Are Swift variables atomic?
Objective CSwiftObjective C Problem Overview
In Objective-C you have a distinction between atomic and nonatomic properties:
@property (nonatomic, strong) NSObject *nonatomicObject;
@property (atomic, strong) NSObject *atomicObject;
From my understanding you can read and write properties defined as atomic from multiple threads safely, while writing and accessing nonatomic properties or ivars from multiple threads at the same time can result in undefined behavior, including bad access errors.
So if you have a variable like this in Swift:
var object: NSObject
Can I read and write to this variable in parallel safely? (Without considering the actual meaning of doing this).
Objective C Solutions
Solution 1 - Objective C
It's very early to assume as no low-level documentation is available, but you can study from assembly. Hopper Disassembler is a great tool.
@interface ObjectiveCar : NSObject
@property (nonatomic, strong) id engine;
@property (atomic, strong) id driver;
@end
Uses objc_storeStrong and objc_setProperty_atomic for nonatomic and atomic respectively, where
class SwiftCar {
var engine : AnyObject?
init() {
}
}
uses swift_retain from libswift_stdlib_core and, apparently, does not have thread safety built in.
We can speculate that additional keywords (similar to @lazy) might be introduced later on.
Update 07/20/15: according to this blogpost on singletons swift environment can make certain cases thread safe for you, i.e.:
class Car {
static let sharedCar: Car = Car() // will be called inside of dispatch_once
}
private let sharedCar: Car2 = Car2() // same here
class Car2 {
}
Update 05/25/16: Keep an eye out for swift evolution proposal https://github.com/apple/swift-evolution/blob/master/proposals/0030-property-behavior-decls.md - it looks like it is going to be possible to have @atomic behavior implemented by yourself.
Solution 2 - Objective C
Swift has no language constructs around thread safety. It is assumed that you will be using the provided libraries to do your own thread safety management. There are a large number of options you have in implementing thread safety including pthread mutexes, NSLock, and dispatch_sync as a mutex mechanism. See Mike Ash's recent post on the subject: https://mikeash.com/pyblog/friday-qa-2015-02-06-locks-thread-safety-and-swift.html So the direct answer to your question of "Can I read and write to this variable in parallel safely?" is No.
Solution 3 - Objective C
It is probably to early to answer this question. Currently swift lacks access modifiers, so there is not obvious way to add code which manages concurrency around a properties getter / setter. Furthermore, the Swift Language doesn't seem to have any information about concurrency yet! (It also lacks KVO etc ...)
I think the answer to this question will become clear in future releases.
Solution 4 - Objective C
Details
- Xcode 9.1, Swift 4
- Xcode 10.2.1 (10E1001), Swift 5
Links
- apple.developer.com Dispatch
- Grand Central Dispatch (GCD) and Dispatch Queues in Swift 3
- Creating Thread-Safe Arrays in Swift
- Mutexes and closure capture in Swift
Implemented types
- QueueSafeValue - a universal wrapper that provides safe multi-threaded access to values
- AtomicArray
- AtomicInteger
- AtomicValue
Main Idea
class Example {
private lazy var semaphore = DispatchSemaphore(value: 1)
func executeThreadSafeFunc1() {
// Lock access. Only first thread can execute code below.
// Other threads will wait until semaphore.signal() will execute
semaphore.wait()
// your code
semaphore.signal() // Unlock access
}
func executeThreadSafeFunc2() {
// Lock access. Only first thread can execute code below.
// Other threads will wait until semaphore.signal() will execute
semaphore.wait()
DispatchQueue.global(qos: .background).async {
// your code
self.semaphore.signal() // Unlock access
}
}
}
Sample of atomic access
class Atomic {
let dispatchGroup = DispatchGroup()
private var variable = 0
// Usage of semaphores
func semaphoreSample() {
// value: 1 - number of threads that have simultaneous access to the variable
let atomicSemaphore = DispatchSemaphore(value: 1)
variable = 0
runInSeveralQueues { dispatchQueue in
// Only (value) queqes can run operations betwen atomicSemaphore.wait() and atomicSemaphore.signal()
// Others queues await their turn
atomicSemaphore.wait() // Lock access until atomicSemaphore.signal()
self.variable += 1
print("\(dispatchQueue), value: \(self.variable)")
atomicSemaphore.signal() // Unlock access
}
notifyWhenDone {
atomicSemaphore.wait() // Lock access until atomicSemaphore.signal()
print("variable = \(self.variable)")
atomicSemaphore.signal() // Unlock access
}
}
// Usage of sync of DispatchQueue
func dispatchQueueSync() {
let atomicQueue = DispatchQueue(label: "dispatchQueueSync")
variable = 0
runInSeveralQueues { dispatchQueue in
// Only queqe can run this closure (atomicQueue.sync {...})
// Others queues await their turn
atomicQueue.sync {
self.variable += 1
print("\(dispatchQueue), value: \(self.variable)")
}
}
notifyWhenDone {
atomicQueue.sync {
print("variable = \(self.variable)")
}
}
}
// Usage of objc_sync_enter/objc_sync_exit
func objcSync() {
variable = 0
runInSeveralQueues { dispatchQueue in
// Only one queqe can run operations betwen objc_sync_enter(self) and objc_sync_exit(self)
// Others queues await their turn
objc_sync_enter(self) // Lock access until objc_sync_exit(self).
self.variable += 1
print("\(dispatchQueue), value: \(self.variable)")
objc_sync_exit(self) // Unlock access
}
notifyWhenDone {
objc_sync_enter(self) // Lock access until objc_sync_exit(self)
print("variable = \(self.variable)")
objc_sync_exit(self) // Unlock access
}
}
}
// Helpers
extension Atomic {
fileprivate func notifyWhenDone(closure: @escaping ()->()) {
dispatchGroup.notify(queue: .global(qos: .utility)) {
closure()
print("All work done")
}
}
fileprivate func runInSeveralQueues(closure: @escaping (DispatchQueue)->()) {
async(dispatch: .main, closure: closure)
async(dispatch: .global(qos: .userInitiated), closure: closure)
async(dispatch: .global(qos: .utility), closure: closure)
async(dispatch: .global(qos: .default), closure: closure)
async(dispatch: .global(qos: .userInteractive), closure: closure)
}
private func async(dispatch: DispatchQueue, closure: @escaping (DispatchQueue)->()) {
for _ in 0 ..< 100 {
dispatchGroup.enter()
dispatch.async {
let usec = Int(arc4random()) % 100_000
usleep(useconds_t(usec))
closure(dispatch)
self.dispatchGroup.leave()
}
}
}
}
Usage
Atomic().semaphoreSample()
//Atomic().dispatchQueueSync()
//Atomic().objcSync()
Result
Solution 5 - Objective C
From Swift 5.1 you can use property wrappers to make specific logic for your properties. This is atomic wrapper implementation:
@propertyWrapper
struct atomic<T> {
private var value: T
private let lock = NSLock()
init(wrappedValue value: T) {
self.value = value
}
var wrappedValue: T {
get { getValue() }
set { setValue(newValue: newValue) }
}
func getValue() -> T {
lock.lock()
defer { lock.unlock() }
return value
}
mutating func setValue(newValue: T) {
lock.lock()
defer { lock.unlock() }
value = newValue
}
}
How to use:
class Shared {
@atomic var value: Int
...
}
Solution 6 - Objective C
Here is the atomic property wrapper that I use extensively. I made the actual locking mechanism a protocol, so I could experiement with different mechanisms. I tried semaphores, DispatchQueues, and the pthread_rwlock_t. The pthread_rwlock_t was chosen because it appears to have the lowest overhead, and a lower chance of a priority inversion.
/// Defines a basic signature that all locks will conform to. Provides the basis for atomic access to stuff.
protocol Lock {
init()
/// Lock a resource for writing. So only one thing can write, and nothing else can read or write.
func writeLock()
/// Lock a resource for reading. Other things can also lock for reading at the same time, but nothing else can write at that time.
func readLock()
/// Unlock a resource
func unlock()
}
final class PThreadRWLock: Lock {
private var rwLock = pthread_rwlock_t()
init() {
guard pthread_rwlock_init(&rwLock, nil) == 0 else {
preconditionFailure("Unable to initialize the lock")
}
}
deinit {
pthread_rwlock_destroy(&rwLock)
}
func writeLock() {
pthread_rwlock_wrlock(&rwLock)
}
func readLock() {
pthread_rwlock_rdlock(&rwLock)
}
func unlock() {
pthread_rwlock_unlock(&rwLock)
}
}
/// A property wrapper that ensures atomic access to a value. IE only one thing can write at a time.
/// Multiple things can potentially read at the same time, just not during a write.
/// By using `pthread` to do the locking, this safer then using a `DispatchQueue/barrier` as there isn't a chance
/// of priority inversion.
@propertyWrapper
public final class Atomic<Value> {
private var value: Value
private let lock: Lock = PThreadRWLock()
public init(wrappedValue value: Value) {
self.value = value
}
public var wrappedValue: Value {
get {
self.lock.readLock()
defer { self.lock.unlock() }
return self.value
}
set {
self.lock.writeLock()
self.value = newValue
self.lock.unlock()
}
}
/// Provides a closure that will be called synchronously. This closure will be passed in the current value
/// and it is free to modify it. Any modifications will be saved back to the original value.
/// No other reads/writes will be allowed between when the closure is called and it returns.
public func mutate(_ closure: (inout Value) -> Void) {
self.lock.writeLock()
closure(&value)
self.lock.unlock()
}
}