This might be an unpopular opinion, but we might be doing more harm than good with interfaces in object-oriented programming languages.
Let me explain.
Interfaces 101
First, disambiguation time: when I’m talking about interfaces here, I’m referring to interface definitions in code, not user interfaces, user experience, or anything of a graphical nature.
Put flatly, an interface is a contract that says a class will have certain characteristics – primarily public methods and properties – that other components can interact with.
The intent of an interface is to provide a programming language some degree of decoupling. For example, if you have a class that needs to read some external data, you might have it take in an IDomainObjectDataProvider instead of a SqlServerDomainObjectDataProvider.
That way your class wouldn’t need to care if it was talking to data in memory, in a database, or provided by some external API call.
This makes sense and its the classical reason for having interfaces.
Another reason might be a class in one layer not having references to a class defined in another layer. In that sense, interfaces can provide a certain degree of indirection. I like this reasoning less, but it can be valid in cases.
So What’s Wrong With Interfaces?
Interfaces on their own are not bad, but they do have some significant tradeoffs and I’m not convinced that developers fully consider the tradeoffs of using them.
Navigation Woes
First of all, using an interface makes it hard to navigate through code.
If I’m in my development environment of choice, most will support control clicking or some keyboard shortcut to navigate to the definition of a type.
With concrete types, navigation will take you directly to the implementation of the method you’re most interested in. With interfaces, you’ll navigate to the interface’s definition of that method.
This might not sound like much, but if you’re “in the flow” and thinking through a process, this is akin to coming around a corner and finding a solid wall where you expect a door. You have to make sense of what you’re seeing, then figure out which concrete type(s) you’re actually working with, find them, and find the relevant definition.
The penalty of this to our productivity is small, but significant, and this is a scenario that happens frequently in interface rich environments.
Obfuscation via Interface
Let’s go back to the definition of an interface as a contract and the earlier example of an interface for a specific sort of data provider.
True, it’s very nice to have our code flexible and decoupled from specific implementations.
But if I’m looking at a class and seeing an interface, I can sometimes lose track of the runtime specifics.
Let’s say we only have one type of IEmalSender in an application, for example. If all I see when navigating code is an IEmailSender reference, I may lose track of what sender we’re actually working with in production and some of the specifics of its implementation.
Some may argue that this is a good thing and that I shouldn’t have to care, and they’re right – to an extent – but the problem comes when we think so much in terms of abstraction that seeing the concrete deployment scenario becomes difficult.
Architectural Cement
I like to think of interfaces as a sort of “architectural cement” in software development.
What I mean by this is that if I’m doing some refactoring (cleaning up the form of code without changing its behavior) and I find I no longer need to pass a certain parameter, or I want to make a method synchronous that was asynchronous (or vice versa), or any number of minor tweaks, interfaces make this harder.
Instead of changing something in one place, I have to navigate to the interface and change it there as well. If there were any other implementations of that interface, I need to seek them out and make sure they’re changed as well.
This means that an operation that might have been trivial to do in passing now takes me out of my natural flow and requires some additional degree of effort and thought to carry out. It might not be a lot, but it’s enough to make me think twice.
Additionally, if a member of an interface is never used, it’s much harder to detect that with code analysis tools than if a method that doesn’t adhere to an interface would be. This means that dead code that’s part of an interface definition stays around longer.
My point here is that we pay for interfaces during the maintenance of our software in the form of little inconveniences.
It’s not a lot, but it’s more than you think, and the more interfaces you use, the more pronounced the problem becomes.
Interface Segregation Principle
Another major issue I see with interfaces are violations of the Interface Segregation Principle (ISP). ISP is part of SOLID programming principles which are 5 principles to produce maintainable software over time.
Specifically, ISP talks about preferring many smaller interfaces around specialized tasks to one larger interface designed for a class that does many general things.
This principle frequently gets violated when developers add interfaces to an existing system. Typically they’ll go into a class and extract an interface for all public members, then replace usages of the class with usages of the interface.
It’s somewhat straightforward and easy to do and so the path of least resistance leads to giant interfaces such as IUserRepository instead of smaller interfaces like IUserValidator and IUserCreator.
There are a number of problems with these larger interfaces including:
- They frequently demonstrate the problems listed in earlier sections
- They make it hard to make new implementations due to the number of members that are part of the interface
- They tend to be the only concrete implementation of that interface
- It tends to promote classes that don’t adhere to the Single Responsibility Principle (another tenant of SOLID)
All told, large interfaces are a bad idea and tend to lead to maintenance headaches in the long term.
Inheritance vs Interfaces
So, if I’m cautioning on interfaces, what am I suggesting might be a better alternative?
Frequently when systems need a degree of flexibility in implementation, they don’t need complete flexibility like an interface provides. Often they just need a base class that can serve as a bit of a mini-contract for purposes of dependency injection or testing.
Because of that, I advocate that any time you think about adding an interface, you consider instead if introducing or using an existing base class might be a better fit.
Some advantages that base classes can provide:
- Navigation to a base class may actually navigate to a concrete or default implementation of a relevant method
- Base classes provide a degree of code reuse / sharing that is not possible via interfaces
- Base classes are slightly easier to refactor than interfaces
Of course, there are disadvantages and tradeoffs to consider:
- Code in your base class will be present in any derived class unless overridden, which can constrain implementations too much
- You don’t always control enough of the code to make base classes a viable option, or layer dependencies make this impossible
- This can lead to too much “depth of inheritance” if there was already inheritance going on in your class hierarchy.
So, it’s a bit of a tradeoff as far as whether you use base classes or interfaces.
In general, I like to use interfaces for very small capabilities and tend to prefer base classes for things like configuring inversion of control containers.
Closing Thoughts
Your preferences are going to match your needs. All I ask is that you don’t just automatically assume “This should be an interface” or “This should be a base class” or even “I shouldn’t pass a concrete class into this method”.
Whether you optimize for flexibility, maintenance, rapid development, or something else is entirely up to you.
Everything has advantages and tradeoffs, and software engineering is about finding the right mix for your codebase.
The post Death by Interfaces? appeared first on Kill All Defects.
Top comments (17)
Yea in my experience Base classes has been horrendous. Because ppl walk all over Liskov substitution principle. First, they have an inheritance in 20 levels and somewhere in that chain they simply don't implement all methods, they become empty or have NotImplementedException as only code line.. This is way worse than fat interfaces imo.. We just need to keep fighting this in code reviews in my opinion, don't just mindlessly extract interfaces but defend why it's needed. Or if base classes is your go to then cap the limit of 2 levels of inheritance before it spirals out of control.
One benefit to use Interface is it provides good support for writing mock tests and today’s dependency injection is tied to interfaces. These are the reasons for their increased usage in today’s applications. 100% agree to your thoughts.
I agree. I thought of this when writing my article but forgot to work it into the base class vs interface discussion - many mocking frameworks can work with base classes instead of interfaces and some can even work with concrete classes.
There are so many things wrong in the article, I don't even know where to start.
Advising against using interfaces and using the base abstract class? Are we going back to the 90s?
I think in general, interfaces are the right choice over base classes. However, putting an interface on every class you write isn't necessary. Only add them when they provide value for testing, DI, or architecture.
But the tooling has improved enough to quickly go to the implementation of the interface. In VS, you can use CTRL + F12.
If we're misusing interfaces - which we are - the questions are how should we use them, how are we using them incorrectly, and perhaps when are we using them without needing them?
Death by.... We're doing more harm than good with ... We can fill in the blanks with pretty much anything.
The underlying problem I see behind the abuse of interfaces is that we don't understand what abstractions are for. We don't design abstractions and then implement them. We build whatever classes we're going to build and keep them wrapped in 1:1 corresponding interfaces. We get the benefit of mocking but not much else.
I get more benefit from interfaces when I use them to describe what another class must depend on, from the perspective of that class. That leads to small, cohesive interfaces.
It also leads to writing the code I need, as opposed to creating dependencies first and then creating their consumers around them.
This discussion probably ties in well with unit testing. Imo, creating fine grained unit tests is a waste of time and a maintenance nightmare.
Here the interfaces come in very handy, we test against the interfaces only and we get decent code coverage and a good handle on code quality without going overboard.
There is certainly a place for interfaces in any good design. However, I completely agree that they are overused and that we should think twice about whether we really need them when we add them.
Sure, this goes against "accepted wisdom" and does not sit well for a lot of people. But, it's good to question what we do and why we do it. Cargo cult driven development is a very easy habit to slip into!
In particular, I absolutely hate introducing an interface just to be able to write some unit tests for some component. I avoid doing that whenever possible.
I think I remember someone blogging or tweeting to say "every time I add an interface to my code I feel like I've failed". I might be wrong but I think was Paul Stovell aka Mr Octopus Deploy.
Thanks for a great article!
would like to share a few awesome videos regarding this subject matter:
i personally try to avoid di, mocking and interfaces as much as possible and only use them when absolutely necessary.
as long as my intergration tests are on point then maintenance and evolving my apps is a breeze.
Fascinating article Matt sand really interesting points.
I love interfaces and think the decoupling they bring are a huge plus to refactoring without worry.
That said, I do massively agree with your point on legibility. Being able to jump to the definition of an object is really useful, with interfaces you lose that.
I find myself jumping between interfaces, the DI creation (startup.cs) for example and actual definitions which can be a pain.
I try to minimise interfaces where possible, only using them where I think there is a true need for easy substitution (database interactions being a primary candidate).
Interface does not define the implementation so change naturally is less often as it's more generic. I think the case when one specific implementation effect the interface used in many other places is in most wrong approach.
Interface defines behavior of some group of objects then change of the interface should effect all, and it is a good thing.
If we have that one member is different from others then we need to think if the member fits the group, maybe not, maybe group was wrongly defined.
My biggest issue with class inheritance is method overload where we can have multiple level of inheritance with overloaded implementations. That means that probably our group is totally broken and the original class has nothing to it's members.
Well, I like your opinion about design implications when you implement interfaces. There is no silver bullet when you are building a system architecture, only implications. All styles, paradigms, frameworks, languages, all of them have good and bad implications.
The real metrics about good design are a system cheap to maintain, build, test and deploy.
I think you hit the nail on the head with my central point. This piece struck a nerve and got dog piled on, but what I really want is for people to recognize the tradeoffs we make and the prices of things. Use what works, but make conscious choices about it.
Stop putting capital Is in front of your interface name. I don't care whether my type is from interface definition or a base class. This just clutters code. If you need to emphasise, that this type is an interface, you are already on the track of being doomed.
🤣
The article doesn't make me too hopeful, but maybe some time in Rust or Go would alter his opinion. I'm certainly mildly curious about his experience with either language and how he might approach architecture there.
i disagree, what about test?
I disagree with almost every point in the article. But we don't need to be all the same right :) Thanks for writing it down.