SukramJ/homematicip_local 2.10.0b0

on GitHub

What's Changed

Integration

• Fix: on Home Assistant 2026.9 no entity was added any more — a startup produced hundreds of Error adding entity None for domain … with RuntimeError: Detected code that calls device_registry.async_get_or_create with a deprecated via_device parameter (aiohomematic#3365). The deprecation notice promises a warning until 2027.8, but that promise only holds while HA can find the integration in the call stack: the via_device of a DeviceInfo is passed on by HA's own entity_platform, so no integration frame is left, and HA then falls back to its core behaviour — which raises. Devices are now linked by registry id: ControlUnit.ensure_via_device_exists() returns the id of the via device it made sure exists, and the central is created on demand so a link never points at an unregistered device
• Breaking: the minimum Home Assistant version is now 2026.8.0. The device registry APIs this release moves to (via_device_id on DeviceInfo and async_get_or_create, plus async_get_device_by_identifier) do not exist in 2026.7
• Fix: devices were no longer removed from the device registry when their last entity disappeared. The removal path tested device_id in device_registry.devices, and on HA 2026.9 devices iterates device entries instead of being a mapping, so the membership test silently never matched. It now asks the registry directly
• Replace device_registry.async_get_device(identifiers=…) with async_get_device_by_identifier(identifier, config_entry_id). Identifiers are no longer unique across config entries, so the lookup is now scoped to this integration's own entry — which is what all three call sites (via-device creation, device-action/-trigger resolution, and the known-device check after a cache clear) meant all along
• Side effect of the id-based links, measured against the full test device set: devices that so far ended up with no via device at all — the virtual remotes among them — now hang off the central like every other device (7 of 439 devices in that run; entity count unchanged at 4705)
• Garage doors (HmIP-MOD-HO, HmIP-MOD-TM) gain a Door Mode select entity that exposes the door's three physical states — closed, open and ventilation position — as a first-class, tappable control. Home Assistant's cover platform has no native ventilation state (architecture#502), so ventilation was previously only reachable by setting a cover position inside an undiscoverable range. The data point itself comes with the aiohomematic bump below; the integration contributes the entity name and its state translations. Suggested in #1252 by @ANierbeck

Config Panel

• The climate schedule editor no longer discards unsaved edits when a second weekday is opened (frontend#95). Each weekday now keeps its own draft, including its undo history, so several days can be edited in one dialog session; changed tabs carry a dot and the footer button becomes "Save all". Closing the dialog with unsaved changes asks first, and the inline row button is labelled "Apply" / "Übernehmen" so that "Save" means only the action that writes to the CCU. This affects the config panel and the climate schedule card alike, since both render the same schedule editor

Development

• Beyond the entries above, the integration's own changes in this release are openccu-loom backend (Beta) work; the direct-CCU backend gains support for two new devices through the aiohomematic bump below. The user-facing details of the loom backend stay out of scope for this changelog until it leaves Beta

Dependencies

Bump aiohomematic to 2026.8.5

• Fix: HM-CC-TC climate entities stayed frozen at their restored state. CustomDpSimpleRfThermostat gated its validity on TEMPERATURE and SETPOINT, but HM-CC-TC reports SETPOINT only when the wheel is turned on the device — never periodically. On a device nobody had touched since the last CCU restart, SETPOINT therefore stayed unrefreshed forever, which dragged the whole climate data point to is_valid=False: the entity kept value_state=restored and froze current_temperature, target_temperature and hvac_mode, while the sibling sensor.* entities kept updating from the very same TEMPERATURE events. Validity now follows TEMPERATURE alone, and target_temperature stays None until a real SETPOINT arrives instead of reporting a stale value
• Fix: classic RF thermostats hung at the restored state while the operating mode stayed untouched. Same failure class for CustomDpRfThermostat (HM-CC-RT-DN, HM-TC-IT-WM-W-EU, HM-CC-VG-1, BC-RT-TRX-*, BC-TC-C-WM), which gated its validity on CONTROL_MODE — a parameter these devices only send when the mode actually changes. After a CCU restart it could stay unrefreshed for as long as nobody switched between AUTO and MANU. Validity now follows TEMPERATURE and SETPOINT; until CONTROL_MODE arrives, mode reports its existing AUTO fallback
• Expose the garage door's discrete mode as a SELECT-category combined data point. CustomDpGarage now declares a CombinedDpGarageDoorMode that reads DOOR_STATE and writes DOOR_COMMAND under its own parameter name DOOR_MODE, so the integration's generic select dispatch picks it up without any platform-side wiring — and the entity is named after what it does rather than borrowing the identity of either source parameter. While the door is travelling the select keeps reporting the mode the door is heading for instead of dropping to unknown; a STOP clears that held mode
• Fix GarageDoorState.POSITION_UNKNOWN, which carried a stray leading underscore and therefore never matched the DOOR_STATE value reported by the CCU
• Add support for the flush-mount switch actuators HmIP-FS6 and HmIP-FSI6. HmIP-FS6 has no input channel, so it shares the channel layout of HmIP-FSM and is now registered as a switch profile on channel 2 — without that registration it only produced generic data points. OPERATING_VOLTAGE is ignored for it as well, consistent with the other mains-powered actuators
• Expose CHANNEL_OPERATION_MODE on channel 1 of HmIP-FSI6. The custom switch data point of that device already worked, but the un-ignore rule was keyed on HmIP-FSI16 only — model keys are matched with str.startswith, and "hmip-fsi6".startswith("hmip-fsi16") is False, so the operation mode of the push-button input stayed hidden

Bump openccu-loom-client to 2026.8.26 (pins openccu-loom-types==0.5.5)

• Bump for the openccu-loom backend (Beta); it has no runtime effect on the direct-CCU backend, where the client is not loaded. Advances the bundled loom client from 2026.8.9 to 2026.8.26 and its transitively pinned openccu-loom-types from 0.3.10 to 0.5.5
• This raises the minimum daemon to openccu-loom 0.65.1 or newer. The client is generated against the daemon's API 7.13.0 and checks that version at connect time, so it refuses a daemon reporting an older API rather than half-initializing against an incompatible one — an older daemon is rejected outright rather than merely missing newer surface. Installations that cannot update the daemon should stay on 2.9.1