RxJava2 `Observable.zip` taking `Iterable` and a `zipper` is broken

[dynaxis]

http://reactivex.io/RxJava/2.x/javadoc/io/reactivex/Observable.html#zip-java.lang.Iterable-io.reactivex.functions.Function-

The zipper signature is Function<? super T[]>, ? extends R> zipper, where use of T[] breaks it. Just invoke it with non Object T, and a ClassCastException is thrown.

I checked with Single and Flowable, they have different signatures for the comparable zip operators. They use Object[] where T[] is used for Observable, which are also consistent with the 1.x zip operator.

I think this is just a bug in the signature.

[akarnokd]

Totally possible. Can you post a small unit test that fails?

[dynaxis]

    @Test
    public void zipIterableOfObservables() {
        List<Observable<Integer>> observables =
                new ArrayList<Observable<Integer>>();
        observables.add(Observable.just(1, 2, 3));
        observables.add(Observable.just(1, 2, 3));

        Observable.zip(observables, new Function<Integer[], Integer>() {
            @Override
            public Integer apply(Integer[] o) throws Exception {
                int sum = 0;
                for (int i : o) {
                    sum += i;
                }
                return sum;
            }
        }).test().assertResult(2, 4, 6);
    }

Please note that if we use lambda as a zipper, it eventually generates what the above test has.

[dynaxis]

FYI, the following works and it's what is generated with the 1.x signature:

    @Test
    public void zipIterableOfObservables() {
        List<Observable<Integer>> observables =
                new ArrayList<Observable<Integer>>();
        observables.add(Observable.just(1, 2, 3));
        observables.add(Observable.just(1, 2, 3));

        Observable.zip(observables, new Function<Object[], Object>() {
            @Override
            public Object apply(Object[] o) throws Exception {
                int sum = 0;
                for (Object i : o) {
                    sum += (Integer) i;
                }
                return sum;
            }
        }).test().assertResult(2, 4, 6);
    }

[akarnokd]

Thanks. The underlying problem is that we can't do new T[n] and such lambdas cast the bridge Object apply(Object[]) argument to Integer[] which fails since zip and combineLatest use Object[] internally.

Currently I can't write a PR. @vanniktech could you change all ? super T[] signatures back to ? super Object[] and add unit tests like above to verify there is no ClassCastException in a PR?

[vanniktech]

Sure 👍

[akarnokd]

Closing via #4525.

[dynaxis]

BTW, why use array instead of List there? Any reason? If List is used, then at least it seems safe with generics, doesn't it?

[akarnokd]

Using array has reduced allocation cost and less indirection.

[hafs-r]

I've Singles of different types to run with Single.zip, and I'm getting ClassCastException as SingleZipIterable. Is it wrong zipping Singles of different type together?