What Makes Code Testable?
Core aspects
- Object-Oriented: Although both procedural and functional programming paradigms support testing, managing test complexity can become a challenge as the system scales. As complexity increases, maintaining test isolation becomes more difficult. An effective strategy is to use object-oriented techniques that encapsulate complex logic within hierarchies of classes, creating testable units that hide their internal complexity behind interfaces. Wrapping procedural or functional code in classes can combine the strengths of several paradigms. Any code base that reaches a certain level of complexity can benefit from an object-oriented approach to improve testability.
- Classes implement interfaces: The class implements a well-designed interface, just as large as necessary to satisfy its requirements.
- Tests run against interfaces, so the implementations details are not relevant as long as the results are correct.
- Minimal number of dependencies in a unit for low complexity.
- Dependencies are injectable via constructor injection or setter injection. Injectable is equivalent to mockable. Mocks make testing with dependencies much easier. IoC containers are good tools for building many applications consisting of classes into which dependencies need to be injected.
- Dependencies are interfaces that can be implemented as 'mocks' for testing purposes, thus enabling the independent, concurrent development of software units. This aligns with the Dependency Inversion Principle (DIP), which promotes reliance on abstractions, not concretions.
Additional Details
- Avoid direct instantiation of objects in units being tested; doing so tightly couples the implementation and restricts substitutability. Use factories and dependency injections instead.
- Favor dependency injection over service lookups.
- Avoid reflections undermining encapsulation.
- Avoid method chaining, which can make it difficult to replace module dependencies.
- Avoid using final and static methods, as they cannot be overridden and reduce testability. This is especially important for methods that may require mocking.
- Keep private methods simple. Their functionality should be tested indirectly through the public methods that call them, not directly.
- No business logic in constructors and static initialization blocks. Unnecessary logic could be inadvertently executed, or unexpected side effects could occur when a subclass calls the superclass constructor. Instead, place this logic in methods or other objects that can be replaced.
- For third-party libraries and immutable, untestable code, create intermediate wrappers with testable interfaces for better mockability.
- Modular design:
- Create distinct, small modules, each with unique functionality and responsibilities. This ensures that changes in one module are isolated from others.
- Maintain loose coupling: Minimize the dependencies between modules.
- Develop self-contained modules: Each module should be a self-contained unit of functionality. New features should be introduced by creating new modules.
- Avoid the singleton design pattern because its private constructor prevents mocking. Consider a getInstance() method that returns an interface type instead.
- Favor composition over inheritance. Use inheritance only for polymorphism, as it is less flexible than composition. Composition should be the primary approach to code reuse.
Testable Architecture
An important part of a testable architecture is the Humble Object Pattern:
- The pattern emphasizes the isolation of complex to test aspects (such as GUI interactions), keeping them 'humble' with minimal logic.
- Minimize complex layers: Keep complex layers such as GUIs and adapters as thin as possible, focusing mainly on user interaction and routing.
- Maximize business logic: Where possible, move logic to the business logic/service layer, which is easier to test, improving overall testability.