Reduce the scope of locking to increase parallelism.

[vigdorchik]

In my rxmon project https://github.com/vigdorchik/rxmon I make heavy use of combineLatest. This change that reduces the scope of locking should help increase combineLatest throughput. I wish I could report the speedup numbers, but I couldn't find how to publish RxJava to maven locally to try it out.

[cloudbees-pull-request-builder]

RxJava-pull-requests #961 FAILURE
Looks like there's a problem with this pull request

[vigdorchik]

I saw this failure without my change, and I believe it's unrelated.

On Wed, Apr 16, 2014 at 5:14 PM, CloudBees pull request builder plugin <
[email protected]> wrote:

RxJava-pull-requests #961https://netflixoss.ci.cloudbees.com/job/RxJava-pull-requests/961/FAILURE
Looks like there's a problem with this pull request

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1043#issuecomment-40596720
.

[vigdorchik]

Review by @benjchristensen and @akarnokd.

[benjchristensen]

I'll take a look ... sounds like a good change.

We have started doing perf testing using JMH. You can take a look at examples here: https://github.com/Netflix/RxJava/tree/master/rxjava-core/src/perf/java/rx/operators

It would be useful to add one for this operator to measure before/after performance.

[akarnokd]

The change looks okay but not certain about the performance gains. Two other things:

• It would be great if you could rewrite it to the new Operator/Subscriber idiom.
• The original had a race condition between next and error, you'd need to wrap the incoming observer/subscriber into a SerializedSubscriber.

[vigdorchik]

This change doesn't work. It allows onCompleted before all onNext have been triggered.