YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context
Rust agentsagiaiai-memoryaudio-recordingcomputer-visionhermeshermes-agentllmlocal-ailocal-firstmachine-learningmcpmultimodalopenclawprivacyrewindscreen-recordingspeech-to-textycombinator
fix(hd): keep meeting HD recording bound to the whole call, not 30 min (#4583)
Louis Beaumont1fa20e4fdccc
diff --git a/crates/screenpipe-engine/src/high_fps_controller.rs b/crates/screenpipe-engine/src/high_fps_controller.rsindex 54d9e39a8..e5fc06149 100644--- a/crates/screenpipe-engine/src/high_fps_controller.rs+++ b/crates/screenpipe-engine/src/high_fps_controller.rs@@ -275,11 +275,19 @@ impl HighFpsController {/// that will never come.pub fn start_meeting_session(&self, meeting_id: i64) -> HighFpsSnapshot {let now = Instant::now();- let in_meeting = self- .detector- .as_ref()- .map(|d| d.is_in_meeting())- .unwrap_or(false);+ // Use the same meeting signal as `snapshot()`: the desktop app builds+ // this controller with NO detector handle and drives meeting state+ // through `meeting_started`/`meeting_ended` events (`event_in_meeting`,+ // set via `set_in_meeting`). Checking only `self.detector` here meant+ // the app always saw "not in a meeting" and downgraded every+ // meeting-triggered session to the bounded 30-min `STALE_MEETING_FALLBACK`+ // timer — so HD stopped 30 min into any longer call instead of staying+ // bound to the meeting. Fall back to the event flag when no detector is+ // wired so a fresh meeting gets the proper meeting-bound session.+ let in_meeting = match self.detector.as_ref() {+ Some(d) => d.is_in_meeting(),+ None => self.event_in_meeting.load(Ordering::Relaxed),+ };let (kind, expires_at, log_reason) = if in_meeting {(SessionKind::Meeting { meeting_id },@@ -691,6 +699,45 @@ mod tests {assert!(remaining >= MAX_MEETING_DURATION.as_secs() - 5);}+ #[test]+ fn app_meeting_session_is_meeting_bound_via_event_flag() {+ // Regression: the desktop app builds the controller with NO detector+ // and signals meetings through set_in_meeting (event_in_meeting). The+ // meeting_started handler calls set_in_meeting(true) and THEN+ // start_meeting_session(id). Before the fix, start_meeting_session only+ // consulted self.detector (None → false), so it downgraded every+ // meeting to the 30-min STALE_MEETING_FALLBACK timer and HD stopped+ // 30 min into any longer call. With the event-flag fallback, a fresh+ // meeting must be meeting-bound with the ~4h cap.+ let c = Arc::new(HighFpsController::new(None, DefaultMode::Always, 100));+ c.set_in_meeting(true);+ c.start_meeting_session(293);+ let s = c.snapshot();+ assert!(matches!(+ s.kind,+ Some(SessionKind::Meeting { meeting_id: 293 })+ ));+ let remaining = s.remaining_secs.unwrap();+ assert!(remaining >= MAX_MEETING_DURATION.as_secs() - 5);+ // And it ends when the matching meeting ends (not only on the cap).+ c.handle_meeting_ended(293);+ assert!(!c.snapshot().active);+ }++ #[test]+ fn app_stale_click_without_meeting_still_falls_back_to_timer() {+ // Counter-test: no detector AND no active meeting (event flag false) is+ // a genuine stale click — must still downgrade to the bounded timer so+ // it can't sit waiting for a meeting_ended that never comes.+ let c = Arc::new(HighFpsController::new(None, DefaultMode::Always, 100));+ c.start_meeting_session(42);+ let s = c.snapshot();+ assert!(matches!(s.kind, Some(SessionKind::Timer)));+ let remaining = s.remaining_secs.unwrap();+ assert!(remaining <= STALE_MEETING_FALLBACK.as_secs());+ assert!(remaining >= STALE_MEETING_FALLBACK.as_secs() - 5);+ }+#[test]fn prewarm_pending_upgrades_to_meeting_preserving_started_at() {let c = ctl();