parse_messages() previously advanced past a 0xFE start byte whenever no
complete frame could be matched, silently consuming partial frames at the
end of each ser.read(64) call as garbage. When the device sends 20-50
rapid knob messages during a fast turn and one message straddles the
64-byte read boundary, the split message was discarded: the daemon skipped
a knob-position update, the LED color lagged, and when the next complete
message arrived the color jumped — producing a visible flicker.
Fix: break out of the parse loop when a 0xFE is followed by a known type
byte (0x02/0x03/0x06/0x07) but the buffer does not yet hold a complete
frame. The partial bytes are returned in the remainder and prepended to
the next ser.read() chunk, so no message is ever lost.
Also sync last_led_colors after the initial send_leds() on connect so the
dedup guard is accurate from the first knob event instead of from the
first heartbeat.
Adds 5 tests covering partial heartbeat, button, and knob frames, the
split-read round-trip, and unknown frame type skipping (61 pass total).
The heartbeat path called send_leds but never updated last_led_colors,
so the dedup guard in handle_knob had no visibility into what was last
actually sent. Heartbeats arriving mid-knob-turn injected extra serial
writes that bypassed the guard and overwhelmed the device's LED renderer,
causing visible flicker.
Two targeted fixes:
1. Deduplicate LED writes in handle_knob: compute new colours and only
call send_leds() when they differ from last_led_colors. A single
physical knob turn produces 20-50 ADC samples in quick succession;
without this guard each sample triggers a 47-byte write and the
firmware can't keep up, causing visible flicker.
2. Gate reapply_app_volumes on a 200 ms knob-quiet period. Calling
playerctl/pulsectl while the user is actively turning a knob stalls
the main loop, backing up serial data, missing heartbeats, and
causing the device LEDs to time out mid-turn.
Both state variables (last_led_colors, last_knob_event) are mutable
lists initialised in main() and threaded through handle_knob, keeping
module-level globals out of the picture.
Adds 6 new unit tests (56 total, all passing).
- Add MPRISController class: wraps playerctl with 3s player-list cache,
fuzzy app-name matching, and thread-safe get/set_volume via MPRIS2 D-Bus
- Add PulseController event watcher thread: separate pulsectl connection
listens for sink-input events and queues them so reapply fires within
~0.1s for PA-only apps instead of waiting up to 1s
- Route set_app_volume and get_app_volume_norm through MPRIS first,
falling back to PA stream for non-MPRIS apps (e.g. Brave/Chromium)
- Update reapply_app_volumes to prefer MPRIS path, preventing turnupd
from fighting Spotify on the PA layer during song transitions
- Wire MPRISController and start_watching() into main(); drain_events()
triggers immediate reapply on any PA sink-input event
- Add per-knob RGB LED configuration, validation, and color interpolation
- Add build_led_packet, send_leds, all_led_colors to turnupd
- Update config with VALID_LED_MODES, _validate_color, _validate_leds
- Update contrib/config.example.json with LED configuration examples
- Remove legacy root-level config.py and turnupd.py (moved to src/)
Sean Doran