Ir al contenido

11 · Changelog

1.3.0 — Fleet + v1.3 protocol

Brought the rbamp Python package in line with the rbAmp v1.3 wire contract and the cross-platform reference model (Arduino / ESP-IDF / ESPHome / STM32 HAL). Bench-validated: 11/11 PASS on a heterogeneous Fix-A fleet (UI1 + I2 + I3) standfw 2026-06-17 + 179/179 mock tests PASS. Schema CRC: 0x5FB3E9F3 (registers_v2).

Added

  • RbAmpFleet — host-side multi-module manager: scan() (with conflict-detect + Tier-2 wedge-canary), add(), find(), count, iteration; batched poll_all() (failure-isolated, MISS-resilient); fleet-wide total_power() / total_energy_wh() / poll_errors(); General-Call sync (enable_gc_all() / gc_latch() / check_sync()); assign_address() / check_conflict(). Soft cap RBAMP_FLEET_MAX_MODULES = 16.
  • POD structures: RbAmpFleetPoll (per-member poll result), RbAmpFleetSync (per-member GC sync witness).
  • Identity / capability: read_variant() (HW_VARIANT 0x55 byte, 1..6 = UI1/UI2/UI3/I1/I2/I3), read_capability() (u16 LE feature bitmap; branch on bits, not version heuristics), read_product_id() (0x01 = rbAmp sensor family), read_uid() (12 bytes), read_label() / set_label() (8-byte ASCII via byte-loop write).
  • Event / error channel (v1.3 two-channel model): read_last_error() (device REG_ERROR byte, one wire read), read_event_flags(), clear_event_flags(mask), has_error() (sticky bit3 helper), clear_error() (CMD_CLEAR_ERROR).
  • Per-channel CT configuration: configure_channels(class, models, n) (batched, one terminal flash SAVE; variant-clamped; client-side per-class validation _validate_ct_code); read_ct_model_ch(ch) (applied-model mirror read from 0x51-0x53).
  • Per-device fleet primitives: enable_gc(enable=True) (RMW REG_FLEET_CONFIG.bit0 + save + reset; ~1 s blocking; idempotent no-op if already correct), read_fleet_config(), set_group_id() / read_group_id(), read_gc_tick() (u16 witness; 0xFFFF = never received), RbAmp.broadcast_latch_group(bus, group, tick) static (5-byte GC frame A5 27 g tl th).
  • Exception hierarchy: RbAmpError base + RbAmpIOError, RbAmpTimeoutError, RbAmpStaleError, RbAmpParamError, RbAmpModeError, RbAmpVersionError.

Changed

  • Energy is integrated against the master wall-clock, never against the chip's diagnostic latch_ms (the chip timer undercounts by ~25-30 %; HW-validated bench measure: master_dt/wall = 0.999 vs latch_ms/wall = 0.743 = 26 % undercount). On CPython use time.monotonic(); on MicroPython use time.ticks_ms() + time.ticks_diff(). On a stale read (PERIOD_VALID = 0) the library raises RbAmpStaleError and does NOT reset the anchor; the next valid snapshot covers the full interval, the firmware preserves the accumulator → the integration math stays exact, with no double- or under-counting.
  • CT model codes are a per-class accepted set (SCT_013 {1,2,3,4,6}, WIRED_CT {1,2,3}, BUILTIN_CT ∅); 5 (SCT-013-100) and 7 (SCT-013-060) are reserved-uncharacterised, and the library raises RbAmpParamError before any I²C operation. REG_CT_MODEL is pure staging: bind via a per-channel command, and multi-channel binds are order-independent on v1.3 firmware (on legacy v1.x the defensive descending pattern is retained in the set_ct_model_ch cycle).
  • Address change is a production-OK two-phase magic commit; develop mode is not required. prepare_address_change() arms the candidate + the 0xA5 magic window of 5 seconds; commit_address_change() issues CMD_COMMIT_ADDR + reset. The library guarantees the arm-state is cleared via try/finally on partial failure; an expired window → RbAmpTimeoutError. Compatible with legacy v1.x firmware via a capability-gated fallback.
  • Variant detection — read_variant() reads REG_HW_VARIANT (the v1.3 canonical SKU byte). The NACK-probe path has been removed (the firmware never NACKs reads; unmapped reads → 0x00).
  • Multi-byte WRITE byte-at-a-time (F.13 wire-canon, HW-confirmed). The library's set_label() writes the 8 LABEL bytes with a byte-loop automatically (rbAmp does not auto-increment writes).

Bench-robustness (Stage 1B)

  • CPython SMBusBackend exposes a generic NACK-retry (retry_attempts=3, retry_gap_ms=2) with retry_count_total + retry_exhaustion_count diagnostics. It is used on all wire operations (config writes follow the same discipline as reads — otherwise a config write that is silently dropped on a contended bus would be invisible).
  • MicroPython-on-ESP32 MachineI2CBackend keeps the IDF i2c_master 50 kHz spin-discipline (NACK-retry with a tight wait between attempts so a contended bus does not silently drop a write). See 09 · API Reference → "Wire-protocol details" for the full contract.

Notes

  • Marginal bus / ESP32-based hosts: the same IDF i2c_master driver used in native ESP-IDF projects — it can hang on a held bus below the library level. We recommend external ~4.7 kΩ pull-ups, no debugger NRST in production, and an app-level task-watchdog on the polling task as a recovery posture. See chapter 10 · Troubleshooting.

1.1.0 — extended period snapshot

Extended period snapshot (max P, latch_ms diagnostic), master-side Wh accounting.

1.0.0 — Initial release

Single-device RT metering (RMS U / I / P / PF / frequency), period energy (Wh) integration, CT-model + sensor-class configuration, and I²C address change. Dual-backend (CPython smbus2 / MicroPython machine.I2C).

1.3.0 Fleet + v1.3 protocol 1.1.0 extended period snapshot 1.0.0 Initial release