fix(component_base): stop components reading calculate_yesterday's faked minutes_now

[springfall2008]

Posted by Claude on behalf of @springfall2008.

Stacked on #5084 — merge that first. Base is set to its branch, so this diff shows only the change itself.

The other half of #4804. #5054 stopped components seeing the midnight_utc that calculate_yesterday() rewinds. It fakes minutes_now to 0 in the same breath (output.py:3291), and that passthrough was deliberately left alone at the time: base.minutes_now was measured from a different clock than now_utc, so deriving it would have made components disagree with the planner wherever the container's timezone differed from the timezone: setting. #5084 removes that second clock, so the derivation is now exactly the calculation update_time() makes — through the same helper.

Why this one matters more than the date

A rewound date is obviously wrong. 0 reads as a legitimate "just after midnight", so it fails quietly. It reaches 39 component-side call sites:

File	Reads	Effect of minutes_now = 0
octopus.py	19	end_minutes <= self.minutes_now stops dropping expired dispatch slots; start_offset = start_minutes - self.minutes_now shifts every published offset by the real time of day
load_ml_component.py	9	minutes_now % 30 slot alignment, history windows, ML feature payload
solcast.py	5	minute_absolute = self.minutes_now - minute, today's actual-PV alignment
web.py	2	SOC history chart range
alphaess.py, deye.py, sunsynk.py	1 each	_now_minutes() / _window_active() pick the live TOU slot while writing to a real inverter — 0 reads as midnight

axle.py also reads base.minutes_now, but load_axle_slot() is called from fetch.py on the base's own thread — it should see the faked values, and this does not touch it.

Tests

ComponentBase gets a test that compares the derived value against update_time()'s on the real base rather than restating the formula, so the two cannot drift, plus the rewind case itself.

The AlphaESS doubles set base.minutes_now on a MagicMock base, so they were answering arithmetic with more mocks rather than failing. They get a real clock and a set_mock_clock() helper that moves it — writing minutes_now directly no longer does anything, which is the point.

Mutation-tested: reverting the property fails with Faked base.minutes_now leaked into the component: 0.

• ./run_all --quick green (314 tests) under both a mismatched host timezone and TZ=Europe/London
• ./run_pre_commit green

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Three moderate findings remain unresolved.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

This follow-up prevents components from reading calculate_yesterday()’s temporary minutes_now = 0 by deriving time from the stable clock.

Changes:

• Derives component minutes from shared time state.
• Adds rewind regression coverage.
• Updates AlphaESS tests with coherent mock clocks.

File summaries

File	Review summary
apps/predbat/component_base.py	Two moderate findings: snapshot now_utc consistently (3 votes) and update Deye/Sunsynk test doubles (2 votes).
apps/predbat/tests/test_component_base.py	Moderate finding (3 votes): restore all fields mutated by update_time().
apps/predbat/tests/test_alphaess_control.py	Reviewed mock-clock updates.
apps/predbat/tests/test_alphaess_api.py	Reviewed coherent mock-clock handling.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 3
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[springfall2008]

Posted by Claude on behalf of @springfall2008.

All three findings held up this time. Fixed in d4eb4b6.

1. minutes_now read base.now_utc twice — real, and the sharpest of the three.

Once directly, once through self.midnight_utc, which goes back to the same field. update_time() runs on the main thread while components run on theirs, so the two reads can straddle it. At a day boundary that subtracts the new day's midnight from the old day's timestamp: (23:59 day 1) - (00:00 day 2) is -60s, which int() truncates toward zero, so it returns 0 — reading as midnight rather than as an error. Exactly the failure mode this PR exists to remove, reintroduced by a second read.

Snapshotted once now. A base whose now_utc rolls into the next day between reads reproduces it deterministically without needing two threads to interleave; mutation-tested (now_utc should be read once per call, was read 2 times).

2. Deye and Sunsynk doubles — real, same shape as AlphaESS.

Confirmed: MagicMock answers (mock - mock).total_seconds() / 60 / 5 with arithmetic of its own — int() of it is 1, so _now_minutes() silently returns 5. Their comment promised "tests set this for time-aware control" while a test setting base.minutes_now would have been ignored. Both doubles get a real clock and the same set_mock_clock() helper AlphaESS got.

3. Incomplete restore in the new test — real. It saved three of the seven fields update_time() writes, leaking this test's wall-clock state into later tests through the shared PredBat. Restores the full set now, matching what tests/test_clock.py already does.

./run_all --quick green (314 tests), pre-commit clean.