Is it assumed that CompositeException is eligible to modify the cause of exceptions passed in constructor of CompositeException?

[zubchenok]

Since RxJava modifies exceptions during call of getCause method of CompositeException instance and connect these exceptions in chains we have finally millions of exceptions and quickly getting out of memory.

    @Test
    public void compositeExceptionIssue() {
        Single
            .just(new Throwable("ROOT ERROR"))
            .flatMapCompletable(rootError -> Observable
                .range(1, 10)
                .flatMapCompletable(testNumber -> Completable
                    .mergeArrayDelayError(
                        Completable.error(new RuntimeException("Test#" + testNumber + "A", rootError)),
                        Completable.error(new RuntimeException("Test#" + testNumber + "B", rootError))
                    )
                    .doOnError(Throwable::getCause)
                    .onErrorComplete()
                )
                .doOnComplete(() -> {
                    rootError.printStackTrace();
                })
            )
            .blockingAwait();
    }

This simple test demonstrates that cause of rootError throwable is changed.

[akarnokd]

Do you mean those "Caused by: java.lang.RuntimeException: Duplicate found in causal chain so cropping to prevent loop" listings?

[akarnokd]

The getCause looks needlessly convoluted. I presume it was implemented that way that IDEs show some usable stacktraces of the individual inner exceptions. Java 7's addSuppressed is better in this regard but we can't use it just yet.

I can probably have such duplicate indicator show up at most once.

[zubchenok]

I mean that original exception is actually chained with other exceptions. rootError.getCause()!=null while expected to be null (in my option).

[akarnokd]

I see. To resolve it, we'd have to remove the custom getCause implementation otherwise it will keep touching all sorts of inner exceptions. We could also probably fake a StackTraceElement array containing all inner exceptions too.

[akarnokd]

Thanks for reporting. This is a shortcoming of the 1.x and 2.x composite design and I'm afraid we can only resolve this for 3.x because it is a breaking change in the structure of the composite verified by unit tests.