throttle operator throws error RangeError: Maximum call stack size exceeded

[OliverJAsh]

RxJS version:
6.6.3 (latest at time of writing)

Code to reproduce:

https://stackblitz.com/edit/rxjs-compat-ywrxqr

import * as Rx from "rxjs";
import { throttle } from "rxjs/operators";

const source = Rx.of(1).pipe(
  throttle(() => Rx.EMPTY, { leading: false, trailing: true })
);

source.subscribe(x => console.log(x));

Expected behavior:
It logs 1 once.

Actual behavior:
It logs 1 many times and then errors with RangeError: Maximum call stack size exceeded.

Additional information:

The reduced test case is a bit contrived. In practice my duration selector has a bit more logic:

throttle(n => n === 10 ? Rx.timer(100) : Rx.EMPTY, { leading: false, trailing: true })

[OliverJAsh]

I was hoping I could workaround it like so:

-throttle(() => Rx.EMPTY, { leading: false, trailing: true })
+throttle(() => Rx.scheduled(Rx.EMPTY, Rx.async), { leading: false, trailing: true })

… so it completes asynchronously rather than synchronously. However IIUC this doesn't work because of another bug with throttle: #5732.

[OliverJAsh]

Note this issue only occurs when trailing: true. Example that uses trailing: false: https://stackblitz.com/edit/rxjs-compat-mygu4y.

[OliverJAsh]

In RxJS v7 beta, the behaviour is different but still incorrect (at least in my mind): https://stackblitz.com/edit/rxjs-compat-mmjf6c. This never logs anything. Update: never mind, see #6034 (comment).

[josepot]

In RxJS v7 beta, the behaviour is different but still incorrect (at least in my mind): https://stackblitz.com/edit/rxjs-compat-mmjf6c. This never logs anything.

That's the behavior that I would expect: given the fact that leading is set to false and that the source is a sync observable that emits once and completes in the same tick.

What's the behavior that you were expecting?

[OliverJAsh]

My understanding was that if trailing is set to true, the throttled Observable should always emit the last value that the source emitted before completion.

#5732

[josepot]

My understanding was that if trailing is set to true, the throttled Observable should always emit the last value that the source emitted before completion.

#5732

I could be completely wrong, of course, but what I would expect from trailing set to true is for the resulting observable to emit the latest value that the source emitted while the durationSelector observable was "active". If the source emits and completes synchronously, then I would expect for the operator to not capture any values during the durationSelector observable... Because well, there is no point to subscribing to the durationSelector, since the source has completed already. Again, I could be completely wrong, of course. I'm just saying how I understand this operator.

[OliverJAsh]

Another weird discovery: in v7.0.0-beta.10 this doesn't throw an error but it does still log 1 over 1000 times.

import * as Rx from "rxjs";
import { throttle } from "rxjs/operators";

const source = Rx.of(1).pipe(
  throttle(() => Rx.of(1), { leading: false, trailing: true })
);

source.subscribe(x => console.log(x));

https://stackblitz.com/edit/rxjs-compat-vswz6n

If someone can confirm this is not expected behaviour, I can try to send a PR with a test and a fix.

If I understand correctly, this is the expected behaviour as @josepot says:

If the source emits and completes synchronously, then I would expect for the operator to not capture any values during the durationSelector observable

So source should emit nothing and complete synchronously?

---

In RxJS v7 beta, the behaviour is different but still incorrect (at least in my mind): https://stackblitz.com/edit/rxjs-compat-mmjf6c. This never logs anything.

Ah, I believe this is because of this change: 4af0227. In my example the duration selector never emits anything, it only completes. That makes sense.

[benlesh]

The original bug is fixed in 7.0.0.

However, the bug reported here is still outstanding.

It begs the question "why would anyone do this?" but it still needs to be fixed.

[benlesh]

Closing this, as it's fixed in 7. I'm uncertain that this needs fixed in 6, as it's sort of an edge case, but I'm happy to reopen it and take anyone's PR on the matter if they want to put in the time.

[benlesh]

Other issue is here: #6058