MicroPython has created a great foundation to build upon and to make it even better for beginners we've created CircuitPython. This guide covers a number of ways the core and libraries are geared towards beginners.

Cookiecutter is a cool tool that lets you bootstrap a new repo based on another repo. We've made one here for CircuitPython libraries that include configs for Travis CI and ReadTheDocs along with a setup.py, license, code of conduct and readme.

Adafruit funded libraries should be under the adafruit organization and have the format Adafruit_CircuitPython_<name> and have a corresponding adafruit_<name> directory (aka package) or adafruit_<name>.py file (aka module).

Community created libraries should have the format CircuitPython_<name> and not have the adafruit_ module or package prefix.

Both should have the CircuitPython repository topic on GitHub.

A driver should be initialized and ready to use after construction. If the device requires deinitialization, then provide it through deinit() and also provide __enter__ and __exit__ to create a context manager usable with with.

For example, a user can then use deinit()`:

import digitalio
import board

led = digitalio.DigitalInOut(board.D13)
led.direction = digitalio.Direction.OUTPUT

for i in range(10):
    led.value = True
    time.sleep(0.5)

    led.value = False
    time.sleep(0.5)
led.deinit()

This will deinit the underlying hardware at the end of the program as long as no exceptions occur.

Alternatively, using a with statement ensures that the hardware is deinitialized:

import digitalio
import board

with digitalio.DigitalInOut(board.D13) as led:
    led.direction = digitalio.Direction.OUTPUT

    for i in range(10):
        led.value = True
        time.sleep(0.5)

        led.value = False
        time.sleep(0.5)

Python's with statement ensures that the deinit code is run regardless of whether the code within the with statement executes without exceptions.

For small programs like the examples this isn't a major concern because all user usable hardware is reset after programs are run or the REPL is run. However, for more complex programs that may use hardware intermittently and may also handle exceptions on their own, deinitializing the hardware using a with statement will ensure hardware isn't enabled longer than needed.

Whenever possible, make sure device you are talking to is the device you expect. If not, raise a ValueError. Beware that I2C addresses can be identical on different devices so read registers you know to make sure they match your expectation. Validating this upfront will help catch mistakes.

When designing a driver for a device, use properties for device state and use methods for sequences of abstract actions that the device performs. State is a property of the device as a whole that exists regardless of what the code is doing. This includes things like temperature, time, sound, light and the state of a switch. For a more complete list see the sensor properties bullet below.

Another way to separate state from actions is that state is usually something the user can sense themselves by sight or feel for example. Actions are something the user can watch. The device does this and then this.

Making this separation clear to the user will help beginners understand when to use what.

Here is more info on properties from Python.

CircuitPython is aimed to be one's first experience with code. It will be the first step into the world of hardware and software. To ease one's exploration out from this first step, make sure that functionality shared with CPython shares the same API. It doesn't need to be the full API it can be a subset. However, do not add non-CPython APIs to the same modules. Instead, use separate non-CPython modules to add extra functionality. By distinguishing API boundaries at modules you increase the likelihood that incorrect expectations are found on import and not randomly during runtime.

When adding extra functionality to CircuitPython to mimic what a normal operating system would do, either copy an existing CPython API (for example file writing) or create a separate module to achieve what you want. For example, mounting and unmount drives is not a part of CPython so it should be done in a module, such as a new storage module, that is only available in CircuitPython. That way when someone moves the code to CPython they know what parts need to be adapted.

Whenever possible, document your code right next to the code that implements it. This makes it more likely to stay up to date with the implementation itself. Use Sphinx's automodule to format these all nicely in ReadTheDocs. The cookiecutter helps set these up.

Use Sphinx flavor rST for markup.

Lots of documentation is a good thing but it can take a lot of space. To minimize the space used on disk and on load, distribute the library as both .py and .mpy, MicroPython and CircuitPython's bytecode format that omits comments.

After the license comment:

"""
`<module name>` - <Short description>
=================================================
<Longer description.>
"""

Documenting what the object does:

class DS3231:
    """Interface to the DS3231 RTC."""

Renders as:

.. py:class:: DS3231
  :noindex:

  Interface to the DS3231 RTC.

Comment is after even though its weird:

lost_power = i2c_bit.RWBit(0x0f, 7)
"""True if the device has lost power since the time was set."""

Renders as:

.. py:attribute:: lost_power
  :noindex:

  True if the device has lost power since the time was set.

First line after the method definition:

def turn_right(self, degrees):
    """Turns the bot ``degrees`` right.

       :param float degrees: Degrees to turn right
    """

Renders as:

.. py:method:: turn_right(degrees)
  :noindex:

  Turns the bot ``degrees`` right.

  :param float degrees: Degrees to turn right

Comment comes from the getter:

@property
def datetime(self):
    """The current date and time"""
    return self.datetime_register

@datetime.setter
def datetime(self, value):
    pass

Renders as:

.. py:attribute:: datetime
  :noindex:

  The current date and time

[BusDevice](https://github.com/adafruit/Adafruit_CircuitPython_BusDevice) is an awesome foundational library that manages talking on a shared I2C or SPI device for you. The devices manage locking which ensures that a transfer is done as a single unit despite CircuitPython internals and, in the future, other Python threads. For I2C, the device also manages the device address. The SPI device, manages baudrate settings, chip select line and extra post-transaction clock cycles.

from adafruit_bus_device import i2c_device

DEVICE_DEFAULT_I2C_ADDR = 0x42

class Widget:
    """A generic widget."""

    def __init__(self, i2c, address=DEVICE_DEFAULT_I2C_ADDR):
        self.i2c_device = i2c_device.I2CDevice(i2c, address)
        self.buf = bytearray(1)

    @property
    def register(self):
        """Widget's one register."""
        with self.i2c_device as i2c:
            i2c.writeto(b'0x00')
            i2c.readfrom_into(self.buf)
        return self.buf[0]

from adafruit_bus_device import spi_device

class SPIWidget:
    """A generic widget with a weird baudrate."""

    def __init__(self, spi, chip_select):
        # chip_select is a pin reference such as board.D10.
        self.spi_device = spi_device.SPIDevice(spi, chip_select, baudrate=12345)
        self.buf = bytearray(1)

    @property
    def register(self):
        """Widget's one register."""
        with self.spi_device as spi:
            spi.write(b'0x00')
            i2c.readinto(self.buf)
        return self.buf[0]

When writing a driver, take in objects that provide the functionality you need rather than taking their arguments and constructing them yourself or subclassing a parent class with functionality. This technique is known as composition and leads to code that is more flexible and testable than traditional inheritance.

.. seealso:: `Wikipedia <https://en.wikipedia.org/wiki/Dependency_inversion_principle>`_
  has more information on "dependency inversion".

For example, if you are writing a driver for an I2C device, then take in an I2C object instead of the pins themselves. This allows the calling code to provide any object with the appropriate methods such as an I2C expansion board.

Another example is to expect a DigitalInOut for a pin to toggle instead of a microcontroller.Pin from board. Taking in the ~microcontroller.Pin object alone would limit the driver to pins on the actual microcontroller instead of pins provided by another driver such as an IO expander.

CircuitPython boards tend to have a small amount of internal flash and a small amount of ram but large amounts of external flash for the file system. So, create many small libraries that can be loaded as needed instead of one large file that does everything.

Speed isn't as important as API clarity and code size. So, prefer simple APIs like properties for state even if it sacrifices a bit of speed.

Although Python doesn't require managing memory, its still a good practice for library writers to think about memory allocations. Avoid them in drivers if you can because you never know how much something will be called. Fewer allocations means less time spent cleaning up. So, where you can, prefer bytearray buffers that are created in __init__ and used throughout the object with methods that read or write into the buffer instead of creating new objects. Unified hardware API classes such as busio.SPI are design to read and write to subsections of buffers.

Its ok to allocate an object to return to the user. Just beware of causing more than one allocation per call due to internal logic.

However, this is a memory tradeoff so do not do it for large or rarely used buffers.

Use ustruct.pack_into instead of ustruct.pack.

The Adafruit Unified Sensor Driver Arduino library has a great list of measurements and their units. Use the same ones including the property name itself so that drivers can be used interchangeably when they have the same properties.

Outside of sensors, having common methods amongst drivers for similar devices such as devices can be really useful. Its early days however. For now, try to adhere to guidelines in this document. Once a design is settled on, add it as a subsection to this one.

The Python API for a new module should be defined and documented in shared-bindings and define an underlying C API. If the implementation is port-agnostic or relies on underlying APIs of another module, the code should live in shared-module. If it is port specific then it should live in common-hal within the port's folder. In either case, the file and folder structure should mimic the structure in shared-bindings.

Keeping compatibility with MicroPython isn't a high priority. It should be done when its not in conflict with any of the above goals.