Exception rethrown inconsistently with/without .flatMap() in 0.16.1

[jayessdeesea]

Maybe I just don't understand JavaRX yet but this seems inconsistent.

Run this, and then re-run without the .flatMap(). Note that the second run passes the unit test.

@Test(expected = RuntimeException.class)
public void verifyExceptionIsThrownIfThereIsNoExceptionHandler() {

    Observable.OnSubscribeFunc < Object > creator = new Observable.OnSubscribeFunc < Object >() {

        @Override
        public Subscription onSubscribe(Observer < ? super Object > observer) {

            observer.onNext("a");
            observer.onNext("b");
            observer.onNext("c");
            observer.onCompleted();

            return Subscriptions.empty();
        }
    };

    Func1 < Object, Observable < Object >> manyMapper = new Func1 < Object, Observable < Object >>() {

        @Override
        public Observable < Object > call(Object object) {

            return Observable.from(object);
        }
    };

    Func1 < Object, Object > mapper = new Func1 < Object, Object >() {

        private int count = 0;

        @Override
        public Object call(Object object) {

            ++count;

            if (count > 2) {
                throw new RuntimeException();
            }
            return object;
        }
    };

    Action1 < Object > onNext = new Action1 < Object >() {

        @Override
        public void call(Object object) {
            System.out.println(object.toString());
        }
    };
    Observable.create(creator).flatMap(manyMapper).map(mapper).subscribe(onNext);
}

[zsxwing]

The problem here is that one of SafeObserver.onError is called twice: one is RuntimeException, the other is OnErrorNotImplementedException. A little to similar to #748

[zsxwing]

Not sure this is a bug. If an Observer throws an Exception in onError, what will happen? I think Rx does not guarantee that it always calls onError when an exception is thrown. It only guarantees that it at least calls onError once even if there are many exceptions. Do I misunderstand anything?

[jayessdeesea]

the behavior is inconsistent. that's the bug.

Document that the throwables are silently swallowed or thrown if there is no onError()

Errors should probably always be thrown because I want my JVM to terminate before my JVM starts misbehaving and serving wrong data

[benjchristensen]

If an onError is not implemented it should always be thrown.

[zsxwing]

@benjchristensen , Looks some operators do not handle OnErrorNotImplementedException correctly. Could you propose a guild about how to handle exception in RxJava? Now we have RuntimeException except OnErrorNotImplementedException, Exception, Error and OnErrorNotImplementedException.

[benjchristensen]

Dealing with in pull request #839

[benjchristensen]

Should be fixed in #839