Pymple - A Simple Inversion of Control Container For Python

Why

If you search for Inversion of Control containers for Python you often encounter the argument "Python is dynamic and does not need those things that static languages need". This is partly true.

Dependency Injection and Inversion of Control is a pattern and not a language feature. It not only makes your code easier to test, but also way more readable. The dependencies are clearly noted in the constructor and your IDEs will give you autocompletion support. If you need to test a class, it is clear where and how to pass in the mocks.

Therefore Inversion and Control and Dependency Injection (which go hand in hand) should also be practiced in dynamic languages. If you think this is not viable, check out AngularJS which also makes use of the above mentioned patterns in a dynamic programming language, namely JavaScript.

For further information watch Google's Clean Code Talks

Limitations

Pymple does currently not support:

• Threadsafety
• Lifetimes

Installation

This library is a Python 3.4+ library. On Python 3.4 however the typings package is required.

Install it via pip for Python 3:

sudo pip3 install pymple

Usage

Pymple knows two types of parameters:

• Singletons: A singleton is a callable that is executed once and the result is saved so future calls to the build method will return the same instance
• Factories: A factory is callable that is executed again every time it is accessed

By default Pymple tries to resolved a singleton based on the annotated type, e.g.:

from pymple import Container

class A:
    def __init__(self):
        pass

class B:
    def __init__(self, param: A):
        self.a = A

container = Container()
b = container.resolve(B)
isinstance(b.a, A) == True

Overriding The Default Behavior

However you can also override it by defining it explicitly:

container = Container()
ccontainer.register(B, lambda c: B('hi'))

b = container.resolve(B)
b.a == 'hi'

The first passed in variable to the lambda is the container instance itself, so you can also resolve other classes on it:

container = Container()
ccontainer.register(B, lambda c: B(c.resolve(A)))

b = container.resolve(B)
isinstance(b.a, A) == True

Registering Factories

If you want to register a factory instead of a singleton, simple pass False as the second parameter:

container = Container()
ccontainer.register(B, lambda c: B('hi'), False)

b = container.resolve(B)
c = container.resolve(B)
b != c

Aliasing

Sometimes a type interface uses an abstract class as type annotation. In that case you can simply define an alias:

container = Container()
ccontainer.alias(ConcreteClass, AbstractClass)

clazz = container.resolve(AbstractClass)
isinstance(clazz, ConcreteClass) == True