dimi - Minimalistic Dependency Injection for Python

“Dependency Injection” is a 25-dollar term for a 5-cent concept.

James Shore

Tired of complex abstractions? Tired of things like "to do dependency injection let's first dive into providers, scopes and binders"?

Just want dead simple syntax for injecting dependencies and mocking them inside tests?

Say no more!

Installation

pip install dimi

Getting Started

from dimi import Container
from typing import Annotated
from flask import Flask, jsonify

# DI container instance stores all the dependencies you place inside
di = Container()


# dependency is a function, async function or a class
# it can be added to the container via the following decorator:

@di.dependency
def get_weather_service_url():
    return 'https://awesome-weather-provider.com/api'


# Dependency may have sub-dependencies defined via typing.Annotated
# For dimi mostly the second part of Annotated matters, the first part is for type checkers

@di.dependency
class WeatherService:
    def __init__(self, url: Annotated[str, get_weather_service_url])
        self.url = url

    def get_weather(self, city): ...


app = Flask(__name__)

# Annotated[MyCls, ...] is the shortcut for Annotated[MyCls, MyCls]

@app.route("/api/weather/<city>")
@di.inject
def get_weather(city: str, weather_service: Annotated[WeatherService, ...]):
    return jsonify(weather_service.get_weather())

The DI container also supports dict-like way of retrieving the dependencies:

weather_service = di[WeatherService]
print(weather_service.get_weather('london'))

OK, but where is the profit?

Someone may ask, "Why not just do this instead?"

# the same API handle but w/o DI

@app.route("/api/weather/<city>")
def get_weather(city: str):
    return jsonify(
        WeatherService('https://awesome-weather-provider.com/api').get_weather(city)
    )

The simplest argument for DI is easier testing.

How would you test the non-DI API handle above? I guess something nasty with monkey patches.

But with DI in action everything becomes much simpler and more obvious:

from myapp import di  # app-wide instance of the Container
from myapp.services import WeatherService


def test_weather_api_handle(test_client):
    class MockWeatherService:
        def get_weather(self, city):
            return {'temperature': 30.5, 'wind_speed': 3.1, 'city': city}

    # override() preserves and restores DI container state after exit from context manager
    with di.override({WeatherService: MockWeatherService}):
        weather = test_client.get('/api/weather/london').json()
        assert weather == {'temperature': 30.5, 'wind_speed': 3.1, 'city': 'london'}

Key Benefits

Simple and concise API:
- Ask for dependencies via Annotated[SomeType, some_callable] type annotation
- Add callables to the DI container via @di.dependency
- Do injections via @di.inject
- override DI contents via di.override()
Explicit dependencies wiring. No magic, no complicated or hidden rules. You define each (sub-)dependency explicitly inside Annotated[]
Auto sub-dependencies resolving. A may depend on B which may depend on C which may depend on D and E and so on. All of this will be correctly tied together and resolved at the time of a function call.
Optional scopes. Just use @di.dependency(scope=Singleton) to cache first call of a function for the lifetime of the app.
Async support
Thread safety

Docs