Design Guide

This guide covers a variety of development practices for CircuitPython core and library APIs. These APIs are both built-into CircuitPython and those that are distributed on GitHub and in the Adafruit and Community bundles. Consistency with these practices ensures that beginners can learn a pattern once and apply it throughout the CircuitPython ecosystem.

Start libraries with the cookiecutter

Cookiecutter is a 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, license, code of conduct, readme among other files.

Cookiecutter will provide a series of prompts relating to the library and then create a new directory with all of the files. See the CircuitPython cookiecutter README for more details.

Module Naming

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).

If the name would normally have a space, such as “Thermal Printer”, use an underscore instead (“Thermal_Printer”). This underscore will be used everywhere even when the separation between “adafruit” and “circuitpython” is done with a -. Use the underscore in the cookiecutter prompts.

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

Both should have the CircuitPython repository topic on GitHub.


As our Code of Conduct states, we strive to use “welcoming and inclusive language.” Whether it is in documentation or in code, the words we use matter. This means we disfavor language that due to historical and social context can make community members and potential community members feel unwelcome.

There are specific terms to avoid except where technical limitations require it. While specific cases may call for other terms, consider using these suggested terms first:



Main (device)





Secondary (device)





Note that “technical limitations” refers e.g., to the situation where an upstream library or URL has to contain those substrings in order to work. However, when it comes to documentation and the names of parameters and properties in CircuitPython, we will use alternate terms even if this breaks tradition with past practice.

Lifetime and ContextManagers

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
import time

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

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

    led.value = False

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
import time

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

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

        led.value = False

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.

Verify your device

Whenever possible, make sure the device you are talking to is the device you expect. If not, raise a RuntimeError. 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.

Exceptions and asserts

Raise an appropriate Exception, along with a useful message, whenever a critical test or other condition fails.


if not 0 <= pin <= 7:
    raise ValueError("Pin number must be 0-7.")

If memory is constrained and a more compact method is needed, use The assert statement instead.


assert 0 <= pin <= 7, "Pin number must be 0-7."

Design for compatibility with CPython

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 a new module for additional functionality related to a CPython module do NOT simply prefix it with u. This is not a large enough differentiation from CPython. This is the MicroPython convention and they use u* modules interchangeably with the CPython name. This is confusing. Instead, think up a new name that is related to the extra functionality you are adding.

For example, storage mounting and unmounting related functions were moved from uos into a new storage module. These names better match their functionality and do not conflict with CPython names. Make sure to check that you don’t conflict with CPython libraries too. That way we can port the API to CPython in the future.


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 unmounting 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.

Document inline

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.

Module description

After the license comment:

`<module name>`

<Longer description>

* Author(s):

Implementation Notes


* `Adafruit Device Description
  <hyperlink>`_ (Product ID: <Product Number>)

**Software and Dependencies:**

* Adafruit CircuitPython firmware for the supported boards:

* Adafruit's Bus Device library:

* Adafruit's Register library:


Version description

After the import statements:

__version__ = "0.0.0+auto.0"
__repo__ = "<repo github link>"

Class description

At the class level document what class does and how to initialize it:

class DS3231:
    """DS3231 real-time clock.

       :param ~busio.I2C i2c_bus: The I2C bus the DS3231 is connected to.
       :param int address: The I2C address of the device. Defaults to :const:`0x40`

    def __init__(self, i2c_bus, address=0x40):
        self._i2c = i2c_bus

Renders as:

class DS3231(i2c_bus, address=64)

DS3231 real-time clock.

  • i2c_bus (I2C) – The I2C bus the DS3231 is connected to.

  • address (int) – The I2C address of the device. Defaults to 0x40

Documenting Parameters

Although there are different ways to document class and functions definitions in Python, the following is the prevalent method of documenting parameters for CircuitPython libraries. When documenting class parameters you should use the following structure:

:param param_type param_name: Parameter_description


The type of the parameter. This could be, among others, int, float, str, bool, etc. To document an object in the CircuitPython domain, you need to include a ~ before the definition as shown in the following example:

:param ~busio.I2C i2c_bus: The I2C bus the DS3231 is connected to.

To include references to CircuitPython modules, cookiecutter creates an entry in the intersphinx_mapping section in the file located within the docs directory. To add different types outside CircuitPython you need to include them in the intersphinx_mapping:

intersphinx_mapping = {
    "python": ("", None),
    "BusDevice":("", None,),
    "CircuitPython": ("", None),

The intersphinx_mapping above includes references to Python, BusDevice and CircuitPython Documentation

When the parameter have two different types, you should reference them as follows:

class Character_LCD:
    """Base class for character LCD

       :param ~digitalio.DigitalInOut rs: The reset data line
       :param ~pwmio.PWMOut,~digitalio.DigitalInOut blue: Blue RGB Anode


    def __init__(self, rs, blue):
        self._rc = rs = blue

Renders as:

class Character_LCD(rs, blue)

Base class for character LCD



Parameter name used in the class or method definition


Parameter description. When the parameter defaults to a particular value, it is good practice to include the default:

:param int pitch: Pitch value for the servo. Defaults to :const:`4500`


Attributes are state on objects. (See Getters/Setters above for more discussion about when to use them.) They can be defined internally in a number of different ways. Each approach is enumerated below with an explanation of where the comment goes.

Regardless of how the attribute is implemented, it should have a short description of what state it represents including the type, possible values and/or units. It should be marked as (read-only) or (write-only) at the end of the first line for attributes that are not both readable and writable.

Instance attributes

Comment comes from after the assignment:

def __init__(self, drive_mode):
    self.drive_mode = drive_mode
    The pin drive mode. One of:

    - `digitalio.DriveMode.PUSH_PULL`
    - `digitalio.DriveMode.OPEN_DRAIN`

Renders as:


The pin drive mode. One of:

Property description

Comment comes from the getter:

def datetime(self):
    """The current date and time as a `time.struct_time`."""
    return self.datetime_register

def datetime(self, value):

Renders as:


The current date and time as a time.struct_time.

Read-only example:

def temperature(self):
    The current temperature in degrees Celsius. (read-only)

    The device may require calibration to get accurate readings.
    return self._read(TEMPERATURE)

Renders as:


The current temperature in degrees Celsius. (read-only)

The device may require calibration to get accurate readings.

Data descriptor description

Comment is after the definition:

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

Renders as:


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

Method description

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:


Turns the bot degrees right.


degrees (float) – Degrees to turn right

Documentation References to other Libraries

When you need to make references to documentation in other libraries you should refer the class using single backticks :class:`~adafruit_motor.servo.Servo`. You must also add the reference in the file in the intersphinx_mapping section by adding a new entry:

"adafruit_motor": ("", None,),

Use adafruit_register when possible

Register is a foundational library that manages packing and unpacking data from I2C device registers. There is also Register SPI for SPI devices. When possible, use one of these libraries for unpacking and packing registers. This ensures the packing code is shared amongst all registers (even across drivers). Furthermore, it simplifies device definitions by making them declarative (only data.)

Values with non-consecutive bits in a register or that represent FIFO endpoints may not map well to existing register classes. In unique cases like these, it is ok to read and write the register directly.

Do not add all registers from a datasheet upfront. Instead, only add the ones necessary for the functionality the driver exposes. Adding them all will lead to unnecessary file size and API clutter. See this video about outside-in design from @tannewt.

I2C Example

from adafruit_register import i2c_bit
from adafruit_bus_device import i2c_device

class HelloWorldDevice:
    """Device with two bits to control when the words 'hello' and 'world' are lit."""

    hello = i2c_bit.RWBit(0x0, 0x0)
    """Bit to indicate if hello is lit."""

    world = i2c_bit.RWBit(0x1, 0x0)
    """Bit to indicate if world is lit."""

    def __init__(self, i2c, device_address=0x0):
        self.i2c_device = i2c_device.I2CDevice(i2c, device_address)

Use BusDevice

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.

I2C Example

from adafruit_bus_device import i2c_device


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)

    def register(self):
        """Widget's one register."""
        with self.i2c_device as i2c:
        return self.buf[0]

SPI Example

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)

    def register(self):
        """Widget's one register."""
        with self.spi_device as spi:
        return self.buf[0]

Class documentation example template

When documenting classes, you should use the following template to illustrate basic usage. It is similar with the simpletest example, however this will display the information in the Read The Docs documentation. The advantage of using this template is it makes the documentation consistent across the libraries.

This is an example for a AHT20 temperature sensor. Include the following after the class parameter:


**Quickstart: Importing and using the AHT10/AHT20 temperature sensor**

    Here is an example of using the :class:`AHTx0` class.
    First you will need to import the libraries to use the sensor

    .. code-block:: python

        import board
        import adafruit_ahtx0

    Once this is done you can define your `board.I2C` object and define your sensor object

    .. code-block:: python

        i2c = board.I2C()  # uses board.SCL and board.SDA
        aht = adafruit_ahtx0.AHTx0(i2c)

    Now you have access to the temperature and humidity using
    the :attr:`temperature` and :attr:`relative_humidity` attributes

    .. code-block:: python

        temperature = aht.temperature
        relative_humidity = aht.relative_humidity


Use composition

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.

See also

Wikipedia 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 Pin from board. Taking in the 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.

Lots of small modules

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 second

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.

Avoid allocations in drivers

Although Python doesn’t require managing memory, it’s 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.

It’s 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 struct.pack_into instead of struct.pack.

Use of MicroPython const()

The MicroPython const() feature, as discussed in this forum post, and in this issue thread, provides some optimizations that can be useful on smaller, memory constrained devices. However, when using const(), keep in mind these general guide lines:

  • Always use via an import, ex: from micropython import const

  • Limit use to global (module level) variables only.

  • Only used when the user will not need access to variable and prefix name with a leading underscore, ex: _SOME_CONST.


from adafruit_bus_device import i2c_device
from micropython import const

_DEFAULT_I2C_ADDR = const(0x42)

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

    def __init__(self, i2c, address=_DEFAULT_I2C_ADDR):
        self.i2c_device = i2c_device.I2CDevice(i2c, address)

Libraries Examples

When adding examples, cookiecutter will add a <name> file in the examples directory for you. Be sure to include code with the library minimal functionalities to work on a device. You could other examples if needed featuring different functionalities of the library. If you add additional examples, be sure to include them in the examples.rst. Naming of the examples files should use the name of the library followed by a description, using underscore to separate them.

Sensor properties and units

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.

Property name

Python type



(float, float, float)

x, y, z meter per second per second


(float, float, float)

x, y, z micro-Tesla (uT)


(float, float, float)

x, y, z degrees


(float, float, float)

x, y, z radians per second



degrees Celsius



measured CO2 in ppm



equivalent/estimated CO2 in ppm (estimated from some other measurement)



Total Volatile Organic Compounds in ppb



centimeters (cm)



non-unit-specific proximity values (monotonic but not actual distance)



non-unit-specific light levels (should be monotonic but is not lux)



SI lux



hectopascal (hPa)






milliamps (mA)



volts (V)



RGB, eight bits per channel (0xff0000 is red)


(time.struct, str)

Sample alarm time and string to characterize frequency such as “hourly”



date and time



16-bit PWM duty cycle (regardless of output resolution)



Hertz (Hz)



Digital logic



16-bit Analog value, unit-less



grams (g)



non-unit-specific sound level (monotonic but not actual decibels)

Driver constant naming

When adding variables for constant values for a driver. Do not include the device’s name in the variable name. For example, in, variables should not start with FANCY123_. Adding this prefix increases RAM usage and .mpy file size because variable names are preserved. User code should refer to these constants as adafruit_fancy123.HELLO_WORLD for clarity. adafruit_fancy123.FANCY123_HELLO_WORLD would be overly verbose.

Adding native modules

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.

To test your native modules or core enhancements, follow these Adafruit Learning Guides for building local firmware to flash onto your device(s):

Build CircuitPython

MicroPython compatibility

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