replay subject should ignore emitted values after completion

[FDIM]

Bug Report

Seems like replay subject behavior is a bit inconsistent. If one subscribes before the subject completes, only values emitted until completion are sent to observers - all good. Now if I subscribe after it completes, it will replay all the values - even if they were emitted after subject completion.

Reproduction
https://stackblitz.com/edit/rxjs-fyepk5

import { ReplaySubject } from 'rxjs'; 

const ready = new ReplaySubject();
ready.next(1);
ready.complete();
ready.next(2);

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

Current Behavior
1 and 2 are logged to console

Expected behavior
Only 1 should be logged

Environment

• RxJS version: 6.3.3

[cartant]

With no buffer size specified, the buffer size is infinite - not one.

[FDIM]

Yes, I know that - but still, if the subject completes - why should it keep collecting values in the buffer?

[cartant]

Because the default behaviour was not changed. If refCount is not specified and is not set to true the subscription to the source is not unsubscribed.

The behaviour is described here and in that PR's initial description. It's likely the case that the docs can be improved - as I was unable to determine how to express the overload signatures in the TSDoc - but I don't believe this is a bug. Feel free to change this issue to an improve-the-docs issue.

[FDIM]

Thanks for technical details, but I'd argue that from API standpoint this seems to be a bug. Easily fixable in client code though so no big issue.

[cartant]

We're well aware that there are two camps here: one deems it a bug; the other does not. However, the behaviour was not changed (back) by the above referenced PR, so this isn't a bug.

[FDIM]

Not sure why you keep referencing the PR, it's not in 6.3.3. A discussion and a change in the future may be nice - so maybe the issue could stay open

[cartant]

That PR is in 6.4 which has just been released.

A discussion and a change in the future may be nice - so maybe the issue could stay open

For the discussions that have been had - ad nauseam - see that PR and all of the issues it references.

As this issue relates to 6.3.3, it can stay closed as a dupe.