Dependency Depression
Async-first dependency injection library based on python type hints
Quickstart
First let's create a class we would be injecting:
class Test:
pass
Then we should create instance of container and register our Test class in it, we would use Callable provider that would simply call our class, since classes are also callables!
from dependency_depression import Depression, Callable
container = Depression()
container.register(Test, Callable(Test))
Then we should create a context and resolve our class from it:
with container.sync_context() as ctx:
ctx.resolve(Test)
# < __main__.Test>
Injecting
To mark parameters for injection mark them with typing.Annotated and Inject marker
from typing import Annotated
from dependency_depression import Callable, Depression, Inject
def create_number() -> int:
return 42
def create_str(number: Annotated[int, Inject]) -> str:
return str(number)
container = Depression()
container.register(str, Callable(create_str))
container.register(int, Callable(create_number))
with container.sync_context() as ctx:
string = ctx.resolve(str)
print(string, type(string))
# 42
Providers
When creating a provider you should specify the type it returns, but it can be inferred from class type or function return type:
from dependency_depression import Callable
provider = Callable(int)
# Is the same as Callable(factory=int, impl=int)
assert provider.provide_sync() == 0
Example using factory function, impl is inferred from return type hint:
from dependency_depression import Callable
def create_foo() -> str:
return "foo"
provider = Callable(create_foo)
assert provider.provide_sync() == "foo"
assert provider.impl is str
This all comes into play when you have multiple implementations for base class and want to retrieve individual providers from a container,
let's register two concrete classes under same interface:
from dependency_depression import Depression, Callable
class Base:
pass
class ConcreteA(Base):
pass
class ConcreteB(Base):
pass
container = Depression()
container.register(Base, Callable(ConcreteA))
container.register(Base, Callable(ConcreteB))
with container.sync_context() as ctx:
a = ctx.resolve(Base, ConcreteA) # <__main__.ConcreteA>
b = ctx.resolve(Base, ConcreteB) # <__main__.ConcreteB>
# This would raise an error since we have two classes registered as `Base`
ctx.resolve(Base)
If you have multiple classes registered under same interface you can specify concrete class using Impl marker:
from typing import Annotated
from dependency_depression import Inject, Impl
class Injectee:
def __init__(
self,
a: Annotated[Base, Inject, Impl[ConcreteA]],
b: Annotated[Base, Inject, Impl[ConcreteB]],
):
pass
You can also just register concrete classes instead:
container.register(ConcreteA, Callable(ConcreteA))
container.register(ConcreteB, Callable(ConcreteB))
class Injectee:
def __init__(
self,
a: Annotated[ConcreteA, Inject],
b: Annotated[ConcreteB, Inject],
):
pass
Generics
Dependency Depression can also be used with Generics:
import dataclasses
from typing import Generic, TypeVar, Annotated
from dependency_depression import Inject, Depression, Callable
T = TypeVar("T")
@dataclasses.dataclass
class User:
id: int
username: str
@dataclasses.dataclass
class Item:
id: int
title: str
class IRepository(Generic[T]):
def get(self, identity: int) -> T:
raise NotImplementedError
class UserRepository(IRepository[User]):
def get(self, identity: int) -> User:
return User(id=identity, username="Username")
class ItemRepository(IRepository[Item]):
def get(self, identity: int) -> Item:
return Item(id=identity, title="Title")
class Injectee:
def __init__(
self,
user_repository: Annotated[IRepository[User], Inject],
item_repository: Annotated[IRepository[Item], Inject],
):
self.user_repository = user_repository
self.item_repository = item_repository
container = Depression()
container.register(IRepository[User], Callable(UserRepository))
container.register(IRepository[Item], Callable(ItemRepository))
container.register(Injectee, Callable(Injectee))
with container.sync_context() as ctx:
injectee = ctx.resolve(Injectee)
injectee.user_repository
# < __main__.UserRepository>
injectee.item_repository
# <__main__.ItemRepository>
Context
Context as meant to be used within application or request scope, it keeps instances cache and an ExitStack to close all resources.
Cache
Context keeps cache of all instances, so they won't be created again, unless use_cache=False or NoCache is used.
In this example passing use_cache=False would cause context to create instance of Test again, however it wouldn't be cached:
from dependency_depression import Callable, Depression
class Test:
pass
container = Depression()
container.register(Test, Callable(Test))
with container.sync_context() as ctx:
first = ctx.resolve(Test)
assert first is not ctx.resolve(Test, use_cache=False)
# first is still cached in context
assert first is ctx.resolve(Test)
Closing resources using context managers
Context would also use functions decorated with contextlib.contextmanager or contextlib.asyncontextmanager, but it won't use other instances of ContextManager.
Note that you're not passing impl parameter should specify return type using Iterable, Generator or their async counterparts - AsyncIterableand AsyncGenerator:
import contextlib
from typing import Iterable
from dependency_depression import Depression, Callable
@contextlib.contextmanager
def contextmanager() -> Iterable[int]:
yield 42
class ContextManager:
def __enter__(self):
# This would never be called
raise ValueError
def __exit__(self, exc_type, exc_val, exc_tb):
pass
container = Depression()
# Without return type hint you can specify impl parameter:
# container.register(int, Callable(contextmanager, int))
container.register(int, Callable(contextmanager))
container.register(ContextManager, Callable(ContextManager))
with container.sync_context() as ctx:
number = ctx.resolve(int) # 42
ctx_manager = ctx.resolve(ContextManager) # __enter__ would not be called
with ctx_manager:
...
# Oops, ValueError raised
In case you need to manage lifecycle of your objects you should wrap them in a context manager:
import contextlib
from typing import AsyncGenerator
from dependency_depression import Callable, Depression
from sqlalchemy.ext.asyncio import AsyncSession
@contextlib.asynccontextmanager
async def get_session() -> AsyncGenerator[AsyncSession, None]:
session = AsyncSession()
async with session:
try:
yield session
except Exception:
await session.rollback()
raise
container = Depression()
container.register(AsyncSession, Callable(AsyncSession))
@Inject decorator
@inject decorator allows you to automatically inject parameters into functions:
from typing import Annotated
from dependency_depression import Callable, Depression, Inject, inject
@inject
def injectee(number: Annotated[int, Inject]):
return number
container = Depression()
container.register(int, Callable(int))
with container.sync_context():
print(injectee())
# 0
Without active context number parameter would not be injected:
injectee()
# TypeError: injectee() missing 1 required positional argument: 'number'
But you still can use your function just fine
print(injectee(42))
You can pass parameters even if you have an active context:
with container.sync_context():
print(injectee()) # 0, injected
print(injectee(42)) # 42, provided by user
Usage with Asyncio
Dependency Depression can be used in async context, just use context instead of sync_context:
import asyncio
from dependency_depression import Callable, Depression
async def get_number() -> int:
await asyncio.sleep(1)
return 42
async def main():
container = Depression()
container.register(int, Callable(get_number))
async with container.context() as ctx:
number = await ctx.resolve(int)
assert number == 42
if __name__ == '__main__':
asyncio.run(main())
Async context also supports both sync and async context managers and factory functions.
276 Jan 05, 2023
14 Sep 11, 2022
29 Dec 12, 2022
26 Jun 20, 2022
1 Feb 10, 2022
1 Nov 06, 2021
1 Jan 07, 2022
8 Sep 02, 2022
0 Feb 23, 2022
4 Oct 28, 2022
5 Nov 21, 2021
136 Dec 23, 2022
3 Jan 07, 2022
1 Jan 15, 2022
1 Sep 01, 2022
243 Dec 30, 2022
5 Nov 02, 2022
1 Jan 20, 2022
2 Jan 17, 2022
47 Jan 01, 2023