Dependency Injection, as a practice, is meant to introduce abstractions (or seams) to decouple volatile dependencies. A volatile dependency is a class or module that, among other things, can contain nondeterministic behavior or in general is something you which to be able to replace or intercept.
For a more detailed discussion about volatile dependencies, see section 1.3.2 of this freely readable introduction of my book.
Because your FileLogger writes to disk, it contains nondeterministic behavior. For this reason you introduced the ILoggable abstraction. This allows consumers to be decoupled from the FileLogger implementation, and allows you—later on—when such requirement comes in, to easily swap this FileLogger implementation with a SqlLogger implementation that logs to a SQL database, or even have an implementation that forwards the call to both the FileLogger or the SqlLogger.
To be able to successfully decouple a consumer from its volatile dependency, however, you need to inject that dependency into the consumer. There are three common patterns to choose from:
- Constructor Injection—Dependencies are statically defined as list of parameters to the class's instance constructor.
- Property Injection—Dependencies are injected into the consumer via writable instance properties. This pattern is sometime also called 'setter injection', especially in the Java world.
- Method Injection—Dependencies are injected into the consumer as method parameters.
Both Constructor Injection and Property Injection are applied inside the startup path of the application (a.k.a. the Composition Root) and require the consumer to store the dependency in a private field for later reuse. This requires the constructor and property to be instance members, i.e. non-static. Constructor Injection is typically preferred over Property Injection, because Property Injection leads to Temporal Coupling. Static constructors can't have any parameters and static properties lead to the Ambient Context anti-pattern (see section 5.3)—this hinders testability and maintainability.
Method injection, on the other hand, is applied outside the Composition Root and it does not store any supplied dependency, but instead merely uses it. Here's an example from the earlier reference:
// This method calculates the discount based on the logged in user.
// The IUserContext dependency is injected using Method Injection.
public static decimal CalculateDiscountPrice(decimal price, IUserContext context)
{
// Note that IUserContext is never stored - only used.
if (context == null) throw new ArgumentNullException("context");
decimal discount = context.IsInRole(Role.PreferredCustomer) ? .95m : 1;
return price * discount;
}
Method injection is, therefore, the only of the three patterns that can be applied to both instance and static classes.
When applying Method Injection, the method's consumer must supply the dependency. This does mean, however, that the consumer itself must have been supplied with that dependency either through Constructor, Property, or Method Injection. For example:
public class ProductServices : IProductServices
{
private readonly IProductRepository repository;
private readonly IUserContext userContext;
public ProductServices(
IProductRepository repository,
IUserContext userContext) // <-- Dependency applied using Ctor Injection
{
this.repository = repository;
this.userContext = userContext;
}
public decimal CalculateCustomerProductPrice(Guid productId)
{
var product = this.repository.GetById(productId);
return CalculationHelpers.CalculateDiscountPrice(
product.Price,
this.userContext); // <-- Dep forwarded using Method Injection
}
}
Your example of the static LogService that created FileLogger inside its constructor is a great example of tightly coupled code. This is known as the Control Freak anti-pattern (section 5.1) or in general can be seen as a DIP violation—this is the opposite of DI.
To prevent tight coupling of volatile dependencies, the best is to make LogService non-static and inject its volatile dependencies into its sole public constructor:
public class LogService
{
private readonly ILoggable _logger;
public LogService(ILoggable logger)
{
_logger = logger;
}
public void WriteLine(string message) ...
}
This likely defeats the purpose of your LogService class, because now consumers would be better of by injecting ILoggable directly instead of injecting LogService. But this brings you back to the reason why you probably wanted to make that class static in the first place, which is that you have many classes that need to log and it feels cumbersome to inject ILoggable into all those constructors.
This, however, might be caused by another design issue in your code. To understand this, you might want to read through this q&a to get some sense of what design changes you can make that allows less classes to depend on your logger class.
