NPE in rx.subjects.AbstractSubject.emitNotification()

[ronanM]

The problem is describe here :

https://groups.google.com/forum/#!topic/rxjava/x6B5bim4uoE

Callstack :

Exception in thread "AWT-EventQueue-0" java.lang.NullPointerException
at rx.subjects.AbstractSubject.emitNotification(AbstractSubject.java:98)
at rx.subjects.AbstractSubject.emitNotification(AbstractSubject.java:116)
at rx.subjects.PublishSubject.onNext(PublishSubject.java:82)
at rx.operators.OperationMulticast$MulticastConnectableObservable$2.onNext(OperationMulticast.java:64)
at rx.operators.SynchronizedObserver.onNext(SynchronizedObserver.java:86)
at rx.operators.OperationMerge$MergeObservable$ChildObserver.onNext(OperationMerge.java:267)
at rm.rx.test.gui.GuiRx$4$1.mouseClicked(GuiRx.java:180)
at java.awt.AWTEventMulticaster.mouseClicked(AWTEventMulticaster.java:270)

[zsxwing]

I checked the AbstractSubject codes and found this bug. The following codes do not check if Observer<? super T> o is null:

    protected static <T> void emitNotification(final SubjectState<T> state, final Action2<SubjectState<T>, Observer<? super T>> onEach) {
        for (Subscription s : snapshotOfObservers(state)) {
            Observer<? super T> o = state.observers.get(s);
            // emit notifications to this observer
            emitNotification(state.currentValue.get(), o);
            // onEach action if applicable
            if (onEach != null) {
                onEach.call(state, o);
            }
        }
    }

I think #605 will fix this issue. @akarnokd, can you confirm it? Thanks.

[akarnokd]

PR #605 doesn't exhibit this bug as I make copy of the observers.values() collection when dispatching values.

[benjchristensen]

Subjects were completely re-written and I don't believe this can happen.

I looked at the sample code on Google Groups but it was far too complicated to quickly pull in and run and had many dependencies.

If you want to provide a simpler test case I can assert that this is truly fixed, otherwise I believe 0.16.0 will fix it as Subjects have a rewritten subscription manager that uses a state machine to correctly keep control of the Observers.

[mttkay]

We just encountered this problem in our live app. This is with version 0.15.1.

Was anyone able to confirm that this is fixed with 0.16.x?

[akarnokd]

This shouldn't happen anymore. Could you check with the latest version and close?

[mttkay]

I can confirm this is fixed for us.