On Thu, 1 Oct 2026 04:31:24 GMT, David Holmes <[email protected]> wrote:
> This seems reasonable based on your description of the situation. I do find > it strange that the problem seems so much more reproducible on these > particular OS versions. The ultimate fix seems more of a workaround for a > still unknown OS problem. Yes, I feel the same. Sadly no one knows this code well. What I know about the process handling I've learned the past few days working on this fix. It's a mix of BSD, mach, and ptrace, so it's no surprise that there can be some sync up issues between the 3. > I'm curious whether you tested the 4th fix on its own. It seems with the > reply after detach the reply would get stuck because of the process still > being suspended. IF you move the reply location then you naturally avoid that > possibility. I tried so many combinations of fixes (there are more that I didn't mention), that I can't recall for sure if I tried just the SIGCONT fix without anything else. I think I always had the relocated reply in place since it was one of the first changes and made a big difference. ------------- PR Comment: https://git.openjdk.org/jdk/pull/33158#issuecomment-5925577336
