I have two observables A and B, and I'd like to create an observable which emits the latest value from B, but if B hasn't emitted anything yet, it emits the latest value from A. I was thinking something like this:
merge(A.takeUntil(B), B)
But this seems a little counterintuitive, and I'd like to have a more readable approach. Is there a more canonical Rx way to do this?
Custom Operator
This solution is very readable in use but complicated by the fact that you hide the complexity within a custom operator. The benefit of a custom operator is that you only subscribe to each source observable once. This means your observables don't have to be "hot".
Unfortunately, this operator breaks synchronously executing observables and has each value pass through the event loop.
function prefer<T, R>(...observables: Observable<R>[]): Observable<R>{
return new Observable(observer => {
const subscrptions = new Array<Subscription>();
const unsub = (index) => {
for(let i = index; i < subscrptions.length; i++){
subscrptions[i].unsubscribe();
}
}
observables
.map(stream => publish()(stream))
.forEach((stream, index) => {
subscrptions.push(stream.subscribe((payload: R) => {
observer.next(payload);
unsub(index + 1);
subscrptions.length = index + 1;
}));
stream.connect();
});
return { unsubscribe: () => unsub(0) }
})
}
Operator in Use
prefer(
interval(10000).pipe(map(_ => "Every 10,000")),
interval(5000).pipe(map(_ => "Every 5,000")),
interval(1000).pipe(map(_ => "Every 1,000")),
interval(250).pipe(map(_ => "Every 250"))
).subscribe(console.log);
Related
I want combineLatest functional but for variable count of observables.
Something like:
// init combineLatest of three observables
[1, 2, 3]
// first observable produced new value "2"
[2, 2, 3]
// third observable ended
[2, 2]
// first observable produced new value "1"
[1, 2]
// new observable added
[2, 2, 4]
Is it possible in RxJS?
If I understand the problem right, the solution is pretty tricky for something that looks innocent.
I try to go step by step to explain a potential solution.
First of all we need understand that there are 3 different events that we need to manage:
the fact that one Observable completes
the fact that one Observable is added to the array which is given to combineLatest
the fact that a new array of Observables has to be passed to combineLatest, either because we are at the beginning of the processing (i.e. with the initial array) or because we have added a new Observable or because one Observable has completed
The second thing that we need to recognize is that we need to store the array of Observables we pass to combineLatest in a variable, otherwise we are not able to add or remove Obaservables from there.
Once these things are clear, we can build a solution in the form of a function that returns 2 things:
the Observable that we want to subscribe to and that should have the behavior that we are looking for
a Subject that we can use to communicate that we want to add a new Observable to the combineLatest function
The last point we need to recognize is that any time we change the list of Observable, either because we add or because we remove an Observable (because it completed), we need to run the combineLatest function with the new fresh list of Observables.
Now that all this has been clarified, this is the code of the function that returns an Observable which behaves as described
function dynamicCombineLatest(startingObservables: Observable<any>[]) {
// this is the variable holding the array of Observables
let observables = startingObservables;
// this is the array that contains the list of Observables which have been, potentially, transformed to emit
// immediately the last value emitted - this happens when a new Observable is added to the array
let observablesPotentiallyWithLastValueImmediatelyEmitted =
startingObservables;
// this is the variable holding the array of values last notified by each Observable
// we will use it when we need to add a new Observable to the list
const lastValues = [];
// this are the Subjects used to notify the 3 different types of events
const start = new BehaviorSubject<Observable<any>[]>(observables);
const add = new Subject<Observable<any>>();
const remove = new Subject<Observable<any>>();
let skipFirst = false;
// this is the chain of operations which must happen when a new Observable is added
const addToObservables = add.pipe(
tap({
next: (obs) => {
console.log("add");
// we need to make sure that the Observables in the list will immediately start to emit
// the last value they emitted. In this way we are sure that, as soon as the new added Observable emits somthing,
// the last value emitted by the previous Observables will be considered
observablesPotentiallyWithLastValueImmediatelyEmitted = observables.map(
(o, i) => {
return startWith(lastValues[i])(o);
}
);
// the new Observable is added to the list
observables.push(obs);
observablesPotentiallyWithLastValueImmediatelyEmitted.push(obs);
},
})
);
// this is the chain of operations which must happen when an Observable is removed
const removeFromObservables = remove.pipe(
tap({
next: (obs) => {
const index =
observablesPotentiallyWithLastValueImmediatelyEmitted.indexOf(obs);
console.log("remove");
// we simply remove the Observable from the list and it "last value"
observablesPotentiallyWithLastValueImmediatelyEmitted.splice(index, 1);
observables.splice(index, 1);
lastValues.splice(index, 1);
// we make sure that the Observables in the list will immediately start to emit with the last value they emitted
observablesPotentiallyWithLastValueImmediatelyEmitted = observables.map(
(o, i) => {
return lastValues[i] ? startWith(lastValues[i])(o) : o;
}
);
// we set that the first value of the new combineLatest Observable will be skipped
skipFirst = true;
},
})
);
// here we merge the 2 chains of operations so that both add and remove logic will be executed
// when the relative Subjects emit
merge(addToObservables, removeFromObservables).subscribe({
next: () => {
console.log("new start");
// we notify that a change in the Observable list has occurred and therefore we need to unsubscribe the previous "combineLatest"
// and subscribe to the new one we are going to build
start.next(observablesPotentiallyWithLastValueImmediatelyEmitted);
},
});
// this is where we switch to a new Observable, result of the "combineLatest" operation,
// any time the start Subject emits a new Observable list
const dynamicObservables = start.pipe(
switchMap((_observables) => {
const _observablesSavingLastValueAndSignallingRemove = _observables.map(
(o, i) =>
o.pipe(
tap({
next: (v) => {
// here we save the last value emitted by each Observable
lastValues[i] = v;
},
complete: () => {
// here we notify that the Observable has completed and we need to remove it from the list
remove.next(o);
},
})
)
);
console.log("add or remove");
// eventually this is the Observable created by combineLatest with the expected array of Observables
const _combineLatest = combineLatest(
_observablesSavingLastValueAndSignallingRemove
);
const ret = skipFirst ? _combineLatest.pipe(skip(1)) : _combineLatest;
skipFirst = false;
return ret;
})
);
// here we return the Observable which will be subscribed to and the add Subject to be used to add new Observables
return { dynamicObservables, add };
}
You can look at this stackblitz for an example.
Buffer and combine based on a key
Here's a slight variant of what you're asking for. It works just like mergeAll, only it keeps a buffer and emits the latest for any observable that have emitted so far.
The varient here is that you need to supply string keys for your values to get attached to. You should be able to see how to turn this into array indices if you so choose.
The reason I haven't done this with an array is because there's no much undefined behavior. For example, if the first observable completes and the second observable emits, your elements are all opaquely re-ordered.
Using keys returns control back to the caller, who can just use Object.keys() if they don't care about indices/labels for their data.
Here you are:
interface LabeledObservable<T> {
label: string,
stream: Observable<T>
}
interface CombinedLatest<T> {
[key:string]: T
}
function combineLatestAll<T>():
OperatorFunction<
LabeledObservable<T>,
CombinedLatest<T>
>
{
return source$ => defer(() => {
const buffer = {};
return source$.pipe(
mergeMap(({label, stream}) => stream.pipe(
map(v => {
buffer[label] = v;
return {...buffer};
}),
finalize(() => {
delete buffer[label];
})
))
);
});
}
Subject for new observables
If you like the idea of a subject you can use to inject new observables into your combineLatest operator, this still allows that. The only alteration needed is that you must supply unique labels for your observables. If you don't care about the labels, you can just use any ID generator pattern (Like incrementing a global id counter or something).
const startingObservables: Observable<any>[] = /*some observables */;
const add = new Subject<LabeledObservable<any>>();
add.pipe(
combineLatestAll()
).subscribe(console.log);
startingObservables.forEach((stream,i) => {
add.next({label: "" + i, stream});
});
I asked a question
Is Observable from chained promises equivalent of observables created with from and chained with concatMap?
on totally false premises. It seems that neither of my solutions had nothing to do with my intention.
I created a method that returns Observable and calls 2 methods returning Promise. I tried 2 ways:
public setItemInfos(itemInfos: IItemInfo[]): Observable<number> {
return from(this.db.selectionItemInfos.clear().then(() => {
return this.db.selectionItemInfos.bulkAdd(itemInfos);
}));
}
public setItemInfos(itemInfos: IItemInfo[]): Observable<number> {
const clear$ = from(this.db.selectionItemInfos.clear());
const bulkAdd$ = from(this.db.selectionItemInfos.bulkAdd(itemInfos));
return clear$.pipe(concatMap(() => bulkAdd$))
}
the use will be:
myService.setItemInfos(itemInfos).subsribe(count => {
console.log(`Cleared the table 1st and then added ${count} new items`);
});
I thought from both versions that:
table clear is execution is finished when bulkAdd starts
when bulkAdd is finished i get the count from that in subscribe
How this should really be done? Or can it be done?
This is (from what I can tell here), how I would do it.
In general, defer (or any higher-order operator) is a better way to create an observable from a promise. Defer lets you take the eager evaluation semantics of promises and turn them into the lazy evaluation semantics of observables.
Then all the usual observable operators and such will function as expected.
public setItemInfos(itemInfos: IItemInfo[]): Observable<number> {
const clear$ = defer(() => this.db.selectionItemInfos.clear());
const bulkAdd$ = defer(() => this.db.selectionItemInfos.bulkAdd(itemInfos));
return concat(clear$, bulkAdd$);
}
Update 1:
So I think I might know what you're after. This isn't really idiomatic RxJS since it's such an interleaving mix of declarative, imperative style of code. Even so, this should work? I haven't tested it fully, but some tinkering and I think this should do what you're after.
There's most assuredly a better way to accomplish the same thing, but without seeing the bigger picture of what you're after, it's hard to say.
interface Tagged<T> {
payload: T,
tag: number
}
class abitraryClass{
private setItemInfoSub: Subject<Tagged<IItemInfo[]>>;
private processItemInfo: Observable<Tagged<number>>;
private itemInfoTag = 0;
constructor(){
this.setItemInfoSub = new Subject<Tagged<IItemInfo[]>>();
this.processItemInfo = this.setItemInfoSub.pipe(
concatMap(({tag, payload: itemInfos}) => this.db.selectionItemInfos.clear().pipe(
ignoreElements(),
concatWith(defer(() => this.db.selectionItemInfos.bulkAdd(itemInfos))),
map(response => ({
payload: response,
tag
}))
)),
shareReplay(1)
);
// Make the processing pipeline live at all times.
this.processItemInfo.subscribe();
}
public setItemInfos(itemInfos: IItemInfo[]): Observable<number> {
const myTag = this.itemInfoTag++;
this.setItemInfoSub.next({
payload: itemInfos,
tag: myTag
});
return this.processItemInfo.pipe(
filter(({tag}) => tag == myTag),
map(({payload}) => payload)
);
}
}
Is there a better way to re-write this code and avoid chaining of subscriptions ?
Why am I chaining? because I need to the output of source1$ in child subscriptions
And also I have if conditions because I want to call child subscriptions conditionally
PS i checked solution in this post
Here is the stackblitz link and code
import { from } from 'rxjs';
//emit array as a sequence of values
const source1$ = from([1]);
const source2$ = from([2]);
const source3$ = from([3]);
const useCond1 = true; // this is dynamic can be false too
const useCond2 = true; // this is dynamic can be false too
source1$.subscribe(val => {
if (useCond1) {
source2$.subscribe(() => {
console.log('val from source1 in source2', val);
});
}
if (useCond2) {
source3$.subscribe(() => {
console.log('val from source1 in source3', val);
});
}
});
Not sure, but it seems that you need switchMap or mergeMap and iif
from rxjx doc:
import { fromEvent, iif, of } from 'rxjs';
import { mergeMap, map, throttleTime, filter } from 'rxjs/operators';
const r$ = of(`I'm saying R!!`);
const x$ = of(`X's always win!!`);
fromEvent(document, 'mousemove')
.pipe(
throttleTime(50),
filter((move: MouseEvent) => move.clientY < 210),
map((move: MouseEvent) => move.clientY),
mergeMap(yCoord => iif(() => yCoord < 110, r$, x$))
)
.subscribe(console.log);
Yes, there is a better way!
RxJS provides many different operators and static functions for combining, filtering, and transforming observables. When you use what the library provides, you do not need to have nested subscriptions.
In general, I find it simpler to not do any logic at all inside the subscribe, but rather design observables that emit the exact data that is needed.
A simplistic example could look like this:
someValue$ = source1$.pipe(
switchMap(val1 => useCond1 ? source2$ : of(val1))
);
someValue$.subscribe();
switchMap will subscribe to an "inner observable" whenever it receives an emission. The logic above says to either return the value emitted from source1$ (val1) or return whatever source2$ emits depending on the value of useCond1.
So source2$ will only get subscribed to when useCond1 is true;
Note: the function inside switchMap should return an observable (because switchMap subscribes to it), so of was used to turn the emitted value into an observable.
In your case, let's assume you want to emit some calculated value, based possibly on the other two sources.
We can use combineLatest to create a single observable based on 3 different sources. Since you only want to optionally call source2$ and source3$, we can define the sources based on your conditions. We can then use map to transform the array of values from the 3 sources, into the desired output:
someValue$ = source1$.pipe(
switchMap(val1 => {
const s1$ = of(val1);
const s2$ = useCond1 ? source2$ : of('default val2');
const s3$ = useCond2 ? source3$ : of('default val3');
return combineLatest([s1$, s2$, s3$]);
}),
map(([val1, val2, val3]) => {
return ... // your logic to return desired value
})
);
combineLatest will emit an array containing the latest emissions from each source whenever any source emits. This means someValue$ will emit the latest calculated value whenever any of the sources change.
Following use case: A user can join 0...* groups. Each group has an ID and contains 0...* posts.
I subscribe to an Observable (to get the groups of the user he joined) and this returns an array of strings (the group IDs).
const groupIds$ = of(['a', 'b', 'c']);
If I only had one I would now use switchMap and return this new observable and subscribe to it to get the posts from the group.
But so I have an array and so this isn't working. Does anyone has an idea which RxJS operator(s) can achieve this to get the posts from all groups?
Or does no operator for such use case exist and I have to do it separately at subscribe?
You can use ideally forkJoin if you know you'll have all source Observables as an array:
groupIds$.pipe(
concatMap(groups => {
const observables = groups.map(id => ...);
return forkJoin(...observables);
})
).subscribe(...);
forkJoin will emit a single array with all the results in the same order as in groupIds$.
Eventually if you don't care about the order and just want to get all the results in parallel you can use merge instead of forkJoin (I mean the "Observable creation method" called merge imported directly from rxjs. Not the merge operator from 'rxjs/operators').
Everyone who arrives at this question, here is the answer for "switchMap for Array of strings" (thanks to martin). You only have to use 'merge' from 'rxjs' (not the operator!). Inside switchMap return merge and you are done:
groupIds$.pipe(
switchMap(groups => {
const observables = groups.map(id => {
// your function that returns the observable
return something.getObservable(id);
});
return merge(...observables);
})
).subscribe(data => {
console.log('DATA', data);
});
Possible solution: https://www.learnrxjs.io/operators/transformation/mergemap.html.
const groupIds$ = of(['a', 'b', 'c']);
const myPromise = id =>
new Promise(resolve => resolve(something.getPostsByGroupId(id)));
const posts$ = groupIds$.pipe(
mergeMap(
id => myPromise(id),
(valueFromSource, valueFromPromise) => {
return valueFromPromise;
}));
posts$.subscribe(...);
Terser improvement on great existing answers. This emits an array of all posts each time the groupIds$ source emits.
groupIds$.pipe(
switchMap(ids => zip(...ids.map(id => service.getPosts(id)))),
map((posts: Array<Post[]>) => [].concat.apply([], posts))
).subscribe(posts => /*is an array of all posts*/);
I have a state object with string keys and values. Events are coming in, containing key-value pairs to change the state.
I need a debounced stream that:
validates the events and drop all modifications in the debounce cycle if they lead to an invalid state
outputs the diff to the last valid state
For example, for the initial state of {k1: "v1"}, and an event of {k2: "v2"}, output {k2: "v2"}.
But for the events: {k3: "v3"} and {k4: "invalid"}, drop both changes. So when a new event {k5: "v5"} comes in, the k3 key is still undefined.
I was able to implement it, but only by using a new Subject that keeps track of the last valid state: (jsfiddle)
const lastValidState = new Rx.Subject();
const res = modifications
.buffer(debounce)
.withLatestFrom(lastValidState.startWith(state))
.map(([mods, last]) => {
// calculate next state
return [Object.assign({}, last, ...mods), last];
}).filter(([newState]) => {
// check new state
return Object.keys(newState).every((k) => !newState[k].startsWith("invalid"));
// update Subject
}).do(([newState]) => lastValidState.next(newState)).share()
.map(([newState, last]) => {
// output diff
return Object.assign({}, ...Object.keys(newState).filter((k) => newState[k] !== last[k]).map((k) => ({[k]: newState[k]})))
}
)
This code works well, but I don't like the new Subject it introduces. I would prefer a solution that does not rely on that and use only RxJS operators.
I've tried to use pairwise, but I could not figure out how to pair a stream with the last value of itself.
Thanks to cartant's comment, using scan is the way to go.
The only trick is to use distinctUntilChanged to prevent emits for invalid changes.
The modified code: jsfiddle
const res = modifications
.buffer(debounce)
.scan((last, mods) => {
const newState = Object.assign({}, last, ...mods);
const valid = Object.keys(newState).every((k) => !newState[k].startsWith("invalid"));
if (valid) {
return Object.assign({}, ...Object.keys(newState).filter((k) => newState[k] !== last[k]).map((k) => ({[k]: newState[k]})));
}else {
return last;
}
}, state)
.distinctUntilChanged()