If CLOSE_EVT never arrives after DISCONNECT_EVT, the client gets
stuck in DISCONNECTING forever, blocking reconnection and scanner
restart. Add a 10s timeout watchdog in loop() that forces IDLE
as a recovery path.
Introduce set_disconnecting_() helper to ensure the timeout
timestamp is always set when entering DISCONNECTING state.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Separate the DISCONNECTING state into its own log message that
mentions waiting for CLOSE_EVT, making it easier to diagnose
stuck connections vs normal in-progress connections.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The CLOSE_EVT handler matches on conn_id_ to call set_idle_().
Resetting conn_id_ to UNSET_CONN_ID before CLOSE_EVT arrives
causes the event to be dropped, leaving the client stuck in
DISCONNECTING state permanently.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When OPEN_EVT arrives with a failure status, the connection was never
established so no CLOSE_EVT may follow. Reset conn_id_ to prevent a
stale value from matching a future CLOSE_EVT.
Instead of a separate early-return for DISCONNECTING state, let the
failed OPEN_EVT fall through to the existing failed-status handler
which already transitions to IDLE.
When a connection fails to establish, the ESP-IDF stack sends
DISCONNECT_EVT followed by OPEN_EVT with a failure status (e.g. 133).
No CLOSE_EVT follows since no GATT connection was established.
Previously the OPEN_EVT in DISCONNECTING state was silently ignored,
leaving the client stuck in DISCONNECTING forever. Now we treat this
failed OPEN_EVT as the terminal event and transition to IDLE.
ESP_GATTC_DISCONNECT_EVT only indicates that the link-layer disconnection
has started, not that the controller has fully freed resources (L2CAP
channels, ATT resources, HCI connection handle). Transitioning to IDLE
immediately allowed reconnection attempts before cleanup was complete,
causing the controller to reject new connections with status=133
(ESP_GATT_CONN_FAIL_ESTABLISH) or crash with ASSERT_PARAM in lld_evt.c.
This applies the same fix that was made for bluetooth_proxy in #10303
to the base BLEClientBase class, which is used by the ble_client
component.
Changes:
- DISCONNECT_EVT now transitions to DISCONNECTING instead of IDLE
- CLOSE_EVT (already existing) handles the final IDLE transition
- connect() now blocks while in DISCONNECTING state
- OPEN_EVT ignores events received during DISCONNECTING state
J. Nick Koston