On Wed, 18 Mar 2026 15:44:24 GMT, Kevin Walls <[email protected]> wrote:
>> I removed unused return value `replaced`.
>>
>> The simplified version is not equivalent - e.g. when `10 <= count <= 20`,
>> the behavior depends on early `replacing` value.
>
> I tested this version and it still causes the crash, so we can make the
> change if you like.
> As long as replaceThreads() gets called we cause trouble.
> Actually this loop can be simpler:
>
> 139 while (true) {
> 140 replaceThreads(ids, infos);
> 141 goSleep(1);
> 142 }
>
>
> My earlier experiment used hundreds of threads, finishing at different times,
> and would stop replacing threads if there were many alive. But we only need
> a few threads active to cause the problem quite reliably. So it can just
> continually keep trying to replace threads, and it is enough to cause the
> crash.
>
> The sleep was to yield and make sure other threads got a chance to crash us.
> It it still required.
(so replaceThreads does not need to return anything, more simplification).
Apologies the file attached to the bug was not polished. 8-)
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/30105#discussion_r2954421332
