Depend on abstractions.1)
A simplified description of DIP is that a variable declaration should always have the (static) type of an abstract class or
interface. By doing so a module depends only on this abstraction. The concrete subclass realizing the details is referenced only once, namely when it is instantiated.
The more elaborate definition by Robert C. Martin reads as follows:
a. High-level modules should not depend on low-level modules. Both should depend on abstractions.
b. Abstractions should not depend on details. Details should depend on abstractions.2)
Following this rule leads to “inverted” dependencies compared to classical procedural approaches. The following diagram shows the classical approach. A high-level module
A uses a low-level module
When applying DIP, both modules depend on the abstraction (note that in UML diagrams all arrows point into the direction of the dependency):
B is not depended upon anymore but it depends on another module. This is the inverted dependency.
When DIP is not applied, only the low-level modules can be reused independently. The higher-level modules depend on the others, so trying to reuse them makes it necessary to either also reuse the lower-level modules or to change the higher-level module. The former is often not wanted because reuse is often done in another context where the lower-level modules do not fit. And the latter is error-prone and requires additional work as it requires changes to already working modules.
interfacetype for every class
interfacetypes so that an object variable generally has an
interfaceas static type and a concrete class as dynamic type
It is normally not helpful to apply DIP to value objects.
See section contrary principles.
An example for a high-level module is a regulator module of a furnace. The classical approach would result in the regulator depending on a thermometer and a heater. in such a case it would not be possible to reuse the regulator module for regulating the fluid level of a reservoir or the speed of a car. A DIP-compliant solution would result in the regulator just depending on a sensor module and an actuator module and thermometer and header implementing these
interfaces. By doing so thermometer, heater, and regulator can be reused independently.
This example is taken from 3) and slightly modified.
Let's say the high-level module (your business logic), wants to be able to add or remove users to the database. Instead of it talking to the database directly, it defines an interface called ClientRepository which contains the methods the business logic needs. Then a MySQLClientRepository concretion, implements that interface and uses a database library to submit the queries. Since the interface is decided by the business logic, the high-level policy is protected from changes in the database library. More over, since the interface was defined by the business logic, it does not reveal anything about the underlying implementation, which allows different types of user repositories, such as a WebserviceClientRepository implementation (OCP). Finally the MySQLClientRepository and business logic can be built as well as deployed independently.
Robert C. Martin: ButUncleBob -- Principles of OOD
Discuss this wiki article and the principle on the corresponding talk page.