Fix for overlapping windows.

[alexwen]

Source was emitting t multiple times while holding queue.

Fixes #2896

[davidmoten]

Could you add the unit test from your gist as well?

[davidmoten]

Ah, someone else's gist. Still, as the unit test is there may as well chuck it in.

[alexwen]

Thanks for the feedback. Is it kosher to do a sleep in the test? Its the
only way I've been able to get the windows to overlap and reproduce the
condition.

Also, it doesn't seem too reliable to depend upon sleep.

[akarnokd]

Good catch! An unit test would be great, but you don't really need concurrency: reentrancy will do just fine:

    @Test
    public void testWindowNoDuplication() {
        final PublishSubject<Integer> source = PublishSubject.create();
        final TestSubscriber<Integer> tsw = new TestSubscriber<Integer>() {
            boolean once;
            @Override
            public void onNext(Integer t) {
                if (!once) {
                    once = true;
                    source.onNext(2);
                }
                super.onNext(t);
            }
        };
        TestSubscriber<Observable<Integer>> ts = new TestSubscriber<Observable<Integer>>() {
            @Override
            public void onNext(Observable<Integer> t) {
                t.subscribe(tsw);
                super.onNext(t);
            }
        };
        source.window(new Func0<Observable<Object>>() {
            @Override
            public Observable<Object> call() {
                return Observable.never();
            }
        }).subscribe(ts);

        source.onNext(1);
        source.onCompleted();

        assertEquals(1, ts.getOnNextEvents().size());
        assertEquals(Arrays.asList(1, 2), tsw.getOnNextEvents());
    }

[alexwen]

Thanks @akarnokd - great test!

[akarnokd]

Thanks!

[benjchristensen]

Should I push a patch release with this today?

[akarnokd]

A verification on #2878 would be great. After that, the patch release can commence.