David Karnok

`grouped.flatMap(..., 1)`. Limiting backpressure such a way will result in a "hot path" from category 0 to processing its items thus you don't get mixing. Please try out the example...

The way you generate the input to `groupBy`, there is no guarantee you get reliable dispatching patterns. Worked for me; apparently doesn't last for you. Small imbalance in threading and...

@KarboniteKream Thanks for the feedback. I'm still thinking about the implementation details and support considerations. Stay tuned. @Mahammadnajaf Suggesting *blocking* operations is 99.99% wrong.

> consider upgrading to a newer version of RxJava3 that has addressed this issue. 🤣🤣🤣🤣

What's with these out-of-blue out-of-context comments of yours?

I found it very funny since I'm responsible for making that "newer version of RxJava3" happen.

I don't fully understand the issue. We have an operator parameter in RxJava to avoid [the retention of the head item](https://github.com/ReactiveX/RxJava/blob/5bb119b1489afe4fbc5bb51b828be95971ee8f1d/src/main/java/io/reactivex/rxjava3/internal/operators/flowable/FlowableReplay.java#L765-L772).

The buffer is fed by a single producer. Each subscriber walks the linked list from a node they captured at the time of subscription. Every new item arriving will indicate...

Is `SinkManyReplayProcessor` supposed to support [concurrent access](https://github.com/MikkelHJuul/reactor-core/blob/issues/3340-single-processor-stresstest/reactor-core/src/jcstress/java/reactor/core/publisher/SinksManyReplayLatestStressTest.java#L28-L36) to its `tryEmitNext` and other such methods?

@dependabot recreate