search

dotnet / reactive

The Reactive Extensions for .NET
http://reactivex.io
MIT License
6.66k stars 745 forks source link

Add Share() as an alias for Publish().RefCount() #394

Closed khellang closed 4 years ago

khellang commented 7 years ago

Doing .Publish().RefCount() is a pretty common combination.

I see the ports to Java, Scala, Groovy, JS, Kotlin and PHP has this alias already; http://reactivex.io/documentation/operators/refcount.html

Is it worth adding it? I can send a PR you're interested.

shiftkey commented 7 years ago

:thumbsup: I'm not opposed to a well-documented extension method here to make this concept easier to understand.

RxDave commented 7 years ago

For the sake of argument, is it worth considering that often the use of Publish().RefCount() is a code smell? And perhaps we shouldn't encourage its usage by making it too easy to use?

The issue is that many people overuse subjects to begin with, and it seems that the next step along their journey to better reactive programming is to replace their usage of subjects with Publish().RefCount(), which is just as bad, if not worse.

Devs (especially those new to Rx) seem to use RefCount without realizing that the side effects they are sharing may occur again, if the reference count drops down to 0 and then back to 1.

Furthermore, in many cases you can still use Publish without RefCount, if you really need to, by using the overload of Publish that accepts a selector function. It's still using a Subject internally to get the job done, but at least it doesn't mess with the temperature of the query as a whole. Only the selector function is hot. This also may encourage better reactive designs because it's more of a functional approach. Devs often use RefCount as a result of breaking apart their queries into many types, and then gluing them back together again. I believe this is a bad approach to reactive programming. It's much more like OOP -- with events.

Anyway, if you're going to introduce a Share operator, then perhaps it's imperative that you also offer a selector overload, and encourage its use over the other overloads. Devs will probably continue to avoid using it because it does require a bit more mental gymnastics, and forces us out of a strictly OOP design, but at least they'll have the option.

akarnokd commented 6 years ago

Now that there is a timeout version of refCount, at least there is some value in refCount. Otherwise, I'm not convinced we should over-encourage the usage of it and I'd probably remove share() from RxJava if it didn't break some important user's codebase.

akarnokd commented 4 years ago

Closing due to inactivity.