Functions

Functions should do one thing

This is by far the most important rule in software engineering. When functions do more than one thing, they are harder to compose, test, and reason about. When you can isolate a function to just one action, they can be refactored easily and your code will read much cleaner. If you take nothing else away from this guide other than this, you'll be ahead of many developers.

Bad:

from typing import List


class Client:
    active: bool


def email(client: Client) -> None:
    pass


def email_clients(clients: List[Client]) -> None:
    """Filter active clients and send them an email.
    """
    for client in clients:
        if client.active:
            email(client)

Good:

from typing import List


class Client:
    active: bool


def email(client: Client) -> None:
    pass


def get_active_clients(clients: List[Client]) -> List[Client]:
    """Filter active clients.
    """
    return [client for client in clients if client.active]


def email_clients(clients: List[Client]) -> None:
    """Send an email to a given list of clients.
    """
    for client in get_active_clients(clients):
        email(client)

Do you see an opportunity for using generators now?

Even better

from typing import Generator, Iterator


class Client:
    active: bool


def email(client: Client):
    pass


def active_clients(clients: Iterator[Client]) -> Generator[Client, None, None]:
    """Only active clients"""
    return (client for client in clients if client.active)


def email_client(clients: Iterator[Client]) -> None:
    """Send an email to a given list of clients.
    """
    for client in active_clients(clients):
        email(client)

Function arguments (2 or fewer ideally)

A large amount of parameters is usually the sign that a function is doing too much (has more than one responsibility). Try to decompose it into smaller functions having a reduced set of parameters, ideally less than three.

If the function has a single responsibility, consider if you can bundle some or all parameters into a specialized object that will be passed as an argument to the function. These parameters might be attributes of a single entity that you can represent with a dedicated data structure. You may also be able to reuse this entity elsewhere in your program. The reason why this is a better arrangement is than having multiple parameters is that we may be able to move some computations, done with those parameters inside the function, into methods belonging to the new object, therefore reducing the complexity of the function.

Bad:

def create_menu(title, body, button_text, cancellable):
    pass

Java-esque:

class Menu:
    def __init__(self, config: dict):
        self.title = config["title"]
        self.body = config["body"]
        # ...


menu = Menu(
    {
        "title": "My Menu",
        "body": "Something about my menu",
        "button_text": "OK",
        "cancellable": False
    }
)

Also good

class MenuConfig:
    """A configuration for the Menu.

    Attributes:
        title: The title of the Menu.
        body: The body of the Menu.
        button_text: The text for the button label.
        cancellable: Can it be cancelled?
    """
    title: str
    body: str
    button_text: str
    cancellable: bool = False


def create_menu(config: MenuConfig) -> None:
    title = config.title
    body = config.body
    # ...


config = MenuConfig()
config.title = "My delicious menu"
config.body = "A description of the various items on the menu"
config.button_text = "Order now!"
# The instance attribute overrides the default class attribute.
config.cancellable = True

create_menu(config)

Fancy

from typing import NamedTuple


class MenuConfig(NamedTuple):
    """A configuration for the Menu.

    Attributes:
        title: The title of the Menu.
        body: The body of the Menu.
        button_text: The text for the button label.
        cancellable: Can it be cancelled?
    """
    title: str
    body: str
    button_text: str
    cancellable: bool = False


def create_menu(config: MenuConfig):
    title, body, button_text, cancellable = config
    # ...


create_menu(
    MenuConfig(
        title="My delicious menu",
        body="A description of the various items on the menu",
        button_text="Order now!"
    )
)

Even fancier

from dataclasses import astuple, dataclass


@dataclass
class MenuConfig:
    """A configuration for the Menu.

    Attributes:
        title: The title of the Menu.
        body: The body of the Menu.
        button_text: The text for the button label.
        cancellable: Can it be cancelled?
    """
    title: str
    body: str
    button_text: str
    cancellable: bool = False


def create_menu(config: MenuConfig):
    title, body, button_text, cancellable = astuple(config)
    # ...


create_menu(
    MenuConfig(
        title="My delicious menu",
        body="A description of the various items on the menu",
        button_text="Order now!"
    )
)

Even fancier, Python3.8+ only

from typing import TypedDict


class MenuConfig(TypedDict):
    """A configuration for the Menu.

    Attributes:
        title: The title of the Menu.
        body: The body of the Menu.
        button_text: The text for the button label.
        cancellable: Can it be cancelled?
    """
    title: str
    body: str
    button_text: str
    cancellable: bool


def create_menu(config: MenuConfig):
    title = config["title"]
    # ...


create_menu(
    # You need to supply all the parameters
    MenuConfig(
        title="My delicious menu",
        body="A description of the various items on the menu",
        button_text="Order now!",
        cancellable=True
    )
)

Function names should say what they do

Bad:

class Email:
    def handle(self) -> None:
        pass


message = Email()
# What is this supposed to do again?
message.handle()

Good:

class Email:
    def send(self) -> None:
        """Send this message"""


message = Email()
message.send()

Functions should only be one level of abstraction

When you have more than one level of abstraction, your function is usually doing too much. Splitting up functions leads to reusability and easier testing.

Bad:

# type: ignore

def parse_better_js_alternative(code: str) -> None:
    regexes = [
        # ...
    ]

    statements = code.split('\n')
    tokens = []
    for regex in regexes:
        for statement in statements:
            pass

    ast = []
    for token in tokens:
        pass

    for node in ast:
        pass

Good:

from typing import Tuple, List, Dict

REGEXES: Tuple = (
    # ...
)


def parse_better_js_alternative(code: str) -> None:
    tokens: List = tokenize(code)
    syntax_tree: List = parse(tokens)

    for node in syntax_tree:
        pass


def tokenize(code: str) -> List:
    statements = code.split()
    tokens: List[Dict] = []
    for regex in REGEXES:
        for statement in statements:
            pass

    return tokens


def parse(tokens: List) -> List:
    syntax_tree: List[Dict] = []
    for token in tokens:
        pass

    return syntax_tree

Don't use flags as function parameters

Flags tell your user that this function does more than one thing. Functions should do one thing. Split your functions if they are following different code paths based on a boolean.

Bad:

from tempfile import gettempdir
from pathlib import Path


def create_file(name: str, temp: bool) -> None:
    if temp:
        (Path(gettempdir()) / name).touch()
    else:
        Path(name).touch()

Good:

from tempfile import gettempdir
from pathlib import Path


def create_file(name: str) -> None:
    Path(name).touch()


def create_temp_file(name: str) -> None:
    (Path(gettempdir()) / name).touch()

Avoid side effects

A function produces a side effect if it does anything other than take a value in and return another value or values. For example, a side effect could be writing to a file, modifying some global variable, or accidentally wiring all your money to a stranger.

Now, you do need to have side effects in a program on occasion - for example, like in the previous example, you might need to write to a file. In these cases, you should centralize and indicate where you are incorporating side effects. Don't have several functions and classes that write to a particular file - rather, have one (and only one) service that does it.

The main point is to avoid common pitfalls like sharing state between objects without any structure, using mutable data types that can be written to by anything, or using an instance of a class, and not centralizing where your side effects occur. If you can do this, you will be happier than the vast majority of other programmers.

Bad:

# type: ignore

# This is a module-level name.
# It's good practice to define these as immutable values, such as a string.
# However...
fullname = "Ryan McDermott"


def split_into_first_and_last_name() -> None:
    # The use of the global keyword here is changing the meaning of the
    # the following line. This function is now mutating the module-level
    # state and introducing a side-effect!
    global fullname
    fullname = fullname.split()


split_into_first_and_last_name()

# MyPy will spot the problem, complaining about 'Incompatible types in
# assignment: (expression has type "List[str]", variable has type "str")'
print(fullname)  # ["Ryan", "McDermott"]

# OK. It worked the first time, but what will happen if we call the
# function again?

Good:

from typing import List, AnyStr


def split_into_first_and_last_name(name: AnyStr) -> List[AnyStr]:
    return name.split()


fullname = "Ryan McDermott"
name, surname = split_into_first_and_last_name(fullname)

print(name, surname)  # => Ryan McDermott

Also good

from dataclasses import dataclass


@dataclass
class Person:
    name: str

    @property
    def name_as_first_and_last(self) -> list:
        return self.name.split()


# The reason why we create instances of classes is to manage state!
person = Person("Ryan McDermott")
print(person.name)  # => "Ryan McDermott"
print(person.name_as_first_and_last)  # => ["Ryan", "McDermott"]

Comments