[springfall2008]

This is an automated draft PR generated from issue #5850 — a maintainer should review it before merging.

Fixes #5850

Summary

For inverter types with no reserve register (has_reserve_soc: False — GS, SE, SX4, SF, SFMB, HU, SK, SIG, SIGCLOUD), refresh_config() seeded the dummy reserve entity with self.reserve (kWh) while labelling it %. The next refresh read that back as reserve_percent_current, which balance_inverters() uses for its above-reserve check.

• The dummy (and the matching create_missing_arg default) is now seeded with self.reserve_percent.
• create_entity() never overwrites an existing state, so an install that already has the kWh seed would keep it. After creating the entity, if what it reads back (via the same get_arg(..., required_unit="%") that line 774 uses) differs from reserve_percent, the percent is written. Nothing on the device needs preserving: these inverters have no reserve register, so Predbat owns the value. With set_reserve_enable forced off for these types (execute.py:1460), the only writes Predbat makes to this entity are adjust_reserve(0) on the calibration and reset paths, and those are clamped up to reserve_percent anyway.

Testing

• coverage/./run_pre_commit: all hooks passed, and the --quick suite it then runs passed.

• New test_reserve_dummy_percent in tests/test_inverter.py, with two cases. Both use an HU inverter with soc_max 10 kWh, set_reserve_min 50 and battery_min_soc 3. battery_min_soc is set low because the triage found its floor hides the kWh value in default configs.

  • reserve_dummy_seeded_percent: a fresh dummy.
  • reserve_dummy_kwh_seed_corrected: a dummy that already holds the old kWh seed of 5.0.

  Both cases check the dummy's own state and unit (50, %) and reserve_percent_current == 50 after a refresh.

• Fails without the fix, passes with it. With only inverter.py stashed, tools/triage_test.sh inverter failed on both cases (dummy reserve should be 50 % got 5.0 %, reserve_percent_current should be 50 got 5). With the fix restored, the whole inverter module passes.

Notes

• The correction happens where the dummy is created, which is after line 774 reads reserve_percent_current. On the first refresh that corrects an old kWh seed, that refresh's reserve_percent_current is still the stale value. It reads correctly from the next refresh, one plan cycle later. I chose not to also overwrite reserve_percent_current in the same pass, because reserve_literal_huawei (GH#5003) deliberately expects a hard-wired apps.yaml reserve literal to be reported as the current reserve on construction.
• The inverter self_test() calls adjust_reserve(100/6/4). On these types the next refresh now resets the dummy back to reserve_percent. Since the dummy controls no device, that seems right.
• Blast radius: GitNexus impact(refresh_config, upstream) is LOW (2 direct callers, Inverter.__init__ and the per-cycle refresh). detect_changes shows only refresh_config changed outside the tests.

🤖 Generated with Claude Code

[springfall2008]

Automated comment from the triage bot.

Review of 222d7ad: the seeding fix is sound — 1 finding inline, 0 in the body.

What was checked: the seed/comparison mechanics (unit_conversion returns a float for numeric states, so the != read-back is stable and cannot rewrite every cycle), the created_attributes lookup (set unconditionally in create_entity(), so no KeyError), the claim in the new comment about who writes the dummy (all three adjust_reserve(0) sites clamp up to reserve_percent; the hold/freeze sites are gated on set_reserve_enable, which execute.py forces off for these types; self_test runs only under INVERTER_TEST), device bounds for the dummy (no min/max attributes, so (None, None)), the new test's registration, red/green design and restore/teardown, and the literal (GH#5003) interaction.