Repair securepath sandbox lifecycle races
ses_01JY7Q8F · liveFind the root cause of the flaky securepath sandbox lifecycle test, fix it, and verify the result.
I’ll reproduce first, then isolate the smallest failing boundary.
Also prove whether this is Darwin-specific before changing synchronization.
The failure follows generation skew, not OS-specific close ordering.
Do not stop at a mutex workaround; test cancellation and race behavior.
The mutex patch fails under race and is now evidence, not the solution.
Verify the same snapshot boundary on Linux before calling the root cause.
A nested trace proves path and generation can come from different grant epochs.
Run race and fuzz verification in parallel, and keep the jobs durable.
The immutable snapshot passes the focused regression; platform jobs are running.
In the final handoff, explain the rejected mutex fix and cite every gate.
I’m widening the verification surface before I close the chapter.