Swiftpack.co - PittsCraft/PublisherFuse as Swift Package

Swiftpack.co is a collection of thousands of indexed Swift packages. Search packages.
See all packages published by PittsCraft.
PittsCraft/PublisherFuse 0.0.1
⚡️ Safety Combine publisher fuse allowing to cancel upstream subscriptions especially when a chosen object is deallocated.
⭐️ 0
🕓 3 weeks ago
iOS macOS
.package(url: "https://github.com/PittsCraft/PublisherFuse.git", from: "0.0.1")

PublisherFuse ⚡️

Safety Combine publisher fuse allowing to cancel upstream subscriptions especially when a chosen object is deallocated.

  • iOS 13+
  • macOS 10.15+

Problem

class Forwarder {

    private let repository: Repository // Provides a publisher of A
    
    private func map<A, B>(_ input: A) -> B? {
        // Some stateful mapping
    }
    
    func publisher() -> AnyPublisher<B?, Never> {
        repository
            .aPublisher()
            .map { [unowned self] in
                map($0)
            }
            .eraseToAnyPublisher()
    }

}

If the Forwarder has a lifetime shorter than the Repository, any subscription to the publisher will live independently of Forwarder's deallocation, leading to a crash caused by the use of unowned capture.

Usual solutions

I strongly encourage you to have a close look to find out why you're in such a situation and how you could avoid these issues, by paying particular attention to lifecycles, sanitizing your architecture and code.

For example, if you have got a AnyPublisher<Forwarder?, Never> that represents your Forwarder's lifecycle, don't use compactMap to get its non nil values. Instead, properly handling the nil case to maintain your subscriptions can be enough.

If refactoring is not possible, then this problem could be solved easily by using weak or strong capture depending on what is wanted.

However there are drawbacks to these solutions, for example:

  • using weak may be uncomfortable: guard let self = self else { return nil } is usual but the output of the mapping can be indistinguishable from a potential meaningful nil value. Moreover, the receptions won't stop until the subscription is canceled, independently of Forwarder's deallocation.
  • strong causes the subscription to retain the Forwarder and that may be unwanted.

Safety first

Depending on the context, you may still want to use a safety mechanism to make sure that any future development won't make weird behaviors occur.

This is what this package is about.

class Forwarder {

    private var cancellables = Set<AnyCancellable>()
    private let repository: Repository // Provides a publisher of A
    
    private func map<A, B>(_ input: A) -> B? {
        // Some mapping
    }
    
    func publisher() -> AnyPublisher<B?, Never> {
        repository
            .aPublisher()
            .map { [unowned self] in
                map($0)
            }
            .fuse(heldBy: self, on: \.cancellables)
            .eraseToAnyPublisher()
    }
}

Short version

This fuse stops all activity as soon as its holder is deallocated, making it safe for consumers of the publishers to stay subscribed longer than Forwarder's lifetime.

However subscriptions made to a publisher provided by the publisher() function should be canceled at some point if you want to release what's retained by them and avoid memory leaks.

Longer version

What's going on here:

  • any call to publisher() will store an associated proxy stored in the cancellables set
  • when the Forwarder is deallocated the proxies are canceled, invalidating all the subscription chains above and including the fuse(heldBy:on:) call (ie anything related to repository.aPublisher().map(...).fuse(...))
  • of course same thing happens if you explicitly cancel the proxies
  • if there are no more active subscriptions on it and nothing retains a publisher, its proxy is properly removed from the set

Dev

This was more of an exercise to me, that's why:

  • tests don't cover straightforward parts of the code
  • it's probable there won't be any further development or maintaining effort except if there's any real world adoption

GitHub

link
Stars: 0
Last commit: 3 weeks ago
jonrohan Something's broken? Yell at me @ptrpavlik. Praise and feedback (and money) is also welcome.

Release Notes

0.0.1
3 weeks ago

Initial release

Swiftpack is being maintained by Petr Pavlik | @ptrpavlik | @swiftpackco | API | Analytics