ViewController: UIViewController {
var someProperty: SomeClass?
override func viewDidLoad() {
super.viewDidLoad()
let c = SomeClass()
let v = SomeView()
c.view = v
v.reference = c
someProperty = c
}
}
I've tried different variations to confirm it's indeed a bug and my findings are:
- When you don't assign
c
in the sample code to someProperty
, both instances will print the string in their respective deinit
s. A true strong reference cycle won't deinit.
- When
SomeClass
doesn't inherit from NSObject
, this bug doesn't happen.
- When using Swift 2.2, this doesn't happen.
- When using iOS 9-, this doesn't happen.
- Once
someProperty
is set to nil
somewhere in the code, both instances are deinit
ed. Xcode 8 memory debugger confirms there are no memory leaks. However in Instruments, this change isn't reflected--rightfully so, since a real memory leak probably won't be resolved.
FYI, this doesn't happen only when it's assigned to a property of UIViewController
. I originally found out about this behavior in a singleton object.
Seems fixed with iOS 10.3 (first beta) and Xcode 8.3 (first beta).