PublishSubject onCompleted from different disposeBag

[grinder81]

Short description of the issue:

Wrong disposeBag triggering onComplete for another publish subject.

Explanation:

So far this is the weirdest problem I'm dealing with my project using RxSwift. Here are the details:

Publish Subject holder class:

class MyPublishHolder: NSObject{
	
	static let shared = MyPublishHolder()

	let actionPublishSubject = PublishSubject<SomeEnum>()
}

Class subscribing publish subject:

class MyPublishSubscriber: NSObject {
	
	static let shared = MyPublishSubscriber()	

	// This is an Rx Initializer only get call once
	func setupRx() {

		MyPublishHolder.shared.actionPublishSubject
			.subscribeNext {( enum ) in 
				print (enum)
			})
			, onError: { (error) in
                           print(error)
                       }, onCompleted: {
                           print("onCompleted !!!!")
                     }) {
                          print("Why !!!!")
                      }
			.addDisposableTo(rx.disposeBag)
	}
}

From my other class, I'm just doing something like

MyPublishHolder.shared.actionPublishSubject
	.onNext(SomeEnum.enum)

That's it!

Problem:
Everything works fine as expected. I publish an enum and MyPublishSubscriber do some work based on that enum. But in a very certain case which is reproducible every time, onCompleted get called though I never did have onComplete call for this publish subject. Then when I started digging memory footprint I found autorelease pool clearing memory which calls deinit of a disposeBag and that's lead to onCompleted and then onDispose.

Please note that I tracked memory location of MyPublishSubscriber rx.disposeBag and then the one is getting free and this one completely different memory location or address.

Right now I don't see a way to dig more.

RxSwift/RxCocoa/RxBlocking/RxTest version/commit

3.x

Platform/Environment

• iOS

How easy is to reproduce? (chances of successful reproduce after running the self contained code)

• easy, 100% repro

Xcode version:

Xcode 9 

Installation method:

• CocoaPods
  I have multiple versions of Xcode installed:
  (so we can know if this is a potential cause of your issue)
  7,8 and 9

[kzaher]

Hi @grinder81 ,

can you please provide me with some code that reproduces this issue?

It is not clear what is .addDisposableTo(rx.disposeBag).

[grinder81]

.addDisposableTo(rx.disposeBag) from another pod but please ignore that. I also replaced it by fileprivate var disposeBag = DisposeBag() in my MyPublishSubscriber class and then .addDisposableTo(rx.disposeBag) become .addDisposableTo(self.disposeBag).

[kzaher]

Can you repro that in a standalone example project? If so, do you mind sending me that project?

[grinder81]

I tried to create a simple project to reproduce it but that doesn't happen :( I wish I could do that.

Is there any particular area I can focus to trace it down? I have other publish subject but it doesn't happen with other cases. I tried to force it to subscribe on the main scheduler to see is that event anomaly or not but didn't help.

[kzaher]

I would assume that there is a faulty rx extension somewhere.

Usually what happens is that somebody is using objc_getAssociatedObject and objc_setAssociatedObject inside rx extension in this form objc_getAssociatedObject(self, ... vs the correct objc_getAssociatedObject(base, ....

Because Swift has struct reusing, those two facts combined usually results in that kind of behavior.

I would also suggest tracing the creation of all of the subjects, printing their addresses and similar. Sorry, can't help you more without repro.

Good luck!

[grinder81]

I think you gave me the biggest hint. I have following pod

https://github.com/RxSwiftCommunity/NSObject-Rx

which is using objc_ creating mechanism. Here it is:

https://github.com/RxSwiftCommunity/NSObject-Rx/blob/master/HasDisposeBag.swift

I'm going to dig. I really appreciate your feedback and I can't expect more than that as I can't give a sample project to you.

Thanks a lot and will let you know how it ends up.

If you see anything in that pod which is bad, please let it know here.

[grinder81]

Just to give an update. I did quite an investigation.

• I removed that pod completely and made sure there is no rx.disposeBag in my code
• Replaced rx.disposeBag by self.disposeBag

But still, I'm seeing that issue. Which means I can't blame this pod for memory leaking. I'll keep digging.