[devilSatou]

Summary

Fixes #4.

On Arduino core 3.2+ (ESP-IDF 5.4+), adding this library to a project aborts before setup() runs. This was observed on an M5Stack CoreS3 with pioarduino platform-espressif32 (stable):

E (1057) i2c: CONFLICT! driver_ng is not allowed to be used with this old driver
abort() was called at PC 0x4204ef83 on core 0

addr2line:

0x4204ef83: check_i2c_driver_conflict at components/driver/i2c/i2c.c:1738
0x4202ab9e: do_global_ctors at components/esp_system/startup.c:104

The "included because of" section of firmware.map:

libesp_driver_i2c.a(i2c_master.c.obj)
    ← libFrameworkArduino.a(esp32-hal-i2c-ng.c.o)  (i2c_new_master_bus)   ← Arduino core's Wire
libdriver.a(i2c.c.obj)
    ← M5Module-GNSS/BMI270.cpp.o  (i2c_master_write_to_device)            ← this library

Root cause

• BMI270.cpp calls i2c_master_write_to_device() and i2c_master_write_read_device() directly. Despite their names, both belong to the legacy I2C driver (components/driver/i2c/i2c.c). Referencing them links i2c.c.obj, and that object carries a constructor, check_i2c_driver_conflict(), which aborts if the new i2c_master driver is linked too.
• Since Arduino core 3.2.0, Wire is built on the new driver (cores/esp32/esp32-hal-i2c-ng.c → i2c_new_master_bus(), IDF ≥ 5.4), so both drivers end up in every binary. Core 3.1.x and earlier still used the legacy driver for Wire, which is why ESP3.2 conflict #4 started with 3.2.

Changes

src/utilities/ (the Bosch BMI270/BMM150 Sensor API) is untouched.

src/BMI270.h / src/BMI270.cpp

• bmi2_i2c_read and bmi2_i2c_write now use TwoWire:
  • read: beginTransmission → write(reg) → endTransmission(false) (repeated start) → requestFrom(addr, len) (checked against len) → read()
  • write: beginTransmission → write(reg) → write(data, len) → endTransmission(true)
  • They still return 0 on success and -1 on failure. On failure, ESP_LOGE now logs the address, the register and Wire's return code.
• The file-scope static uint8_t _data_buffer[32], static esp_err_t _i2c_ret and struct dev_info* _dev_info were mutable state shared by all instances and tasks. They are now locals. The 32-byte limit (bmi2.read_write_len = 30) is kept.
• dev_info::i2c_port and _i2c_port (i2c_port_t) are replaced by wire and _wire (TwoWire*).
• New: int init(TwoWire& wire, uint8_t bmiAddr).
• int init(i2c_port_t, uint8_t) is kept, deprecated, for source compatibility. I2C_NUM_0 maps to Wire, and I2C_NUM_1 maps to Wire1 on chips with a second HP I2C controller.
  • On IDF 5.x, i2c_port_t comes from the header-only hal/i2c_types.h.
  • On IDF 4.x (core 2.x), it comes from driver/i2c.h, which is where I2C_NUM_x are defined there. IDF 4.x has no new driver to conflict with.
  • Including a header links nothing; only calling the legacy functions does.
• Behavior change: I2C transactions now use Wire's timeout (50 ms by default) instead of portMAX_DELAY.
• The library still does not call Wire.begin(). Pins and clock remain the sketch's responsibility.

Example, docs and metadata

• examples/getSensorData: bmi270.init(I2C_NUM_0, ...) → bmi270.init(Wire, ...)
• README: documents the change and the new init().
• Version → 1.0.3

This branch is stacked on #7; its first commit is the version bump from #7. If #7 is merged first, only the Wire port remains here. Merging this PR alone brings in both.

Testing

I couldn't build with the ESP32 toolchain in my environment, so this branch has not had a hardware run or a firmware.map check yet. What I did instead:

• Compile check: compiled src/BMI270.cpp, plus a sketch that calls both init() overloads, with host g++ (-Wall -Wextra). The stubs reproduce the relevant real declarations:

  • the TwoWire I/O overload sets and the Wire/Wire1 declaration conditions of arduino-esp32 2.0.17, 3.0.7, 3.1.3 and 3.3.12
  • the i2c_port_t / I2C_NUM_x definitions of ESP-IDF 4.4.7, 5.1, 5.3 and 5.5
  • the SOC I2C counts of ESP32-S3, ESP32-C3 and ESP32-C6

  All 10 configurations compile, including core 3.3 with NO_GLOBAL_WIRE. There are no new warnings; the three remaining -Wunused-parameter warnings were already there.

• Symbol check: nm on the resulting BMI270.o finds no references to i2c_master_write_to_device or i2c_master_write_read_device. The object built from current main references both, and those references are what make the linker pull in libdriver.a(i2c.c.obj).

To confirm end to end, build examples/getSensorData for ESP32-S3 on core 3.2+ and run grep 'libdriver.a(i2c.c.obj)' firmware.map. It should find nothing.

🤖 Generated with Claude Code