Add Single, Maybe and Completable

[passsy]

RxJava 2 has the concept of Single, Maybe and Completable. They got quickly adopted by RxSwift 3.3.0.

I couldn't find anything about those reactive types in rxjs. Are they planned?

[kwonoj]

No, it's not planned. It is not a goal of v5 to create full-compliant operator set to other languge bindings or even v4, and there are some added opeator or dropped operator between. Once there is suffecient usecase and demand it can be obviously discussed in further.

[aluanhaddad]

I'm also not sure I would look to RxJava for inspiration. Java is severely lacking in expressive power, even with its recent editions. JavaScript has long had better expressiveness as have most other languages.

[kwonoj]

I'm closing this issue for now, haven't got subsequent details of operator to initiate discussion to be included in core sets. If this operator seems need to be included in core operator sets, please refer #2299 and #2524 as starting baseline to include details for discussions.

[passsy]

Just want to make sure this issue didn't got closed accidentally. It's not about new operators. It's about other base classes besides Observable

• Observable: 0..N flows, no backpressure
• Single: a flow of exactly 1 item or an error
• Completable: a flow without items but only a completion or error signal
• Maybe: a flow with no items, exactly one item or an error

[kwonoj]

@passsy yes, while attached references are saying only about operators, you can consider same for any other cases like this one, or either removing, changing existing operator behavior as well. There is no big differences of having new operator, or either new base class in perspective of discussing it to add into core set.

[whymarrh]

If y'all are open to a bit of discussion on the topic, I do think there are use cases for a type analogous to RxJava's Completable:

Represents a deferred computation without any value but only indication for completion or exception.

That said, the workaround isn't too bad right now, one can (assuming the stream is hot) filter all of the values (e.g. .filter(() => true)).

Example use case

In a redux-observable epic, if you want to dispatch an event after some chain of other actions.

A bit of setup using the example they have on their website:

const PING = 'PING';
const PONG = 'PONG';

const ping = () => ({ type: PING });

const pingEpic = action$ =>
  action$.ofType(PING)
    .delay(1000)
    .mapTo({ type: 'PONG' });

const pingReducer = (state = { isPinging: false }, action) => {
  switch (action.type) {
    case 'PING':
      return { isPinging: true };

    case 'PONG':
      return { isPinging: false };

    default:
      return state;
  }
};

Now if you wanted to dispatch an action after count number of PONG actions have occurred:

const START_ROUND_TRIPS = 'START_ROUND_TRIPS';
const END_ROUND_TRIPS = 'END_ROUND_TRIPS';

const endRoundTrip = (count) => ({type: END_ROUND_TRIPS, payload: {count}})

const roundTripEpic = action$ =>
  action$.ofType(START_ROUND_TRIPS)
    .concatMap((action) => Observable.concat(
      Completable.from(action$.ofType(PONG).take(action.payload.count)), // Up for discussion
      Observable.of(endRoundTrip(action.payload.count))
    ))

It's important that the inner observable not actually contain the actions$.ofType(PONG) values because that would re-dispatch those actions. It's arguable whether this is preferable over .filter(() => true) (I think it's a touch more readable) but I do think it is a use case.

[trxcllnt]

@passsy @whymarrh isn't this already possible w/ the single(), defaultIfEmpty() and ignoreElements() operators?

[whymarrh]

(I had forgotten about ignoreElements(), that works better than the .filter(() => true), thanks!)

[I]sn't this already possible w/ the single(), defaultIfEmpty() and ignoreElements() operators?

Some combination of those work as a replacement for a hypothetical Completable, yup.

[WowMuchName]

From a pure technical point of view one might argue that a completeable is unnecessary since an observable can be used for the job.
However when designing an api it is useful to know whether 0, 1 or N objects can be emitted by a stream. While this should be documented in the jsdoc or javadoc, knowing that by looking at the class of stream makes it a lot easier. It is also useful if those classes enforce those cardinality-constraints so you don't accidentally break your api-contract.
As the rx guy in our team I would also like to point out that there is a barrier of entry you tend to forget about when having worked with rx for some time. Having classes that are called completable, single and maybe - even if they are all essentially observables under the hood - is much easier for rx beginners to understand and work with.

[darkyelox]

You can simple create the objects in next releases, is useful to have them for some cases and also makes the code more readable (knowing for what a Completeable, Single and a Maybe are for)

[ptitjes]

@trxcllnt Just wanted to point out something. Yes, all of this is possible with combinations of single(), defaultIfEmpty() and ignoreElements(). But I guess having types like Single, Maybe and Completable isn't about runtime expressiveness. It is about what it conveys in the type system: why have only void when you can also have null, undefined, never, etc. Whether they should be expressed as classes or type aliases is another problem (I'd bet on type aliases however).

[sawthompson]

Would it be possible to revisit this? I think @WowMuchName makes some compelling points. Personally I have found that using Observables as Singles, while technically possible, creates ambiguity and leads to bugs.

[jaumard]

+1 on this, Single/Maybe and Completable make it more understandable and maintainable by explicitly saying what behavior you'll have. Having just Observable is for me as bad as saying Any on return type of a method, you don't know what you'll have except if you actually look the implementation...

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.