Leaky execution related to a race condition when a listener is added to a MemoryStream

[wclr]

http://www.webpackbin.com/Ny3m5bc4-

See console output.

There is three ending listeners attached, one of the them (in the middle) takes 0 (or 1 item), and this causes kind of a leak after all all listeners detached
.

[wclr]

Not pointing on urgency but commenting the weight (could) - this issue needs to be checked and fixed because it may mean that there is something wrong with whole approach to inner cycles detection.

[Hypnosphi]

In fact, this bug has nothing to do with imitate and can be simplified to this: http://www.webpackbin.com/Vywzoz0E-

the problem here is the combination of MemoryStream with synchronous starting value and take(1).

Let's take a look at MemoryStream#_add method

_add(il: InternalListener<T>): void {
    if (this._has) { il._n(this._v); }
    super._add(il);
}

Here, if a value is already present (which is the case if there are already any listeners attached), it is emitted before the listener is really added. In case of take(1) it leads to instant completion of outer stream. TakeOperator therefore stops and tries to remove itself from the source$ MemoryStream internal listeners list. But it isn't present at that list yet, so it's a no-op. Then, in the second line of MemoryStream#_add, it finally adds itself and stays there forever.

A simple solution would be if MemoryStream just added listeners before emitting the current value, but I'm not sure it wouldn't lead to other bugs, so maybe a better option would be if operators removed itself asynchronously.

[staltz]

Fixed in v7.0