Addresses action item 3 from #118: gate the ESPHome motion publish behind a successful Security+2.0 sync, rather than binding the gdolib callback straight to publish_state.

Why

Motion is a Security+2.0-only feature — the Security+1.0 protocol has no motion command, and decode_v1_packet() has no motion path. But gdolib's protocol probe can transiently run the Security+2.0 decoder against a Security+1.0 bus (gdo_sync_task() falls back to V2 emulation at gdo.c:911-929 when the V1 handshake doesn't resolve in ~7.5 s), and a corrupt frame read in that window can decode as GDO_CMD_MOTION. That is how a Security+1.0 opener with no motion sensor ended up publishing motion. Full trace in #118.

Downstream, a single publish is unrecoverable: the cloud infers a motion sensor exists the moment the entity leaves its unknown state, so one spurious blip permanently marks a sensor that isn't there — and it self-clears after 3 s, so it looks perfectly innocuous in the web UI.

Before this change there was no gate anywhere between a decoded 0x285 and an MQTT publish: binary_sensor/__init__.py binds the parent callback directly to GDOBinarySensor::publish_state, and GDOBinarySensor is an empty class.

What

• Track synced_ and protocol_ on GDOComponent, populated from the existing GDO_CB_EVENT_SYNCED handler.
• set_motion_state() moves out of the header and drops the update (with a warning log) unless we are synced and on Security+2.0.

Behavior by case:

Ordering is safe: in gdo_event_handler, set_protocol_state() is called before set_sync_state() and only when the sync actually succeeded, so protocol_ is never populated from a failed probe.

Scope

This is defense in depth, not the root-cause fix. The parser-side fix belongs in gdolib — filed as konnected-io/gdolib#30 (update_motion_state() has no protocol guard). This PR only ensures the ESPHome layer won't forward a motion state it has no basis to believe. Note it does not address spurious motion on a genuine Security+2.0 opener with no motion sensor — on such a device the gate is open by construction, and that case depends on the gdolib parser hardening.

Only motion is gated. Obstruction, button, motor, light, lock and door all exist in the Security+1.0 protocol and are left alone.

Testing

Compiles clean for garage-door-GDOv2-Q.yaml (ESP32-S3) against the local component tree.

Not yet exercised on hardware — the interesting path needs a Security+1.0 opener whose V1 handshake fails into the V2 probe window, which I can't stage here. The confirming signal on a real device is the new Ignoring motion state ...; not synced on Security+2.0 warning appearing where a Motion: detected would previously have published, correlated with gdolib's secplus V1 panel emulation failed, trying secplus V2 panel emulation.

🤖 Generated with Claude Code