kotlinx.coroutines: Optimize `runBlocking` and `EventLoop` implementations

EventLoopBase implementation class should allocate both queue and delayed lazily, moreover, a queue shall not be allocated as long as there is at most one queued task (optimization that is similar to JobSupport)
DispatchedContinuation shall extend QueuedTask, so that common case of EventLoop usage does not involve creation of objects at all.
Consider providing a reusable version of EventLoop implementation, so that performance-sensitive code can create a single instance that is reused for multiple sequential invocations of runBlocking.

About this issue

Original URL
State: closed
Created 7 years ago
Comments: 15 (12 by maintainers)

Commits related to this issue

EventLoop and runBlocking performance is improved; node-less lock-free MPSC resizeable queue is implemented for that. Fixes #190 — committed to Kotlin/kotlinx.coroutines by elizarov 7 years ago

Most upvoted comments

No. Saving small object to ThreadLocal is usually counter-productive. Access to ThreadLocal is typically more expensive than allocation/collection of small objects. TheadLocal pooling is usually beneficial only for very large objects.

elizarov on Dec 19, 2017