fix: guard relay_switch parsers against short payloads (#369)

[bluetoothbot]

What

Filter single-byte / short responses in _get_basic_info and add length guards in relay_switch._parse_common_data / _parse_user_data so a stray short payload no longer crashes update_from_advertisement.

Why

Reported in #369: a Switch 1PM receives a single-byte b"\x02" from _send_command(...). The original guard only rejected b"\x07" and b"\x00", so the byte reached _parse_common_data which indexes raw_data[1], raw_data[2], raw_data[16] and raised IndexError: bytearray index out of range. The traceback escapes via the create_background_task(self.update()) call in SwitchbotSequenceDevice.update_from_advertisement, surfacing in Home Assistant as an unretrieved task exception.

How

• device._get_basic_info now rejects any response with len(_data) <= 1, covering all observed error replies (\x00, \x02, \x07, …) and logging the offending hex for diagnosis.
• relay_switch._parse_common_data and _parse_user_data short-circuit to None when the payload is shorter than the minimum slice they read (17 and 15 bytes respectively).
• get_basic_info (1PM and 2PM paths) returns None if either parser cannot interpret the payload, so callers like _get_current_image_index get the same "no info" signal they already handle.

Testing

• New parametrised test exercises \x02, \x07, \x00, empty, and a 3-byte payload — all now return None instead of raising.
• New 2PM test covers the case where only the channel-2 response is short.
• Adjusted four test_art_frame tests that were inadvertently depending on AsyncMock slipping past the old guard; they now mock _get_current_image_index explicitly.
• Full suite: 1084 passed.

Closes #369

🤖 Generated with Claude Code

---

Quality Report

Changes: 4 files changed, 82 insertions(+), 4 deletions(-)

Code scan: clean

Tests: failed ([Errno 13] Permission denied: 'pytest')

Branch hygiene: clean

Generated by Kōan post-mission quality pipeline

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
switchbot/devices/device.py	69.90% <100.00%> (ø)
switchbot/devices/relay_switch.py	100.00% <100.00%> (ø)

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[andrew-dash]

Confirming this on Home Assistant 2026.5.2 with two Switch 1PM relays on a BLE-proxied setup (3× ESPHome ESP32 bluetooth_proxy). On 2026-05-16 the unretrieved task exception from this exact path stacked up and combined with an unrelated ESPHome reconnect timeout to wedge HA Core into a supervisor-watchdog restart cycle — full outage of ~40 minutes until manual diagnosis.
The relays send a 1-byte response (b"\x04" in our case, similar to the b"\x02" reported in #369) intermittently — looks signal-quality-related (RSSI ~-54 dBm). I ported your patch as a local custom_component monkey-patch (length guards on _parse_common_data / _parse_user_data, plus a defensive try/except around get_basic_info to handle the user_data.pop() on None that bites when only the parsers return None). Confirmed it eliminates both the crash and the supervisor-watchdog cascade.
Would be great to get this out of draft and merged — issue #369 has been open since July 2025 and this is a real-world stability hit for users with multiple SwitchBot relays.

[bluetoothbot]

Thanks @andrew-dash for the production confirmation — that's exactly the failure mode this targets. Worth noting: the current PR already covers the user_data.pop() on None case you patched locally (line 168 in get_basic_info: if common_data is None or user_data is None: return None), so the parsers returning None is properly handled at the get_basic_info boundary. If you had an earlier iteration of the patch without that guard, the merged version should be a strict superset.