Should an rxjs operator return a function or an observable?

Question

I'm a bit confused by the various definitions of an operator in rxjs.

Below I provide some of the definitions:

1 A Pipeable Operator is a function that takes an Observable as its input and returns another Observable.
Creation Operators are the other kind of operator, which can be called as standalone functions to create a new Observable

2 An operator is a function that takes one observable (the source) as its first argument and returns another observable (the destination, or outer observable)

3 Operators take configuration options, and they return a function that takes a source observable.

4 Operators should always return an Observable [..] If you create a method that returns something other than an Observable, it's not an operator, and that's fine.

Since 1,2,4 seem to be conflicting with 3, which definition is the correct one. Is there a better definition of rxjs operators?

For example: in case of map. Is map() itself the operator? Or the operator is the return value of map()?

Answer 1

Yes. map itself is an operator.

Every operator returns an Observable so later on you can subscribe to the Observable you created.

Of course when I say 'you created' I mean that you created via a creation operator or using the Observable class: new Observable

A pipeable operator is just an operator that would be inside the pipe utility function.

Any function that returns a function with a signature of Observable can be piped and that's why you can create your own Observables and pipe operators to it.

I am pretty much sure that you already know all of this and all you want to know is why there is a conflict in what you are reading.

First, you don't have to pipe anything to your Observable. You can create an Observable and subscribe to it and that's it.

Since you do not have to pipe anything to your Observable, what should we do with the operators?

Are they used only within a pipe?

Answer is NO

As mentioned, an operator takes configuration options, in your example:

const squareValues = map((val: number) => val * val);

and then it returns a new function that takes the source Observable so you can later on subscribe to it:

const squaredNums = squareValues(nums);

As we can see, map took (val: number) => val * val and returned a function that now gets the Observable nums: const nums = of(1, 2, 3);

Behind the scenes map will take the source observable and transform according to the configuration.

Now the important thing: Instead of doing that in this way (which you can but no need for that), it is better to pipe your operators so you can combine all of the functions (assuming you are using more operators) into one single function.

So again, behind the scenes you can refer to pipe as a cooler way to make operations on your source Observable rather then declaring many variables that uses different operators with some configuration which return a function that takes a source Observable.

Answer 2

Is map() itself the operator? Or the operator is the return value of map()?

Current implementation of the map() looks like this:

export function map<T, R>(project: (value: T, index: number) => R, thisArg?: any): OperatorFunction<T, R> {
  return function mapOperation(source: Observable<T>): Observable<R> {
    if (typeof project !== 'function') {
      throw new TypeError('argument is not a function. Are you looking for `mapTo()`?');
    }
    return source.lift(new MapOperator(project, thisArg));
  };
}

So, map() is a function. It is an operator in RxJS terms, yes, but it's still a regular JavaScript function. That's it.

This operator receives projection callback function which gets called by the map operator. This callback is something you're passing to map(), e.g. value => value.id from this example:

source$.pipe(map(value => value.id))

The return value of the map is also a function (declared as OperatorFunction<T, R>). You can tell that it is a function since map returns function mapOperation().

Now, mapOperation function receives only one parameter: source which is of type Observable<T> and returns another (transformed) Observable<R>.

To summarize, when you say:

A Pipeable Operator is a function that takes an Observable as its input and returns another Observable.

This means that an RxJS operator (which is a function) is pipeable when it takes an Observable as its input and returns another Observable which in our case is true: a map operator indeed returns a function (mapOperation) whose signature ((source: Observable<T>): Observable<R>) indicates exactly that: it takes one Observable and returns another.

An operator is a function that takes one observable (the source) as its first argument and returns another observable (the destination, or outer observable)

I already mentioned couple of times that an operator is a just function.

Operators take configuration options, and they return a function that takes a source observable.

Yes, in this case, a map() could be called operator since it receives configuration option - a projecttion callback function. So, there's really no conflicts here since many operators are configurable.

I'd say that there's conflict in this one:

Operators should always return an Observable [..] If you create a method that returns something other than an Observable, it's not an operator, and that's fine.

I guess that this is an old definition when pipeable operators weren't pipeable. Before pipeable operators were introduced (I think in version 5 of RxJS), operators were returning Observables. An old map() implementation indicates just that.

For more information about why creators of RxJS decided to introduce pipeable operators, please take a look at this document.

Another great article about what are Observables can be found here.

Also:

Creation Operators are the other kind of operator, which can be called as standalone functions to create a new Observable.

of() is an example of creation operator which returns (creates) an Observable. Please take a look at the source code.

TL;DR: A map() is a function that usually has one parameter (a projection callback function) which also returns a function that receives a source Observable and returns a destination Observable.

EDIT: To answer your question from comments, I'd like to do it here.

Yes, in RxJS 6 you can create a function that accepts observable and returns another one and that would be the operator. E.g.

function myOperatorFunction(s: Observable<any>) {
  return of(typeof s);
}

and you'd call it like

source$.pipe(myOperatorFunction);

Please notice that I didn't call myOperatorFunction in pipe(), I just passed the reference to it, i.e. I didn't write myOperatorFunction with parenthesis, but without them. That is because pipe receives functions.

In cases where you need to pass some data or callback functions, like in map example, you'd have to have another function that would receive your parameters, just like map receives projection parameter, and use it however you like.

Now, you may wonder why there are operators that don't receive any data, but are still created as functions that return function, like refCount(). That is to coincide with other operators that mostly have some parameters so you don't have to remember which ones don't receive parameters or which ones have default parameters (like min()). In case of refCount, if it was written a bit different than it is now, you could write

source$.pipe(refCountOperatorFunction);

instead of

source$.pipe(refCount());

but you'd have to know that you have to write it this way, so that is the reason why functions are used to return functions (that receives observable and returns another observable).

EDIT 2: Now that you know that built in operators return functions, you could call them by passing in source observable. E.g.

map(value => value.toString())(of())

But this is ugly and not recommended way of piping operators, though it would still work. Let's see it in action:

of(1, 2, true, false, {a: 'b'})
  .pipe(
    map(value => value.toString()),
    filter(value => value.endsWith('e'))
  ).subscribe(value => console.log(value));

can be also written like this:

filter((value: string) => value.endsWith('e'))(map(value => value.toString())(of(1, 2, true, false, {a: 'b'})))
  .subscribe(a => console.log(a));

Although this is a completely valid RxJS code, there's no way you can think of what it does when you read the latter example. What pipe actually does here is that it reduces over all the functions that were passed in and calls them by passing the previous source Observable to the current function.