Zephyr wifi stop_station returns without disconnecting (station lifecycle stubs) #40

On a Pico W using the Zephyr port, wifi.radio.stop_station() returns without disconnecting an associated station. Both start_station() and stop_station() currently contain only commented-out set_mode_station() calls.

This was physically observed on CircuitPython 11.0.0-alpha.2-30-g7d4ef1aac7 during the Env Hub firmware refresh. With no managed workload running and no AP or BLE connection active, the state was:

BEFORE: wifi_connected=true, ap_active=false, ble_connected=false, advertising=false
wifi.radio.stop_station()
AFTER:  wifi_connected=true, ap_active=false, ble_connected=false, advertising=false

The same function bodies remained in the preceding integration head 6b117bdce576abf927554a4946b8610149150aa0. Upstream main was checked on 23 October 2026 and also has the stubs, in ports/zephyr-cp/common-hal/wifi/Radio.c (Git blob 1570d57d63a7855081d8097685487de1461d165c). This is distinct from the socket-service DHCP scheduling fix in #32.

Minimal reproduction on an available test AP:

import wifi
wifi.radio.connect("test-ap", "test-password", timeout=20)
assert wifi.radio.connected
wifi.radio.stop_station()
print(wifi.radio.connected)  # observed True; the method did nothing

Expected: stopping the station ends its station connection, with state and events reflecting the result. A subsequent start_station() should restore station operation. Shared station/AP interface behaviour must be handled deliberately rather than blindly taking down an active AP.

An explicit wifi.radio.enabled = False made the connected check false in a separate physical control. It disables the interface globally and is not a substitute implementation of station lifecycle semantics. The connection was present again in a later console session; that re-enable path has not been fully isolated. The firmware updater now quiesces the radio in the same session as bootloader entry. Both refused update attempts occurred before any flash read/write and were retained.

Update: candidate fix 9e153ad624bcd1559211dfad8eb54ef8841f15e9 is in draft PR #32. It implements start/stop, waits for disconnect completion, preserves state on failure and retains a shared-interface AP. The actual-source harness passes 46 checks; preceding 6b117 fails 38. Physical Pico W now passes three connect/stop/reconnect cycles, repeated stop, disabled-start, externally observed departure and AP-preservation/ping, with successful cleanup. Both Pico images build; Pico 2 rollout and physical failure injection remain open. Keep this issue open pending upstream disposition.

Reproduction, evidence and limits. Two earlier fixture failures are preserved separately; they were not relabelled as passes.