(cIndex - pIndex) / 2 == 0 return null；If it is being expanded, wait a little while for the expansion to succeed, there will be a JUMP, and should not return null about jctools HOT 5 CLOSED

@nitsanw Is that okay? see OOM can cause linked array qs to break #355

from jctools.

Please tell me the version of JCTools you are using, I can’t find the relevant code in the latest 4.0.1.

from jctools.

@nitsanw
Currently only modify BaseMpscLinkedArrayQueue.java poll() and peek, several other classes BaseMpscLinkedAtomicArrayQueue, BaseMpscLinkedUnpaddedArrayQueue have not been modified, if my solution is correct I'll then update the others as well. I also submitted a PR. #381

from jctools.

In this PR #371, the author also wants to get data as much as possible. But @franz1981 thinks that we should follow the principle of "fast failure", instead of using spin to consume CPU. Therefore, I think your suggestion may not be accepted.

from jctools.

Thanks for reporting, repeating my comment from the PR:

Your error is here:

cIndex-pIndex==0 , Empty

No, (cIndex-pIndex/2==0) is the empty condition. That's why we have that as the entry condition for emptiness with the comment // isEmpty? and why isEmpty() is implemented the way it is.

The edge case being decided here is that if the queue is resizing when empty, it is still empty.

More reasoning:
This is an avenue we have intentionally decided not to go down. It is consistent between poll/peek/size/isEmpty/iterator etc. And yes, people with chunk size 2 can hit this frequently, but they are also making a configuration error. These queues don't really make much sense with chunk size less than cache line (8 or 16 elements, depending on oop size).

from jctools.