Static methods, static classes, utility classes¶
- A utility class MUST contain only static methods
- Utility classes MUST NOT have state, no local properties, no DB access, …
- Utility methods MAY call other utility methods
- Utility class methods MUST NOT have dependencies to non static methods like other class instances or global variables
- Utility class methods MUST have high unit test coverage
- Utility class scope MUST be small and domain logic MUST NOT be encapsulated in static methods
- Utility classes MUST be located in a utility sub folder and MUST end
- Static methods MUST be located in utility classes and SHOULD NOT be added to other classes, except a specific pattern has a hard requirement to a static helper method within its class. All classes outside a utility folder MUST be instantiated and handled as object instances.
Static methods as cross-cutting concern solution have been in the core ever since. They are an easy way to extract recurring coding problems to helper methods.
Static methods however have a list of issues that need to be taken into consideration before deciding to use them. First, they can not be extended in a sane way and the core framework has no way to re-route a static method call to a different implementation. They are a hard coded dependency in a system. They can not be easily “mocked away” in unit tests if a class uses a static method from a different class. They especially raise issues if a static method keeps state in static properties, this is similar to a de-facto singleton and it is hard to reset or manipulate this state later. Static properties can easily result in side effects to different using systems. Additionally, static methods tend to become too complex, doing too much at a time and becoming god methods in long run. Big and complex utility methods doing too much at a time is a strong sign something else was not modeled properly at a different place.
The core has a long history of static utility class misuse and is in an ongoing effort to model stuff correctly and getting rid of static utility god classes that happily mix different concerns. Solving some of these utility methods to proper class structures typically improves code separation significantly and renders core parts more flexible and less error prone.
With this history in mind, core development is rather sensible when new static utility classes should be added. During reviews, a heavy introduction of static classes or methods should raise red lights, it is likely some abstraction went wrong and the problem domain was not modeled well enough.
A “good” static method in a utility class can be thought of as if the code itself is directly embedded within the consuming classes. It is mostly an extraction of a common programming problem that can not be abstracted within the class hierarchy since multiple different class hierarchies have the same challenge. Good static methods contain helper code like dedicated array manipulations or string operations. This is why the majority of static utility classes is located within the core extension in the core and other extension have little number of utility classes.
Good static methods calls are not “mocked away” in unit tests of a system that calls a static method and are thus indirectly tested together with the system under test as if the code is directly embedded within the class. It is important to have good test coverage for the static method itself, defining the method behaviour especially for edge cases.
- Clear scope - array manipulation helpers
- Well documented, distinct and short methods doing only one thing at a time with decent names and examples
- High test coverage taking care of edge case input output scenarios acting as additional documentation of the system
- No further dependencies
- Clear scope - a group of helper methods to process version number handling
- Good test coverage defining the edge cases
- Defines how version handling is done in TYPO3 and encapsulates this concern well
- Global access, third party dependencies
- Stateful methods
- No clear concern
- God methods
- Good: Relatively clear focus, but:
- Stateful, external dependencies to objects, depends on configuration
- Relatively inflexible
- This should probably “at least” be a service
- Not static
- Should probably be a dedicated class construct, probably a service is not enough. Why is this not part of a tree structure?
$GLOBALS: Utilitycode should not have dependencies to global state or global objects.