Split creation functions from creation functions with Schedulers

[benlesh]

Currently we have creation functions like so:

• from(obj) and from(obj, scheduler)
• range(a, b) and range(a, b, scheduler)
• (... and many more)

Problem

Schedulers and scheduling adds a LOT of bloat to bundles and we can't currently avoid it.

If you look at the implementations of these you'll see an obvious pattern, especially in the ones I more recently created as "just functions":

function fromSomething<T>(something: Something<T>, scheduler?: IScheduler) {
  if (!scheduler) {
     return new Observable(/* a little bit of code */);
  } else {
    return new Observable<T>(subscriber => {
      /* OMG WHAT?!! A HUGE MOUNTAIN OF CODE AND A BUNCH OF TYPES
          WE NEED TO IMPORT FOR THIS SO WE CAN DO THINGS
          WITH SCHEDULERS, WHICH DON'T GET USED 95% OF THE TIME */
    });
  }
}

/* MORE STUFF DOWN HERE FOR SCHEDULING SUPPORT */
function dispatch() {
  /* you get the idea */
}

Proposal

• Split creation methods into "normal" and "scheduled" flavors: from and fromScheduled, defer and deferScheduled, range and rangeScheduled etc.

If we do this, then the majority of RxJS usage (which doesn't involve schedulers) can avoid pulling in scheduler-related classes and scheduling-related logic.

That would make the above into:

95% use case

function fromSomething<T>(something: Something<T>) {
     return new Observable(/* a little bit of code */);
}

and else where

5% use case

function fromSomethingScheduled<T>(something: Something<T>, scheduler: IScheduler) {
    return new Observable<T>(subscriber => {
      /* OMG WHAT?!! A HUGE MOUNTAIN OF CODE AND A BUNCH OF TYPES
          WE NEED TO IMPORT FOR THIS SO WE CAN DO THINGS
          WITH SCHEDULERS, WHICH DON'T GET USED 95% OF THE TIME */
    });
}

/* MORE STUFF DOWN HERE FOR SCHEDULING SUPPORT */
function dispatch() {
  /* you get the idea */
}

Risks

It will be a breaking change for people using schedulers today. Easy to fix, but annoying.

[kwonoj]

Do we have guesstimate of user numbers uses scheduler when create? I do obviously, and all of my test cases are. I'm feeling it might be confusing number of api surface even if separated out as own export 🤔

[benlesh]

@kwonoj I can do a search of google projects to get a pretty good ballpark number.

[kwonoj]

Plus if fns are separated, what's recommended pattern to inject testscheduler for observables?

[benlesh]

Plus if fns are separated, what's recommended pattern to inject testscheduler for observables?

That wouldn't change... from(whatever) with no Scheduler would never need a TestScheduler.

The change proposed above would obviously not include things like timer and interval.

[cartant]

Is the proposal for a public rxjs/schedule import location? Is that really needed if there aren't name conflicts? Introducing another location seems to be at odds with #3262.

Otherwise, sounds good to me.

[benlesh]

@cartant you're right, we can probably keep them all in the same spot. 🤷‍♂️

(I changed the OP to reflect this)

[benlesh]

Done.