fix(tests): pin midnight_utc in the multi-car IOG tests so they don't fail for an hour a day

[springfall2008]

Two multi_car_iog tests fail for roughly one hour a day. They took down CI on #4995 — a docs-only PR (tools/debug-journal.md + one cspell word) whose test step happened to run at 23:11:37 UTC:

ERROR: Expected car 0 to have charging slots from IOG, got none
ERROR: Expected car 1 to have charging slots from IOG, got none
Test: multi_car_iog_load_slots_regression FAILED
ERROR: car 0 should have IOG slots, got none
ERROR: car_charging_limit_model should be set with consider_full off and IOG slots present, got None
Test: multi_car_iog_model_limit_fetch_4967 FAILED

Cause

run_multi_car_iog_load_slots_test and run_iog_model_limit_fetch_test pin now_utc to noon on the current UTC date, but leave midnight_utc on the real clock. update_time() derives midnight_utc from datetime.now(Europe/London) — London-local midnight, independent of the runner's TZ.

Between 23:00 and 00:00 UTC under BST the London date is already the next day, so midnight_utc lands on the following UTC date, after the pinned now_utc. On #4995 that was midnight_utc = 7 Sept 23:00 UTC against a pinned now_utc of 7 Sept 12:00 UTC — eleven hours earlier.

Every IOG dispatch built as now+1h..now+3h therefore decodes to negative minutes relative to midnight_utc, and decode_octopus_slot() clamps them away:

start_minutes = max(start_minutes, 0)
end_minutes = max(min(end_minutes, self.forecast_minutes + self.minutes_now), start_minutes)
if start_minutes == end_minutes:
    return 0, 0, 0, source, location

kwh > 0 never holds, no slots load, and the assertions fail. Outside that hour the London date matches the UTC date, midnight_utc is at or before the pinned noon, and the slots land at +780/+840 minutes. Under GMT there is no offset, so the window is BST-only.

Fix

• Shared pin_test_clock() / restore_test_clock() helpers. The pin sets midnight_utc = now_utc.replace(hour=0) alongside now_utc/minutes_now, so slot offsets are a fixed 780/840/900 minutes whatever the wall clock says.
• run_multi_car_iog_adhoc_dispatch_test uses the helper too — it had the same latent hazard. In the failing window its IOG slot was silently clipped away and the test passed only on the car_charging_now fallback slot, which is not what it means to assert.
• The clock is restored on the way out instead of left pinned: restore_test_clock() in the two tests without a snapshot harness, and midnight_utc added to snapshot_attrs in run_iog_model_limit_fetch_test. Previously now_utc leaked out pinned while midnight_utc stayed real.

No production code is touched.

Verification

Ran the module under the real clock and under a faithful emulation of the CI instant (00:11 Europe/London on the day after the current UTC date = 23:11 UTC), before and after the change:

	real clock	simulated 23:11 UTC
before	8/8 pass	2 fail (identical error text to the CI log)
after	8/8 pass	8/8 pass

Full quick suite: All tests passed (4 slow tests skipped, total time: 83.03s). Pre-commit hooks on the changed file all pass.

#4995 needs only a CI re-run once this lands.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Clock restoration is not exception-safe in the updated tests, so an unexpected error can still leak pinned time into later tests via the shared PredBat fixture.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR stabilizes the multi-car Intelligent Octopus Go (IOG) unit tests by pinning midnight_utc alongside now_utc, avoiding a BST-only 23:00–00:00 UTC window where dispatch slots can be decoded as negative minutes and clamped away.

Changes:

• Introduces pin_test_clock() / restore_test_clock() helpers to pin now_utc, midnight_utc, and minutes_now consistently for these tests.
• Updates run_multi_car_iog_load_slots_test and run_multi_car_iog_adhoc_dispatch_test to pin and then restore the clock state.
• Extends the snapshot/restore harness in run_iog_model_limit_fetch_test to include midnight_utc, so the pinned value is restored after the test.

File summaries

File	Description
apps/predbat/tests/test_multi_car_iog.py	Adds clock pin/restore helpers and uses them to prevent time-dependent IOG slot decoding failures during BST.

Review details

Suppressed comments (2)

apps/predbat/tests/test_multi_car_iog.py:488

• Similar to the load-slots test: restore_test_clock() is not guaranteed to run if an exception is raised after pin_test_clock(), which can leak the pinned clock into later tests because unit_test.py shares a single PredBat instance across the suite.

    saved_clock = pin_test_clock(my_predbat)

    my_predbat.args["car_charging_loss"] = 0.0

apps/predbat/tests/test_multi_car_iog.py:586

• pin_test_clock() mutates now_utc/midnight_utc/minutes_now but this test only restores state at the end; if fetch_sensor_data_cars() (or later assertions) throws unexpectedly, the pinned clock can still leak into subsequent tests via the shared PredBat fixture. Consider putting the restore of saved_args/saved attrs into a finally block so it always executes.

    pin_test_clock(my_predbat)

• Files reviewed: 1/1 changed files
• Comments generated: 1
• Review effort level: Lite

---

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