Program the ESP32 using Nim! This library builds on the esp-idf. Nim now has support for FreeRTOS & LwIP. Combined with the new ARC garbage collector makes Nim an excellent language for programming the ESP32.

See Releases for updates.

This project is fairly stable and even being used in shipping hardware project. The documentation is rough and primarily only includes this README. As such, it may still require understanding the underlying ESP-IDF SDK for various use cases. However, it is useable and several people unfamiliar with ESP-IDF and embedded programming have added wrappers for esp-idf modules.

Note: It's recommended to use the ESP-IDF.py v4.0 branch (as of 2020-11-24). Branch v4.1 has multiple serious bugs in I2C.

1. Install ESP-IDF
   • TLDR: git clone -b release/v4.0 --recursive https://github.com/espressif/esp-idf.git
   • esp-idf version 4.0 is recommended for now since its more stable
   • esp-idf version can be set using the defines: -d:ESP_IDF_V4_0 or -d:ESP_IDF_V4_1
2. Install Nim 1.4
3. Use Nimble to install Nesper (nimble install https://github.com/elcritch/nesper or for the devel branch nimble install 'https://github.com/elcritch/nesper@#devel' )
4. Create a new Nimble project nimble init --git esp32_nim_example
5. In the new project directory edit the Nimble file and add the lines:

requires "nesper >= 0.6.1"
# includes nimble tasks for building Nim esp-idf projects
include nesper/build_utils/tasks

• Make sure not to include a bin option like bin = @["src/esp32_nim_example"] as this will override the nimble esp_build and result in a broken idf.py build.

6. Run nimble esp_setup to setup the correct files for building an esp32/esp-idf project

1. Run nimble esp_build to build the esp-idf project
2. Flash and monitor the esp32 board using: idf.py -p </dev/ttyUSB0> flash monitor

Notes:

• Running nimble esp_build will both compile the Nim code and then build the esp-idf project
• During development it's often handy just to run nimble esp_compile to check your Nim code works
• Sometimes the Nim build cache gets out of sync, use nimble esp_build --clean to force a full Nim recompile
• Sometimes the esp-idf build cache gets out of sync, use nimble esp_build --dist-clean to force a full Nim recompile

This code shows a short example of setting up an http server to toggle a GPIO pin. It uses the default async HTTP server in Nim's standard library. It still requires the code to initialize the ESP32 and WiFi or ethernet.

import asynchttpserver, asyncdispatch, net
import nesper, nesper/consts, nesper/general, nesper/gpios

const
  MY_PIN_A* = gpio_num_t(4)
  MY_PIN_B* = gpio_num_t(5)

var 
  level = false
  
proc config_pins() =
    MOTOR1_PIN.setLevel(true)

proc http_cb*(req: Request) {.async.} =
    level = not level 
    echo "toggle my pin to: #", $level
    MY_PIN_A.setLevel(level)
    await req.respond(Http200, "Toggle MY_PIN_A: " & $level)

proc run_http_server*() {.exportc.} =
    echo "Configure pins"
    {MY_PIN_A, MY_PIN_B}.configure(MODE_OUTPUT) 
    MY_PIN_A.setLevel(lastLevel)
    
    echo "Starting http server on port 8181"
    var server = newAsyncHttpServer()
    waitFor server.serve(Port(8181), http_cb)

TLDR; Real reason? It's a bit of fun in a sometimes tricky field.

I generally dislike programming C/C (despite C's elegance in the small). When you just want a hash table in C it's tedious and error prone. C is about 5 different languages and I have to idea how to use half of them anymore. Rust doesn't work on half of the boards I want to program. MicroPython? ... Nope - I need speed and efficiency.

The library is currently a collection of random ESP-IDF libraries that I import using c2nim as needed. Sometimes there's a bit extra wrapping to provide a nicer Nim API.

Caveat: these features are tested as they're used for my use case. However, both Nim and the esp-idf seem designed well enough that they mostly "just work". PR's are welcome!

Supported ESP-IDF drivers with Nim'ified interfaces:

• Nim stdandard library support for most basic POSIX network API's!
• Most of the basic FreeRTOS.h header
• NVS Flash
• UART
• SPI (don't mix half-duplex & duplex devices)
• I2C

Other things:

• Nim standard library wrapping of FreeRTOS semaphore's, mutexes, etc
  • include pthread in your CMakeLists.txt file and use Nim's POSIX lock API's
• Nim support for xqueue and other "thread safe" data structures
  • Raw C Wrappers exist, see `rpcsocket_queue_mpack.nim for proper usage. Nim Channel's appear to work as well.
• Nim standard library support for FreeRTOS tasks using thread api's
  • include pthread in your CMakeLists.txt file and use Nim's POSIX Pthread API's

Things I'm not planning on (PR's welcome!)

• I2S
• PWM
• LCDs
• Built-in ADC

This is the more manual setup approach:

3. It's recommend to copy nesper/esp-idf-examples/simplewifi example project initially, to get the proper build steps.
   • git clone https://github.com/elcritch/nesper
   • cp -Rv nesper/esp-idf-examples/simplewifi/ ./nesper-simplewifi
   • cd ./nesper-simplewifi/
   • make build (also make esp_v40 or make esp_v41 )
4. Nesper wrapper API names generally match the C names directly, usually in snake case

• FreeRTOS functions usually are camel case and start with an x, e.g. xTaskDelay
• These api's are found under nesper/esp/* or nesper/esp/net/*, e.g. nesper/esp/nvs

6. Nesper Nim friendly api, usually in camel case

• These api's are found under nesper/*, e.g. nesper/nvs

See SimpleWifi Example

The async code really is simple Nim code:

import asynchttpserver, asyncdispatch, net

var count = 0

proc cb*(req: Request) {.async.} =
    inc count
    echo "req #", count
    await req.respond(Http200, "Hello World from nim on ESP32\n")
    # GC_fullCollect()

proc run_http_server*() {.exportc.} =
    echo "starting http server on port 8181"
    var server = newAsyncHttpServer()

    waitFor server.serve(Port(8181), cb)

when isMainModule:
    echo "running server"
    run_http_server()

import nesper, nesper/consts, nesper/general, nesper/gpios

const
  MOTOR1_PIN* = gpio_num_t(4)
  MOTOR2_PIN* = gpio_num_t(5)

proc config_pins() =
  # Inputs pins use Nim's set `{}` notation
  configure({MOTOR1_PIN, MOTOR2_PIN}, GPIO_MODE_INPUT)
  # or method call style:
  {MOTOR1_PIN, MOTOR2_PIN}.configure(MODE_INPUT)
  
  MOTOR1_PIN.setLevel(true)
  MOTOR2_PIN.setLevel(false) 

import nesper, nesper/consts, nesper/general, nesper/spis

proc cs_adc_pre(trans: ptr spi_transaction_t) {.cdecl.} = ... 
proc cs_unselect(trans: ptr spi_transaction_t) {.cdecl.} = ...

proc config_spis() = 
  # Setup SPI example using custom Chip select pins using pre/post callbacks 
  let
    std_hz = 1_000_000.cint()
    fast_hz = 8_000_000.cint()
    
  var BUS1 = HSPI.newSpiBus(
        mosi = gpio_num_t(32),
        sclk = gpio_num_t(33),
        miso = gpio_num_t(34),
        dma_channel=0,
        flags={MASTER})

  logi(TAG, "cfg_spi: bus1: %s", repr(BUS1))

  var ADC_SPI = BUS1.addDevice(commandlen = bits(8),
                               addresslen = bits(0),
                               mode = 0,
                               cs_io = gpio_num_t(-1),
                               clock_speed_hz = fast_hz, 
                               queue_size = 1,
                               pre_cb=cs_adc_pre,
                               post_cb=cs_unselect,
                               flags={HALFDUPLEX})

Later these can be used like:

const 
  ADC_READ_MULTI_CMD =  0x80
  ADC_REG_CONFIG0 = 0x03

proc read_regs*(reg: byte, n: range[1..16]): SpiTrans =
  let read_cmd = reg or ADC_READ_MULTI_CMD # does bitwise or
  return ADC_SPI.readTrans(cmd=read_cmd, rxlength=bytes(n), )

proc adc_read_config*(): seq[byte] =
  var trn = read_regs(ADC_REG_CONFIG0, 2)
  trn.transmit() # preforms SPI transaction using transaction queue
  result = trn.getData()

See more in the test SPI Test or the read the wrapper (probably best docs for now): spis.nim.

import nesper, nesper/esp_vfs_fat

var
  base_path : cstring = "/spiflash"
  s_wl_handle : wl_handle_t = WL_INVALID_HANDLE

mount_config = esp_vfs_fat_mount_config_t(format_if_mount_failed: true,
                                          max_files: 10, allocation_unit_size: 4096)
err = esp_vfs_fat_spiflash_mount(base_path, "storage", mount_config.addr, s_wl_handle.addr)

if err != ESP_OK:
  echo "Failed to mount FATFS."
else:
  echo "FATFS mounted successfully!"

writeFile("/spiflash/hello.txt", "Hello world!")
echo readFile("/spiflash/hello.txt") # Hello world!

Notice: file extension of files on FAT filesystem is limited to maximum of 3 characters.

Nim is a flexible language which compiles to a variety of backend "host" languages, including C and C . Like many hosted languages, it has excellent facilities to interact with the host language natively. In the embedded world this means full compatability with pre-existing libraries and toolchains, which are often complex and difficult to interface with from an "external language" like Rust or even C . They often also require oddball compilers, ruling out LLVM based lanugages for many projects (including the ESP32 which defaults to a variant of GCC).

Nim has a few nice features for embedded work:

Language:

• High level language and semantics with low level bit fiddling and pointers
• Flexible garbage collector or manual memory management
  • ARC GC allows using native-C debuggers, meaning any embedded debuggers should work too!
  • ARG GC doesn't use locks, and utilizies move semantics -- it's fast
• Simple FFI's around to import and/or wrap C/C libraries
• Async/Event support
• Real hygenic language macros, and collections with generics!
• Very flexible and hackable standard library!

Libraries:

• Simplified network wrappers around native sockets (i.e. use select w/o a PhD)
• Sane standard library, including JSON, datetime, crypto, ...
  • Efficient compiler that eliminates un-needed code (i.e. json support using a few extra kB's)
  • Package library manager

Compiler:

• Fast compilation, generated C compiles fast
• Deterministic exception handling using a non-malloc friendly goto technique
• Object-oriented like programming that's not based on vtables

There are a few cons of Nim:

• Lack of documentation for many parts of the standard library
• Understanding the differences between stack/heap based objects is a bit tricky
• Compiler options are often incompatible and can require some experimentation
• Small community (e.g. lack of some libraries)
• You likely won't get to use it at XYZ Megacorp
• It will require some pioneering!