Chandra Sivaraman
Software Engineering Notes

Interface Segregation Principle31 Mar 2021

Interface Segregation Principle Photo by Paweł Czerwiński on Unsplash

The Interface Segregation Principle (ISP), another Bob Martin creation, says:

No client should be forced to depend on methods it does not use.

Martin advocates splitting fat/general-purpose interfaces into thin/client-specific ones (what Martin Fowler calls role interfaces).3 The idea is to reduce the dependency surface area between object and caller to the minimum required.

ISP vs SRP

Some argue that ISP is just SRP(Single Responsibility Principle) applied to interfaces8. Bob Martin himself clarified the difference between ISP and SRP in a tweet:

ISP can be seen as similar to SRP for interfaces; but it is more than that. ISP generalizes into: “Don’t depend on more than you need.” SRP generalizes to “Gather together things that change for the same reasons and at the same times.”

ISP, thus, is a somewhat stoic principle, that urges consumers (objects, screens, web pages) to practise minimalism, to pare down their dependencies to the bare minimum needed to function. Providers correspondingly are urged to expose interfaces that are tuned to the needs of each client. SRP, on the other hand, calls for a partitioning of code based on different users and their needs. The modules thus partitioned by SRP may be consumed by different clients that each use different parts of the module. ISP ensures that these clients don’t have to know about parts of the module they don’t consume. A difference worth mentioning is that the consumer/client of ISP is a code module whereas the user of SRP is a person(s). Stated another way, ISP is the principle of least knowledge similar to the principle of least privilege in security.

Examples:

  1. A system for realtors has Property objects that represent individual properties. The realtor is the user in this case and the Property class adheres to SRP because it only serves the realtor. The Property class implements an IDetail interface that provides detailed property information for the property detail screen. It also implements an ISummary interface for a master screen that lists properties matching search criteria along with property summary info. The summary screen doesn’t need to know about property details, and therefore shouldn’t be forced to depend on a fat interface that has both detail and summary info.

  2. CRUD(Create-Read-Update-Delete) Repository functionality can be split into IReader and (IReadWriter : IReader) interfaces. Clients that only need to query can use IReader. Clients needing to do both (typically no clients would only want to update) can use IReadWriter.7

  3. A CQRS9 (Command Query Request Segregation) repository splits reads and writes into IQuery and ICommand for query-only clients and clients that need to update the repository.

The polyad vs the monad:

This is an example from Robert Martin’s ISP article1. If a function foo depends on interfaces IA and IB that are part of a container interface IG, then which of these function signatures is preferable:

// polyadic form
void foo(IA a, IB b) 
{
    // call methods on a and b	
}

// monadic form
void foo(IG g) 
{
    a = g.A;
    b = g.B;

    // call methods on a and b
}

According to Martin, the polyadic form is preferable to the monadic form because it ensures that function foo’s dependencies are clearly articulated, and it isn’t saddled with dependencies it doesn’t use. In the monadic example, if unrelated methods are later added to interface IG, then the module containing function foo would need to be recompiled. The less function foo knows about its dependencies, the more decoupled it is from them. Less cohesion between parts results in less friction from change and a more plastic system that is resilient and even welcoming to change.

Finally, some food for thought. What do you think about apps that are built on some least-common denominator framework (like Electron or even the World Wide Web)? Do they violate ISP by providing a general-purpose client that works on all types of devices using a single codebase? Do the clients (the apps themselves) have dependencies that they don’t really need? Would having client-specific apps (native apps built for each client) result in a more decoupled and change-resilient system? Or do the benefits of the least common denominator framework outweigh the costs?

Takeaways

References

  1. ^ The Interface Segregation Principle by Robert Martin
  2. Wikipedia
  3. ^ RoleInterface
  4. Interface-Segregation Principle (ISP) - Principles of Object-Oriented Class Design
  5. What is the distinction between SRP and ISP
  6. https://twitter.com/unclebobmartin/status/996739060348653568
  7. ^ Applying the Interface Segregation Principle to a Repository
  8. ^ SOLID or Colds
  9. ^ CQRS