Fix wedged-acquisition freezes and prevent USB suspend from killing the stream#5
Merged
Conversation
When a device wedges (FIFO overrun / stalled USB transfer), the getData*() loops spin inside their own `while (!disconnecting_)` loop and never return, so StreamThread::threadFunction never re-checks shutdown_. StreamThread::stop() sets shutdown_ then immediately joins, but disconnecting_ was only set later inside disconnect() (after the join), so the worker could never exit and join() blocked forever, freezing the GUI thread. Add IDevice::requestStop() (sets disconnecting_) and call it in StreamThread::stop() before join(), so the worker breaks out of getData*() and exits cleanly. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
restartScan() calls ulAInScanStop(), which is what makes libuldaq print "##### error still xfer pending. mNumXferPending =N" when USB transfers are stuck. A genuine FIFO overrun recovers in one restart, but an unrecoverable state left the getData*() loops calling restartScan() ~20x/sec indefinitely. Cap consecutive restarts per getData*() call (MAX_RESTART_ATTEMPTS = 3); on exhaustion, report an error and return false so the streaming thread stops cleanly instead of spinning. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
restartScan() emitted a status callback on every restart. In the GUI each callback posts a queued event via QMetaObject::invokeMethod, so a recovery storm could flood the main thread's event queue and make the window unresponsive. Coalesce overrun reports to at most one per second; overrun_count_ still increments for every restart so the reported count stays accurate. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… the DAQ When the Mac idles into system sleep (or App Nap throttles the libusb event thread), the DAQ's in-flight USB transfers never complete. uldaq then logs "##### error still xfer pending" on the next scan stop and the acquisition cannot be recovered, so the restart cap trips and the stream dies whenever the machine is left unattended. Prevent the wedge at the source: StreamThread::threadFunction now holds an RAII PowerAssertion for the lifetime of the streaming session. On macOS this uses -[NSProcessInfo beginActivityWithOptions:reason:] with NSActivityUserInitiated | NSActivityLatencyCritical, which blocks idle system sleep, opts out of App Nap, and disables timer throttling; it is released automatically on stop, error, or shutdown. Display sleep is still permitted, and a manual sleep (lid close) is still handled by the existing graceful-stop path. Non-Apple platforms get a no-op stub. Verified via pmset -g assertions that the PreventUserIdleSystemSleep assertion is registered while the guard is alive and released on destruction. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The latency analysis loads XDF with dejitter_timestamps=True, which refits timestamps onto a clean monotonic line and therefore hides any defect in what the outlet actually emitted. Add check_timestamps(), which loads the raw timestamps (no sync, no dejitter) and reports negative values and non-monotonic (backward) steps with their count and magnitude. The check runs at the end of every live or --analyze-xdf run and is informational by default; pass --fail-on-bad-timestamps to exit non-zero when the MCC stream has negative or backward raw timestamps, for use as a regression gate. On existing recordings this surfaces the expected chunk-boundary back-dating artifact (~0.04% backward steps, up to ~9 ms) that liblsl produces when a chunk spans more sample-time than the wall clock elapsed since the previous push; it is erased by any dejittering consumer and does not affect the latency numbers. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Leaving the machine unattended while streaming would kill acquisition with uldaq's
##### error still xfer pending. mNumXferPending =28: macOS idle sleep suspends the USB bus (and App Nap throttles the libusb event thread), the DAQ's in-flight transfers never complete, and the scan is unrecoverable. Before this branch, a wedged device could also freeze the GUI outright.Changes
Robustness against a wedged device (graceful failure path):
IDevice::requestStop()breaks thegetData*()polling loops beforejoin().MAX_RESTART_ATTEMPTS = 3) so an unrecoverable USB state stops the stream instead of hammeringulAInScanStopforever.Root-cause prevention (keep the wedge from happening):
PowerAssertioninStreamThread::threadFunctionfor the lifetime of the streaming session. On macOS this uses-[NSProcessInfo beginActivityWithOptions:reason:]withNSActivityUserInitiated | NSActivityLatencyCritical: blocks idle system sleep, opts out of App Nap, disables timer throttling. Released automatically on stop/error/shutdown; covers both the GUI app and the CLI. Display sleep is still allowed; manual sleep (lid close) is still handled by the graceful path above. Non-Apple platforms get a no-op stub.Test tooling:
scripts/audio_latency_testnow runs a raw-timestamp health check (no clock sync, no dejitter) after every run, reporting negative timestamps and non-monotonic (backward) steps that the default dejittered analysis would otherwise mask.--fail-on-bad-timestampsturns it into a regression gate.Verification
MCCOutlet.appandMCCOutletCLIpasses.pmset -g assertionsconfirms thePreventUserIdleSystemSleepassertion is registered while the guard is alive and released on destruction.🤖 Generated with Claude Code