[ikalchev]

Stacked on #561 (APIs: #562). The registration side of the delegate interface — delegates can now be assigned and validated, still without being dispatched.

What

• Characteristic.delegate / Service.delegate properties — assign a CharacteristicDelegate / ServiceDelegate instance, or its class (instantiated with no arguments). None or del removes it.
• Assignment is when everything is decided ("handled in init only", as the interface promises): the overridden methods are recorded, and an overridden method that is not a coroutine function fails at the assignment with a TypeError naming the method — no sync/async detection later, ever.
• Service.configure_char(..., delegate=...) wires a delegate while configuring the characteristic, alongside the legacy callbacks.

class MyLock(CharacteristicDelegate):
    async def on_set(self, value, context):
        ...

char = service.configure_char("LockTargetState", delegate=MyLock)  # class or instance

• The legacy attributes become storage properties: setter_callback / getter_callback on Characteristic and Service keep returning the assigned callable unchanged, so the driver, the camera stack and tests that patch.object them see no difference — including the delattr teardown of patch.object, which is why the properties implement deleters. One instance may serve many characteristics; the context says which one a call is about.

Not in this PR (deliberately): dispatch. The write and read pipelines call delegates starting with the next PR in the stack (the gate pipeline); until then a delegate is wired, validated, but not called. Which of delegate vs. legacy callback wins is also decided there. CHANGELOG stays untouched per review.

Tests

+14 tests: instance/class/removal/replacement round-trips, wrong-type and sync-override TypeErrors with the right messages, hook recording, legacy/delegate coexistence, legacy removal semantics, configure_char wiring at both the instance and class level, and the service-level equivalents. Three existing tests reach assert_called_once_with through the old plain attributes; they now keep their mock in a local, which is cleaner anyway.

Stack tip: 1743 tests pass, ruff clean, ty at the pre-existing baseline with zero new findings, docs -W clean.

[codecov]

Codecov Report

❌ Patch coverage is 96.10390% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.08%. Comparing base (e64d6e5) to head (08e6cd1).

Files with missing lines	Patch %	Lines
pyhap/service.py	91.30%	2 Missing ⚠️
pyhap/characteristic.py	96.66%	1 Missing ⚠️

Additional details and impacted files

@@                         Coverage Diff                         @@
##           refactor/getter-error-isolation     #563      +/-   ##
===================================================================
+ Coverage                            92.06%   92.08%   +0.02%     
===================================================================
  Files                                   54       54              
  Lines                                10920    10996      +76     
  Branches                              1407     1413       +6     
===================================================================
+ Hits                                 10053    10126      +73     
- Misses                                 579      582       +3     
  Partials                               288      288              

Files with missing lines	Coverage Δ
pyhap/delegates.py	100.00% <100.00%> (ø)
pyhap/characteristic.py	97.85% <96.66%> (-0.15%)	⬇️
pyhap/service.py	97.72% <91.30%> (-2.28%)	⬇️

🚀 New features to boost your workflow:

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

[ikalchev]

Closing per maintainer decision — continuing locally; the work continues on local branches and will return as PRs when the design settles.