django-boot-rhazes 0.3.0

A Django DI manager

Rhazes

A Dependency Injection and IoC container library for Django Framework.

Versions and Requirements

Written for Django 4.2 using python 3.9.

Other python versions (3.6+) should also be supported. It may work with other Django versions as well (but requires changes being applied to setup.cfg).

How it works

Rhazes works by scanning for bean (AKA services) classes available in the project. Ideally to make the scan faster, you shall define the packages you want to scan in a configuration inside settings.py (explained in usage section).

Scanning for bean classes works by creating graphs of bean classes and their dependencies and choose random nodes to do DFS traversal in order to find edges and possible dependency cycles. Then from edges back to to top, there will be builder functions created and registered in ApplicationContext (which is a static class or a singleton) for a specific bean class or interface. A builder is a function that accepts ApplicationContext and returns an instance of a bean class (possibly singleton instance). The builder can use ApplicationContext to access other dependent beans, and this is only possible since we cover dependency graph edges first and then go up in the graph.

Eventually, all bean classes will have a builder registered in ApplicationContext. You can directly ask ApplicationContext for a bean instance of specific type, or you can use @inject decorator so they are automatically injected into your classes/functions.

Usage and Example

Let's assume we have bean classes like below in a package named app1.services:

from abc import ABC, abstractmethod
from rhazes.decorator import bean


class UserStorage(ABC):

  @abstractmethod
  def get_user(user_id: int):
    pass


@bean(_for=UserStorage, primary=False)  # primary is False by default too
class DatabaseUserStorage(UserStorage):

  def get_user(user_id: int):
    return None


@bean(_for=UserStorage, primary=True)  # set as primary implementation of UserStorage
class CacheUserStorage(UserStorage):

  def get_user(user_id: int):
    return None


@bean()
class ProductManager:

  def __init__(self, user_storage: UserStorage):
    self.user_storage = user_storage

  def get_user_products(user_id):
    user = self.user_storage.get_user(user_id)
    # Do something to find products of user?

To make Rhazes scan these packages, we should define RHAZES_PACKAGES in our settings.py:

RHAZES_PACKAGES = [
  "app1.services"
]

Now assuming you have the above classes defined user some packages that will be scanned by Rhazes, you can access them like this:

from rhazes.context import ApplicationContext
from somepackage import UserStorage, DatabaseUserStorage, CacheUserStorage,  ProductManager


application_context = ApplicationContext
# scan packages at settings.INSTALLED_APPS or settings.RHAZES_PACKAGES
application_context.initialize()

# Get ProductManager bean using its class
product_manager: ProductManager = application_context.get_bean(ProductManager)

# Get UserStorage (interface) bean
# this will be CacheUserStorage implementation since primary was set to true
user_storage: UserStorage = application_context.get_bean(UserStorage)

# Specifically get beans of classes (not the interface)
cache_user_storage: CacheUserStorage = application_context.get_bean(CacheUserStorage)  # to directly get CacheUserStorage
database_user_storage: DatabaseUserStorage = application_context.get_bean(DatabaseUserStorage)  # to directly get DatabaseUserStorage

Important note: in order for Rhazes to understand the dependency of the beans and be able to graph it you have to use type hints.

This means that Rhazes fails to understand the type of dependency (service_a) of such bean:

@bean
class ServiceB:
  def __init__(self, service_a):  # Rhazes can't guess what should be injected here
    pass

As a general rule, use type hints anywhere you expect Rhazes to do something for you!

Bean factory

Bean factories are just classes that produce a bean. They are beans themselves!

from rhazes.protocol import BeanFactory

@bean
class SomeBeanFactory(BeanFactory):

    # optional: if you haven't defined "_for" in @bean, you can determine it here
    @classmethod
    def produces(cls):
        return SomeBean

    def produce(self):
        return SomeBean()

Singleton

You can define beans as singleton.

@bean(singleton=True)
class SomeBean:
    pass

At this point this bean will always be the same instance when being injected into another class (another bean or @inject (read further))

Lazy Bean Dependencies

If the bean you are defining is depended on another bean but you don't want to immediately instantiate that other bean you can mark it as lazy.

@bean
class DependencyA:
    pass


@bean(lazy_dependencies=[DependencyA])
class DependencyB:
    def __int__(self, dependency_a: DependencyA):
        self.dependency_a = dependency_a

Now dependency_a will not be instantiated (built) until there is a call to it from inside DependencyB instances.

Injection

You can inject beans into functions or classes as long as your function (or class __init__ function) has good support for **kwargs.

These classes or functions need to be called with determined input parameter names. Example:

@bean
class SomeBean:
    pass


@inject()
def function(bean: SomeBean, random_input: str):
    ...

# You can call it like this:
function(random_input="something")  # `bean` will be injected automatically

Example for classes:

@bean
class SomeBean:
    pass


@inject
class MyClazz:
    def __init__(self, bean: SomeBean, random_input: str):
        ...

MyClazz(random_input="something")  # `bean` will be injected automatically

To explicitly inject some beans and not others:

@bean
class SomeBean1:
    pass


@bean
class SomeBean2:
    pass


@inject(injections=[SomeBean1])
def function(bean1: SomeBean1, bean2: SomeBean2, random_input: str):
    ...

# You can call it like this:
function(bean2=SomeBean2(), random_input="something")  # `bean1` will be injected automatically

Inject into Django views

At this stage only injection into class views are tested. Example:

@inject()
class NameGeneratorView(APIView):
    # You could optionally use @inject() here or at .setup()
    def __init__(self, string_generator: StringGeneratorService, **kwargs):
        self.string_generator = string_generator
        super(NameGeneratorView, self).__init__(**kwargs)

    def get(self, request, *args, **kwargs):
        qs: dict = request.GET.dict()
        return Response(data={
            "name": self.string_generator.generate(
                int(qs.get("length", 10))
            )
        })

This example is taken from here.

When to initialize ApplicationContext

Application Context can be initialized either in a .ready() method of an app in your Django project, or in main urls.py.

Dependency Cycles

Dependency cycles are detected during Application Context initialization and will raise error. So you cant have beans like below:

@bean
class ServiceA:
  def __init__(self, service_b: "ServiceB"):
    pass


@bean
class ServiceB:
  def __init__(self, service_a: "ServiceA"):
    pass

Override beans for tests

In case you need to override a bean class in a test case method you can use one of the following ways:

from rhazes.test.context import TemporaryContext, TemporaryContextManager

# Way 1: using TemporaryContext
temporary_context = TemporaryContext()
mock = Mock()
temporary_context.register_bean(SomeInterface, mock)
# your other code here
# Call this in the end:
temporary_context.reset()


# Way 2
with TemporaryContextManager() as temporary_context:
    mock = Mock()
    temporary_context.register_bean(SomeInterface, mock)
    # your other tests here

Contribution

Read the contribution guidelines.