Dependency Inversion Principle (DIP)

Clean Architecture by Robert Martin states, "Any source code dependency, no matter where it is can be inverted." This is known as the Dependency Inversion Principle (DIP), which is illustrated in the first diagram. Martin argues that this is the primary benefit of Object-Oriented Programming.

He states, "It allows the architect to create a plugin architecture, in which modules that contain high-level policies are independent of modules that contain low-level details."

DIP Diagram from Clean Architecture

Robotics Example

To illustrate this point, we can create two fictional classes:

• PathFollower: A high-level class for directing overall robot motion.
• PIDController: A mid-level class for controlling wheel velocity.

A naive implementation of how we might design these classes is shown in the second diagram.

Naive Implemention (without dependency inversion):

The dependency-inverted implementation is shown in the third diagram. Notice that PathFollower does not inherit from PIDController.

Dependency Inverted Implementation

A few consequences of this inversion are:

• Any controller that adheres to the interface (an abstract class) defined by ControllerInterface can be swapped in with minimal effort. For instance, MPCController can be linked to our executable (instead of PIDController) without any changes to the existing source code. We just compile MPCController as a shared library and modify the linking behavior at compile-time via CMake arguments.
• main and PathFollower do not need to be recompiled if PIDController changes. main will just be linked with the recompiled PIDController.
• As promised above, "high-level policies (PathFollower) are independent of modules that contain low-level details (PIDController).
• Both modules (PathFollower and PIDController) depend on abstractions, not implementation details.

C++ and Python examples of both the Naive and DIP implementations were built using the references linked below.

C++

In C++, an abstract class with virtual methods defines an interface (ControllerInterface) which specifies a contract that all concrete implementations of that abstract class (PIDController) must meet. A concrete implementation (ControllerImpl) of the abstract class is declared and promised within controller_interface.h. The actual definition of the concrete implementation is provided via linked library at compile-time.

To compile:

cd dependency-inversion
mkdir build && cd build
# Set cmake arguments as you desire
cmake .. -DLINK_PID=OFF -DLINK_MPC=ON
make

To run the naive example:

cd dependency-inversion/build
./naive_approach

To run the dependency inverted example:

cd dependency-inversion/build
./di_approach

Python

In Python, an abstract (ABC) class with @abstractmethod methods defines an interface (ControllerInterface) which specifies a contract that all concrete implementations of that abstract class (PIDController) must meet.

To run the naive example:

cd dependency-inversion/python
python naive.py

To run the dependency inverted example:

cd dependency-inversion/python
python di_approach.py

References

• C++
  • http://codingadventures.org/2021/10/16/dependency-inversion-in-c/
  • https://github.com/plusangel/DependencyInversion_pattern_cpp
• Python
  • https://youtu.be/Kv5jhbSkqLE
  • https://github.com/ArjanCodes/betterpython/tree/main/2%20-%20dependency%20inversion