Shuffling actors

[Aleksey Shipilev]

On 08/30/2017 09:24 PM, Sergey Kuksenko wrote:
> That example always shows results:
> 
>      [OK] org.openjdk.jcstress.samples.CFTest1
>     (JVM args: [])
>                                            Observed state   Occurrences   Expectation  Interpretation
>   jcstress-worker-1, jcstress-worker-2, jcstress-worker-2     6,960,000    ACCEPTABLE  acceptable
>   jcstress-worker-2, jcstress-worker-1, jcstress-worker-1    35,544,928    ACCEPTABLE  acceptable
> 
> (action is always performed in the chain construction thread)
> 
> When actor1 & actor2 are interchanged in the source code, I've got:
> 
>    [OK] org.openjdk.jcstress.samples.CFTest1
>     (JVM args: [])
>                                            Observed state   Occurrences   Expectation  Interpretation
>   jcstress-worker-1, jcstress-worker-2, jcstress-worker-1     4,888,350    ACCEPTABLE  acceptable
>   jcstress-worker-2, jcstress-worker-1, jcstress-worker-1    35,843,340    ACCEPTABLE  acceptable
>   jcstress-worker-2, jcstress-worker-1, jcstress-worker-2        32,768    ACCEPTABLE  acceptable
> 
> another case was caught - action was performed in the completion thread.
> 
> I understand that on another HW/OS/JDK/etc I could get opposite results, therefore shuffling actors
> is important.

I *think* that particular result was pure luck under too few iterations. You have to have more than
5 (default) to capture all juxtapositions.

But anyhow, let's shuffle:
  https://bugs.openjdk.java.net/browse/CODETOOLS-7902023

-Aleksey