Run 42274262300 (b5cc94b1b2) failed agc_locks ("users got through between
replacements") and guest_job_manager_worker_spin_scale_0.1 (upper CPU bound).
Neither is touched by #14407, #14373 or #14600: both are scheduling-sensitive
tests (3/189 and 1/189 earlier failures on main).
agc_locks: the 600 replacements could finish before any reader thread was
scheduled; replacers now start after the first reader use.
worker spin: parked and spinning samples take the quietest of three runs, as
the no-pending sample already did.
Co-Authored-By: Claude <[email protected]>
Culprit: none by code; run 40167585870 (main 2400596055) failed guest_job_manager_worker_spin
(check at line 688: noPending 27 ms vs parked 10.6 ms + interval/3 8.5 ms). The sample
only adds CPU when the runner is noisy, so the minimum of 3 is the right estimator.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
MinGW-w64's UCRT has no aligned_alloc, so the Windows build of
guest_job_manager_tests fails since #3170. Use the aligned operator new
and delete instead, which are standard C++17 on every host.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
