Code quality audit: THZ integration (iterative assessment)

[Copilot]

Three successive assessments of the THZ integration codebase, tracking quality metrics and providing a prioritised backlog of issues. No code was changed in this PR — findings fed directly into separate fix PRs merged to development.

Assessment Snapshots

Metric	Initial	Post-COP-cleanup	Post-refactor (current)
Lint errors	86	319 (register map expansion)	0
Failing tests	15	15	0
except Exception (bare)	13	9	4 (annotated)
High-CC functions (≥10)	10	14	13
Test/code ratio	0.37x	0.38x	0.41x

Issues Identified and Since Fixed

• decode_value triplication — sensor.py, cop_sensor.py, test_helpers.py all had independent copies; consolidated into value_codec.decode_raw_value with a compat wrapper
• hass.data multi-instance unsafe — global hass.data[DOMAIN]["device"] storage would corrupt a second config entry; moved to hass.data[DOMAIN][entry_id]
• Dead code — calendar.py, schedule.py, sensor_meta.py were never loaded by the integration (not in async_forward_entry_setups); deleted
• asyncio.sleep(0.01) inside async with device.lock — held the lock for an extra 10ms on every write; removed all 7 occurrences
• send_request() / _list_serial_ports() CC≈26 — decomposed into focused helper methods (_do_handshake_1/2, _exchange_once, _build_by_id_map, etc.)
• decode_raw_value() CC≈14 — replaced 14-branch if/elif chain with a _DECODE_DISPATCH dict; CC now ≈3
• pytest-asyncio missing from requirements_test.txt — 13 async tests silently never ran
• 319 E501 regressions from inline register-map metadata tuples — fixed via per-file-ignores for register_maps/*.py

Still Open (low priority)

• _async_handle_read_raw_register (184 lines) is a nested closure inside _async_setup_services — untestable in isolation; candidate for extraction to module level
• has_entity_name=True conditionally gated on translation_key presence — inconsistent with HA naming conventions for legacy entities
• No unit tests for write-entity platforms (number, switch, select, time)

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[bigbadoooff]

@copilot

analyze again with the current state of development branch

[Copilot]

@copilot

analyze again with the current state of development branch

Here is the updated assessment against the current tip of development (commit 0f4d664).

---

Code Quality Assessment — development branch (2026-02-24)

What Changed Since Last Assessment

Four commits landed since the previous analysis:

Commit	Change	Quality impact
1d8340f	Remove Monthly + Yearly COP sensors	✅ cop_sensor.py shrunk from 886 → 577 lines, 5 classes → 3
992b25b	Fix area schema to use vol.In	✅ Minor correctness fix
710f2bb	Add BLOCK_LABELS dict to const.py + translations	➡ Neutral — adds 61-line dict to const.py, improves UI
0f4d664	Fix ventilator speed unit/device_class	✅ Correctness fix in sensor_meta.py

---

Metrics at a Glance (dev branch vs previous)

Metric	Previous	Now	Δ
Production lines	11,704	11,486	−218 ✅
Test functions	386	393	+7 ✅
Tests passing	371	378	+7 ✅
Tests failing	15	15	0 (same failures)
Lint errors (ruff)	86	80	−6 ✅
Broad except Exception	13	13	0
Functions > 50 lines	20	20	0
High-CC functions (≥10)	10	14	+4 ⚠️
Type hint coverage	80%	81%	+1% ✅
Test/code ratio	0.37x	0.38x	≈same

---

Findings — Status by Category

✅ Resolved / Improved

• cop_sensor.py reduced from 886 → 577 lines by removing unused Monthly/Yearly COP classes. The native_value CC≈11 offender from the previous report (the monthly/yearly classes) is now gone.
• Ventilator speed sensor now has correct device_class and unit.
• Area schema fixed to use vol.In (stricter, correct).
• 7 new tests added, all passing.

⚠️ Still Open (unchanged from previous report)

1. decode_value duplicated — still copy-pasted in sensor.py and cop_sensor.py. value_codec.py exists for this. The cop_sensor.py copy still has the comment "duplicated to avoid circular imports" even though importing from value_codec has no circularity at all.

2. hass.data multi-instance safety — device, device_id, write_manager, register_manager are still stored at hass.data[DOMAIN] (global, not per entry_id). A second config entry would overwrite the first. The duplicate setdefault is now actually three calls (lines 44, 86, 156) instead of two.

3. sensor_meta.py — still 971 lines / 97 entries. The file's module docstring still claims to be the metadata store, while register map tuples also carry metadata via a 6th element. No single source of truth.

4. 13 bare except Exception — identical count and locations as before. No narrowing of exception types.

5. 20 functions > 50 lines — same list. send_request() (136 lines, CC≈26) and _list_serial_ports() (96 lines, CC≈26) remain the worst offenders.

6. High-CC function count increased from 10 → 14 — _async_enable_integration_disabled_entities() (CC≈14) and should_hide_entity_by_default() (CC≈11, grown due to new BLOCK_LABELS content in const.py) are newly flagged. The function should_hide_entity_by_default at CC≈11 / 58 lines is itself a smell — hiding logic should be data-driven (a set of keywords), not a chain of if branches.

7. asyncio.sleep(0.01) inside async with device.lock — still present in 9 places across switch.py, select.py, number.py, time.py, calendar.py, schedule.py.

8. 15 failing tests — identical set. 13 fail due to pytest-asyncio not being in requirements_test.txt. 2 fail due to incomplete mock in test_thz_device_extended.py. These are non-functional tests that appear to be written but never run.

9. has_entity_name only conditionally set (when a translation_key exists) — entity naming is inconsistent between translated and untranslated entities.

10. 1 live f-string in a _LOGGER call — thz_device.py:630 uses _LOGGER.error(f"...") instead of _LOGGER.error("...", e). 6 commented-out _LOGGER debug lines also remain in thz_device.py.

11. 4 internal markdown files in repo root — BUGFIX_CONNECTION_TIMEOUT.md, CODE_REVIEW_SUMMARY.md, IMPLEMENTATION_SUMMARY.md, REFACTORING_SUMMARY.md are developer artefacts, not user docs. Should move to docs/ or be removed.

---

Priority Order for Next Steps

Priority	Item
High	Fix hass.data global-vs-per-entry (multi-instance safety)
High	Add pytest-asyncio to requirements_test.txt to un-fail 13 tests
Medium	Consolidate decode_value into value_codec.py
Medium	Narrow the 13 except Exception to specific types
Medium	Make should_hide_entity_by_default data-driven (set of keywords/prefixes) to reduce CC
Low	Move the 4 internal .md files out of repo root
Low	Fix the lone f-string logger call + remove 6 commented-out debug lines in thz_device.py

Overall grade: C+ → B− (slight improvement due to COP cleanup and test additions, but the structural issues remain untouched).

[bigbadoooff]

@copilot assess again.

[Copilot]

@copilot assess again.

Here is the updated assessment against the current tip of development (commit 1014460, v0.3.0-alpha).

---

Code Quality Assessment — development branch (2026-03-02)

What Changed Since Last Assessment (24 commits)

Area	Change	Impact
sensor_meta.py	Deleted	✅ Single source of metadata truth (register map tuples)
schedule.py	Deleted (dead code)	✅ Removes ~250 lines of never-loaded code
calendar.py	Deleted (dead code)	✅ Removes ~425 lines of never-loaded code
decode_value	Consolidated — cop_sensor.py now imports from value_codec.py; test_helpers.py is now a 1-line re-export	✅ Duplication fully resolved
base_entity.py	_attr_should_poll=False with async_track_time_interval	✅ Correct HA polling pattern
hass.data.setdefault	Reduced from 3 → 2 calls	✅ One duplicate removed
cop_sensor.py	Added THZBaseCOPSensor base class; Monthly/Yearly removed	✅ Better structure, 309 fewer lines
value_codec.py	Added somwinmode, weekday, hexdate, clockdate decoders	➡ Growth adds CC complexity
register_map_all.py	Massively expanded with inline metadata tuples	➡ Generates 183 new E501 errors
thz_device_extended tests	Fixed — all 27 now pass	✅
Dead-code helpers removed	get_device_from_hass, read_block_cached deleted	✅

---

Metrics at a Glance

Metric	Previous (0f4d664)	Now (1014460)	Δ
Production Python files	37	34	−3 ✅
Production lines	11,486	10,518	−968 ✅
Test functions	393	375	−18 (removed tests for deleted modules)
Tests passing	378	362	−16 (same ratio — deleted module tests removed)
Tests failing	15	13	−2 ✅
Lint errors (ruff)	80	319	+239 ⚠️
Broad except Exception	13	9	−4 ✅
Functions > 50 lines	20	19	−1 ✅
High-CC functions (≥10)	14	12	−2 ✅
Type hint coverage	81%	79%	−2%
Test/code ratio	0.38x	0.41x	+0.03x ✅

---

Findings — Status by Category

✅ Resolved Since Last Assessment

1. decode_value duplication — fully resolved. value_codec.py owns decode_raw_value(). sensor.py exports it as a compat wrapper. cop_sensor.py imports directly. test_helpers.py is now a clean 1-line re-export.
2. sensor_meta.py — deleted. Register map tuple 6th element is now the sole metadata source.
3. schedule.py / calendar.py — dead code removed (~675 lines).
4. thz_device_extended mock failures — fixed; all 27 tests pass.
5. _attr_should_poll — correctly False; write entities use async_track_time_interval.
6. Duplicate setdefault — down to 2 calls (still one redundant on line 43 and 157, but much better).
7. except Exception count reduced from 13 → 9 (removed with calendar/schedule).

⚠️ Newly Introduced Issues

Lint errors spike: 80 → 319 — almost entirely E501 (line too long). This comes from register_map_206.py (146 violations) and register_map_all.py (34 violations) where inline metadata tuples produce very long lines like:

("mixerOpen:", 45, 1, "bit0", 1, {"icon": "mdi:gate-open", "translation_key": "mixer_open"}),

The pyproject.toml excludes docstring rules for register_maps/ but not E501. Consider either raising the line limit for register map files or wrapping the metadata dict onto its own line. This is the single biggest quality regression.

decode_raw_value() (value_codec.py) CC≈14, 63 lines — the function gained new decode branches (somwinmode, weekday, hexdate, clockdate, opmodehc, 8party). With 14+ branches it is hard to read and test each path independently. A dispatch dict {decode_type: handler} would cut this to CC≈1.

⚠️ Still Open (unchanged)

1. hass.data multi-instance safety — device, device_id, write_manager, register_manager still stored at hass.data[DOMAIN] (global, not per entry_id). Two config entries would still collide.
2. 9 bare except Exception — still in thz_device.py (6×), __init__.py (2×), config_flow.py (1×).
3. send_request() CC≈26, 136 lines and _list_serial_ports() CC≈26, 96 lines — unchanged.
4. asyncio.sleep(0.01) inside async with device.lock — still in 7 places (switch, select, number, time ×4).
5. 13 tests still failing (all 13 due to missing pytest-asyncio in requirements_test.txt).
6. 1 live f-string logger call — thz_device.py:603 still uses _LOGGER.error(f"...").
7. 6 commented-out _LOGGER lines in thz_device.py — still present.
8. 4 internal .md files in repo root — BUGFIX_CONNECTION_TIMEOUT.md, CODE_REVIEW_SUMMARY.md, IMPLEMENTATION_SUMMARY.md, REFACTORING_SUMMARY.md. A CHANGELOG.md was added (appropriate), but the others remain clutter.

---

Priority Order

Priority	Item
High	Fix E501 spike in register maps (319 → ~0 lint errors)
High	Add pytest-asyncio to requirements_test.txt (un-fail 13 tests)
Medium	Fix hass.data multi-instance safety (per-entry_id storage)
Medium	Refactor decode_raw_value() as a dispatch dict to reduce CC
Medium	Narrow remaining 9 except Exception catches
Low	Remove 4 internal .md files from repo root
Low	Fix lone f-string logger + 6 commented-out debug lines in thz_device.py

Overall grade: C+ → B− (significant structural improvements, but the lint regression from the register map expansion needs attention).