Fix: SPH 3-6kW (DTC 3502) still auto-detects as sph_8000_10000_hu due to floating PV3 input (#337)
The v1.0.2 detection reorder was correct but incomplete. On 2-string SPH 3-6kW inverters the unconnected PV3 input floats at ~0.1 V, so legacy register 11 reads raw 1. The > 0 threshold treated this residual as "PV3 present", triggering the HU branch.
Two-part fix:
- Trust V2.01 when readable — if the V2.01 PV3 register (31018) returns any value (including 0), that response is authoritative and the legacy reg-11 fallback is skipped entirely.
- Noise-floor threshold on reg-11 — when the legacy fallback is still needed, threshold raised from
> 0to> 30(3 V at ×0.1 scale). Floating inputs read ~1 raw; any energised PV string reads 300+ under daylight.
Fix: Optional VPP range re-logs WARNING every ~5 minutes forever on single-battery SPH/MIN (#341)
The retry logic deleted the failure entry before re-reading the range. If the re-read failed again, the fail count reset to 1, re-triggering the WARNING every retry cycle. The "warn once, then DEBUG" intent was unreachable for permanently absent ranges (e.g. battery cluster-2/3/4 on a single-battery unit).
Fix: the entry is no longer deleted before the retry read. The count survives and increments past 1, logging at DEBUG only. The entry is cleared on a successful read.
Fix: TOU time writes cross-contaminate sibling registers on back-to-back writes (#342)
The atomic FC16 write for SPH and MOD TOU periods sourced sibling registers from coordinator.data (up to 60 s stale). Writing start then end within one poll window caused the end-write to read the old start from cache and revert it. Predbat and other controllers that program start+end ~1 s apart were consistently affected.
Fix: both GrowattGenericTime and GrowattModTouTime now perform a fresh hardware read of the full register triple/pair immediately before the FC16 write. Cached data is used only as a fallback if the fresh read fails.