Skip to content

Changelog

What changed in each release of the driver. Entries under Breaking need source changes in code that used the affected API.

[Unreleased]

Added

  • One C++20 driver for the Sensirion SFM digital mass flow meters on I2C, 32 variants in six protocol families: SF06 (SFM4300, SFM3003 with -CET, SFM3013, SFM3019, SFM3119), SF06 gen 2 (SFM3304), SF06 dual (SFM3505), SF05 (SFM3000, SFM3200, SFM3300, SFM3400, SFM4200), SF04 (SFM4100) and the preliminary SFM6000D.
  • A part table with a preset and a build-time checked configuration per variant; calls that return the wait they need (Poll(), AwaitPending()); CRC-8 checked before any value is decoded; status, range and staleness checks; device-loss detection after the datasheet's silence bound (a NACK between samples is not a fault); per-family recovery rules; no empty I2C transfer anywhere (polls send the next real command).
  • A port contract with optional hooks found at compile time: delay, clock, log, clock-stretching read, general call, power cycle, reset pin.
  • An independent sensor model, host tests for every variant, compile-fail tests, a 17-invariant fuzz target, three ESP-IDF examples (compiled only), a host example, and the documentation set with the connector view of the hardware.
  • A reply frame the sensor delivered counts for the silence clock even when a later step of the same call fails (the restart write after a read, the restore after a temperature read).
  • A temperature is checked against the plausible band (-60 to +150 degC) only; a raw word is never refused for its pattern, because 0xFFFF is a real reading on the SFx6000 (-0.005 degC).
  • The silence clock follows the last reply frame the sensor delivered (accepted or rejected, and raw reads), so an error burst followed by a NACK does not read as a reset. Without NowUs() the driver never reports a reset from a count of NACKed reads; the caller times the silence and calls Stop(). A reported reset holds the next command for the stop time. Options for the open hardware questions: Config::poll_with_identifier (K-STOPIDLE) and Config::sf05_serial_low_word (K-SF05-SERIAL); on the SFM3000 a refused 0x31AF keeps the high word of the serial number. Gas::N2 is offered again at the air scale on the SFM3200-AW and SFM4200 (the datasheets list it); oxygen stays refused on the air-calibrated parts. Wake and reset polls give up a quarter after the datasheet figure.
  • A temperature outside -60 to +150 degC (a stuck bus reads as -200 degC on the SF05 parts) is InvalidReply. A failed SetAveraging() leaves StateUnknown (it may have been applied). The optional second SFM3000 serial read is skipped when its pointer was refused. Gas::N2 is offered at the air scale on every SF05 part. The fuzz target asserts invariants 9, 13 and 16 as documented and has a seventeenth for the temperature band.
  • Not verified on hardware; the SFM6000D protocol is preliminary. See the contract page.