2.x: Should schedulers have RxThreadFactory constructor parameters?

[ZacSweers]

RxJava 1 allowed for specifying a custom thread factory, but it looks like this feature did not make it to RxJava 2. Was this intentional or something that a PR would be welcome for?

[akarnokd]

Which scheduler? Computation? It was intentional. You can now configure the priority of each scheduler so there was little benefit from a customizing the factory. Use the ParallelScheduler from extensions which allows more customization.

[ZacSweers]

We used it for the IO scheduler as well. It was useful for tracking our own thread naming and setting their priority to android-specific priorities. We'd like to use it for that and SingleScheduler. Can ParallelScheduler be used as a substitute for non-computation schedulers too?

[akarnokd]

Since scheduler implementations are internal, it is not recommended anyway to use them directly.

ParallelScheduler is an individual scheduler with a specified number of threads that doesn't change over time. So unlike io(), ParallelScheduler won't stop its threads when they become idle for too long.

[ZacSweers]

Since scheduler implementations are internal, it is not recommended anyway to use them directly.

I'm not sure I follow. As in we shouldn't create a new IoScheduler instance?

[ZacSweers]

Prior discussions/work for reference - #3724 #3879

[akarnokd]

Anything in the io.reactivex.internal.* is considered private and not part of any binary or API compatibility contracts.

[akarnokd]

So either use Schedulers.from(Executor) with your own pool or copy the internal code and change it locally.

[ZacSweers]

I missed that they were internal now. That's disappointing to see, and seems like a bit of a step backward compared to RxJava 1. We'll just copy then :|

[JakeWharton]

They're internal in 1.x too. The only remnants are public for binary compatibility that no one should be using anyway. Are you referring to the factory methods? Those could easily be replicated on 2.x.

…

On Fri, Jan 13, 2017, 4:22 AM Zac Sweers ***@***.***> wrote: I missed that they were internal now. That's disappointing to see, and seems like a bit of a step backward compared to RxJava 1. We'll just copy then :| — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4993 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEEERdKOOoaZajpQio5Pgsyac0sCa0Lks5rR2x0gaJpZM4LiyuF> .

[ZacSweers]

Yeah I was referring to the factory methods. Basically wanted to do the same thing you added in the linked PR above

[ZacSweers]

Those could easily be replicated on 2.x.

@akarnokd any thoughts on this? This is what I was thinking of in opening this issue, in case I wasn't clear before

[akarnokd]

They will be still internal classes. However, we could expose them, for example, Schedulers.newComputation(), Schedulers.newIO() etc where extra parameters can be passed in, similar to how Project Reactor's schedulers were exposed.

[ZacSweers]

That sounds reasonable to me. The APIs they expose cover pretty much everything we'd want, would be happy to make a PR matching those if you're up for it.

[akarnokd]

Sure.