fetch2.js

// By default not ref counted
const source = fetchObservable('someurl');

// Everyone gets their own request
const disp1 = source.subscribe(x => console.log(`First ${x}`));
const disp2 = source.subscribe(x => console.log(`Second ${x}`));

// Disposing does not affect one another
disp1.dispose();
disp2.dispose();

// Adding ref counting
const refSource = fetchObservable('someurl').share();

// Add two references
const refDisp1 = refSource.subscribe(x => console.log(`First ${x}`));
const refDisp2 = refSource.subscribe(x => console.log(`Second ${x}`));

// Once the ref count hits zero, the whole operation is cancelled
refDisp1.dispose();
refDisp2.dispose();

// To mimic Promises, we could use AsyncSubject which acts as a thenable
const subject = new Rx.AsyncSubject();

// Will not resolve until both onNext and onCompleted have been called (or onError)
subject.onNext(42);
subject.onCompleted();

subject.subscribe(x => `From AsyncSubject ${x}`)

Assuming cancelUnderlyingFetch() cancelled the underlying fetch, when would fetchObservable's implementation call this (and how)?

@jakearchibald if it cancelled, then it would immediately call onCompleted which then calls dispose and not allow anything new on that stream as per the AutoDetachObserver

And the AbstractObserver is also the gatekeeper to prevent any re-entry which is wrapped by the AutoDetachObserver