Fix quirks v2 entity non-values

[TheJulianJES]

Proposed change

Move super().__init__() in the Sensor class, so _attribute_name is set by the base class before we set _attr_def based on that in the Sensor class.

TODO:

• Maybe clean up comments a bit
• Review test / switch to using Frient sensor device from diagnostics
• Add test coverage for the now missed line of if attr_def is None in _is_non_value
• Check if other classes/entities are also affected

---

AI summary

ZCL numeric measurement attributes use a sentinel value to mean "no measurement available" (e.g. 0xFFFF for uint16, -32768 for int16). Sensor._is_non_value is supposed to map those to None so the entity reports as Unknown in HA. For specialized sensor classes (Temperature, Humidity, etc.) this works, but for quirks v2 entities — which use the generic sensor.Sensor class — the sentinel was leaking through as the raw integer state.

Reported via diagnostics on a frient AQSZB-110: the device's humidity sensor (standard RelativeHumidity cluster, specialized Humidity class) correctly reported None when the measurement was unavailable, but its VOC sensor (custom develco_voc_level cluster 0xfc03, generic Sensor class via quirks v2) reported 65535 instead of None.

Root cause

Sensor.__init__ looked up the attribute definition before calling super().__init__():

if self._attribute_name is not None:
    self._attr_def = self._cluster_handler.cluster.find_attribute(self._attribute_name)

super().__init__(cluster_handlers, endpoint, device, **kwargs)

For specialized subclasses, _attribute_name = "measured_value" is a class attribute, so this works. For quirks v2 entities, _attribute_name is set inside _init_from_quirks_metadata, which runs from super().__init__() — i.e. after the lookup. So _attr_def stayed None, and _is_non_value short-circuited to False (no attr def → can't know the type's non_value), letting the sentinel through.

Fix

Move the find_attribute lookup after super().__init__() so _attribute_name is resolved (whether from a class attribute or from quirks v2 metadata) before we use it.

Tests

• test_ignore_non_value_quirks_v2 — builds a quirks v2 humidity sensor via QuirkBuilder, sends 0xFFFF as the measured value, asserts the entity state is None. Confirmed to fail without the fix.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.63%. Comparing base (f443146) to head (99efd3f).

Additional details and impacted files

@@            Coverage Diff             @@
##              dev     #758      +/-   ##
==========================================
- Coverage   97.64%   97.63%   -0.01%     
==========================================
  Files          62       62              
  Lines       10873    10873              
==========================================
- Hits        10617    10616       -1     
- Misses        256      257       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[puddly]

We don't rely on self._attr_def anywhere within __init__, right?

[TheJulianJES]

Yeah, it's only assigned there and then used exclusively in _is_non_value. Are you suggesting to always re-find the attribute definition in _is_non_value instead? (and it's not used in the base class at all.)

(Side note: I think at some point in the future, we may want to (and be able to?) provide ZCL entities directly with the attribute definition, instead of a string?)

[puddly]

No, just checking. I remember there being some reason for __init__ not being at the very top of the overriden __init__ but I think we're refactored that reason away at some point in the past.