Fix issue #1522

[zsxwing]

TakeLast should ignore other requests if it's requested with Long.MAX_VALUE. #1522

[cloudbees-pull-request-builder]

RxJava-pull-requests #1441 SUCCESS
This pull request looks good

[cloudbees-pull-request-builder]

RxJava-pull-requests #1442 SUCCESS
This pull request looks good

[cloudbees-pull-request-builder]

RxJava-pull-requests #1443 SUCCESS
This pull request looks good

[cloudbees-pull-request-builder]

RxJava-pull-requests #1445 SUCCESS
This pull request looks good

[benjchristensen]

This can still just use getAndAdd without the CAS retry loop.

[zsxwing]

Here I try to ignore the further requests after request(Long.MAX_VALUE). I cannot use getAndAdd because if requested == Long.MAX_VALUE, I wanna ignore n. If adding n to requested, and if it happens to execute in while (true) loop in emit method, the loop won't stop because requested < 0.

[zsxwing]

For the overflow, I mean sending a request(n) after request(Long.MAX_VALUE) may make requested overflow easily. c+n is rarely overflow if request != Long.MAX_VALUE (However, we still can handle c+n overflow problem by adding a check c > Long.MAX_VALUE - n).

[benjchristensen]

If we keep the fast path we can just check that request is MAX_VALUE then skip. For example: https://github.com/Netflix/RxJava/blob/bb56093c00185dbd39785a95197abe4b66f3bbca/rxjava-core/src/main/java/rx/internal/operators/OnSubscribeRange.java#L59

[cloudbees-pull-request-builder]

RxJava-pull-requests #1450 FAILURE
Looks like there's a problem with this pull request

[cloudbees-pull-request-builder]

RxJava-pull-requests #1453 SUCCESS
This pull request looks good

[benjchristensen]

Good catch.

[benjchristensen]

I think this looks good, though I'm still curious if the complexity here is worth us having a fast-path. Would be interesting to do JMH testing. In OnSubscribeFromRange it does make a difference, but here I question it because every onNext is passing through a Queue.