fix(race): unsubscribe raced observables with immediate scheduler

[ktkiiski]

Description:

This fixes a bug with the race operator: whenever a raced Observable emitted a value immediately when subscribed (with the default scheduler), any following Observables were subscribed but never unsubscribed.

Now, if any raced Observable emits a value immediately, any latter Observables are ignored, because for them "the race is already lost".

Also, it is now ensured that the immediately emitting Observable is still added to the RaceSubscriber and will be unsubscribed once the outer Observable is unsubscribed.

[coveralls]

Coverage remained the same at 97.686% when pulling cb98652 on ktkiiski:fix-race-unsubscription into f51b8f9 on ReactiveX:master.

[kwonoj]

Looks good to me.

[trxcllnt]

LGTM, but can the unit tests be implemented as marbles like the rest of them?

[kwonoj]

haven't tried it yet but seems some of the test may able to use marbles. @ktkiiski, what do you think?

[ktkiiski]

I actually first wrote the specs with marbles, but they did not bust the original bug. This is probably because the marble tests are based on VirtualTimeScheduler (?), and it seems that the bug only occurred with the default (immediate) scheduler, when the value was immediately emitted on subscription.

[kwonoj]

Ahhhh forgot that. VirtualTimeScheduler is kind of AsyncScheduler, so if test aims for default scheduler it may not work. @trxcllnt, I think PR's good as-is. any suggestions?

[trxcllnt]

After reviewing the source, it seems the test cold Observables are missing the ability to synchronously emit values on subscription. I was expecting something like cold('(x)--|') to emit x upon subscription. Don't want to block this PR until that gets implemented, so LGTM.

[benlesh]

LGTM as well.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.