MvRxStateStore based on kotlin coroutines about mavericks HOT 10 CLOSED

@ATizik Do you have a MvRxStateStore implementation with coroutines? The main reason we used rx is our own familiarity with rx but if coroutines perform better, it would be worth investigating for sure.

cc @BenSchwab @elihart @hellohuanlin

from mavericks.

@gpeal
I do have something similar in our projects, since you're interested I'll adapt it to MvRx api and share with you

from mavericks.

@ATizik Would love to see a PR. If you have a good way to benchmark it, that would be great too!

from mavericks.

@gpeal
Here it is #84
Speed difference was negligible on existing test, maybe I need a heavier state and operations, i'm in the process of writing more comprehensive benchmark. I'll also test a memory consumption later.

from mavericks.

I still do not understand Why can't simply use the Main Thread to avoid race conditions of setting the state? The stateReducer is just updating the state, why can't put that in the main thread. And if the state changes very frequently, I would prefer to use the LiveData to avoid creating new states.

from mavericks.

@luozejiaqun
It's better to offload it to the background, there is nothing specific about main thread that's better for avoiding race conditions than any other thread.
I'm not sure how LiveData will help avoid creating new states, we still have to run stateReducer the same amount of times

from mavericks.

First question, if there is nothing specific about main thread, why don't use the main thread. Since every ViewModel owns a StateStore, using main thread will avoid creating other threads. I'm not familiar with kotlin coroutines, but the fundamental thing is still threads, less threads achieve the same goal, this my rough understanding. If the stateReducer is simple, there is no reason why can't put it in main thread.
Second question, there is not contradiction between State and LiveData, why can't put a property in LiveData and remove it from the State. In this way, the property can be mutable, changes it and sets it to LiveData's value, then done. The fact is there are some properties we can't use in State, such as SparseArray and LongSparseArray. I put these properties in LiveData and that works well for me.

from mavericks.

@luozejiaqun A fundamental principle of MvRx is that you observe a stream of immutable state. Storing a SparseArray in LiveData actually enables you to write buggy code. If you modify an element in the array, your LiveData won't update. That's why MvRx enforces immutability.

Regarding the background thread, there are performance advantages in keeping it off the main thread but we have implemented it in such a way that there should be no cognitive overhead for the developer. Why not offload things off the main thread if we can?

from mavericks.

@gpeal I know MvRx enforces immutability, because of the reason, I can't use SparseArray in the State. The MvRx's suggestion is using immutable list, but I don't think that makes sense. Because of performance advantages we use SparseArray, immutable list can't meet that requirements. LiveData maybe buggy, but that's my only choice for now.

Regarding the background thread, I know the performance advantages.But using the main thread and no more new threads also has performance advantages.

from mavericks.

@luozejiaqun Let's separate that discussion from the coroutines state store

from mavericks.