System Design (LLD)Role-Based Interface Design

Role-Based Interface Design

LevelIntermediate

Duration60 mins

TopicRole-Based Interface Design

1 / 4

Interfaces Representing Roles

The Role Revolution in Interface Design

When most developers think about interfaces, they think about APIs—a collection of methods that define what an object can do. This implementation-centric view leads to a natural but problematic tendency: creating interfaces that mirror class implementations rather than expressing meaningful contracts.

The Interface Segregation Principle invites us to a paradigm shift: stop thinking of interfaces as implementation contracts and start thinking of them as roles. This seemingly subtle change in perspective transforms how we design systems, dramatically improving flexibility, testability, and maintainability.

What You Will Learn

By the end of this page, you will understand how to conceptualize interfaces as roles rather than implementation blueprints. You'll learn why this mental model is central to ISP, how it prevents fat interfaces before they form, and how role-based thinking enables the compositional flexibility that distinguishes expert object-oriented design from merely functional code.

The Conventional View of Interfaces

To appreciate role-based interface design, we must first understand what it replaces. The conventional view of interfaces treats them primarily as abstraction mechanisms for implementations. Under this view, we create interfaces by:

Designing a class with all its functionality
Extracting an interface that exposes that functionality
Using the interface wherever we want to abstract away the concrete class

This approach seems logical—after all, we're "programming to interfaces, not implementations," right? But there's a critical flaw in this reasoning.

conventional-interface-design.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
// The Conventional Approach: Interface mirrors implementation
 
// Step 1: Design the class with all its functionality
class UserService {
    // Authentication methods
    login(email: string, password: string): User { /* ... */ }
    logout(userId: string): void { /* ... */ }
    resetPassword(email: string): void { /* ... */ }
    changePassword(userId: string, newPassword: string): void { /* ... */ }
    
    // Profile management methods
    getProfile(userId: string): UserProfile { /* ... */ }
    updateProfile(userId: string, profile: UserProfile): void { /* ... */ }
    uploadAvatar(userId: string, image: Buffer): string { /* ... */ }
    
    // User administration methods
    listAllUsers(): User[] { /* ... */ }
    banUser(userId: string): void { /* ... */ }
    deleteUser(userId: string): void { /* ... */ }
    promoteToAdmin(userId: string): void { /* ... */ }
    
    // Notification methods
    sendWelcomeEmail(userId: string): void { /* ... */ }
    sendPasswordResetEmail(email: string): void { /* ... */ }
    notifyLoginFromNewDevice(userId: string, device: DeviceInfo): void { /* ... */ }
}
 
// Step 2: Extract an interface that "abstracts" this class
interface IUserService {
    login(email: string, password: string): User;
    logout(userId: string): void;
    resetPassword(email: string): void;
    changePassword(userId: string, newPassword: string): void;
    getProfile(userId: string): UserProfile;
    updateProfile(userId: string, profile: UserProfile): void;
    uploadAvatar(userId: string, image: Buffer): string;
    listAllUsers(): User[];
    banUser(userId: string): void;
    deleteUser(userId: string): void;
    promoteToAdmin(userId: string): void;
    sendWelcomeEmail(userId: string): void;
    sendPasswordResetEmail(email: string): void;
    notifyLoginFromNewDevice(userId: string, device: DeviceInfo): void;
}
 
// Step 3: Use the interface everywhere
class LoginController {
    constructor(private userService: IUserService) {}
    
    handleLogin(email: string, password: string) {
        // Uses only 2 of 14 methods!
        const user = this.userService.login(email, password);
        this.userService.notifyLoginFromNewDevice(user.id, this.getDeviceInfo());
        return user;
    }
}

The problem is now visible: The LoginController depends on IUserService, but it only uses 2 of the 14 methods. This means:

Unnecessary coupling: LoginController is coupled to concepts it doesn't care about (profile management, admin operations, avatar uploads)
Testing overhead: Mocking IUserService for LoginController tests requires stubbing all 14 methods
Change ripple effects: Adding a method to handle user subscriptions forces recompilation of LoginController, even though it has nothing to do with subscriptions

The fundamental mistake? We designed the interface around the implementation, not around the clients' needs.

The Implementation-Mirroring Trap

When interfaces are derived by "extracting" from implementations, they inevitably inherit the structure and scope of those implementations. This creates fat interfaces that force clients to depend on methods they never use—the precise violation ISP warns against. The solution isn't to add more discipline to extraction; it's to change how we think about interfaces entirely.

Interfaces as Roles — The Paradigm Shift

Role-based interface design inverts the conventional approach. Instead of asking "What does this class do?" and extracting an interface, we ask "What role does this object play in this particular context?" and define interfaces from the client's perspective.

A role is a coherent set of related responsibilities that a client expects from a collaborator. Roles are:

Context-specific: The same class might play different roles in different contexts
Client-defined: Roles are discovered by examining what clients actually need, not what implementations provide
Minimal: A role includes only what's necessary for that specific collaboration
Semantic: Roles carry meaning about why something is needed, not just what is provided

Let's revisit our example with role-based thinking:

role-based-interface-design.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
// Role-Based Approach: Interfaces defined by what clients need
 
// Role: Something that can authenticate users
interface Authenticator {
    login(email: string, password: string): User;
    logout(userId: string): void;
}
 
// Role: Something that can manage passwords
interface PasswordManager {
    resetPassword(email: string): void;
    changePassword(userId: string, newPassword: string): void;
}
 
// Role: Something that provides user profile access
interface ProfileProvider {
    getProfile(userId: string): UserProfile;
    updateProfile(userId: string, profile: UserProfile): void;
}
 
// Role: Something that can manage avatars
interface AvatarManager {
    uploadAvatar(userId: string, image: Buffer): string;
}
 
// Role: Something with administrative powers over users
interface UserAdministrator {
    listAllUsers(): User[];
    banUser(userId: string): void;
    deleteUser(userId: string): void;
    promoteToAdmin(userId: string): void;
}
 
// Role: Something that can send user notifications
interface UserNotifier {
    sendWelcomeEmail(userId: string): void;
    sendPasswordResetEmail(email: string): void;
    notifyLoginFromNewDevice(userId: string, device: DeviceInfo): void;
}
 
// The implementation can fulfill multiple roles
class UserService implements 
    Authenticator, 
    PasswordManager, 
    ProfileProvider, 
    AvatarManager, 
    UserAdministrator, 
    UserNotifier {
    // Implementation of all methods...
}
 
// Now LoginController depends only on the roles it needs!
class LoginController {
    constructor(
        private authenticator: Authenticator,
        private notifier: UserNotifier
    ) {}
    
    handleLogin(email: string, password: string) {
        const user = this.authenticator.login(email, password);
        this.notifier.notifyLoginFromNewDevice(user.id, this.getDeviceInfo());
        return user;
    }
}

Notice the transformation:

Each interface represents a coherent role with a clear name that expresses its purpose
Interfaces are small and focused — typically 1-4 methods
Clients depend only on the roles they need — LoginController takes an Authenticator and UserNotifier
The class fulfills multiple roles by implementing multiple interfaces
Testing is simpler — mock only the 2 interfaces with 5 total methods, not 14

This is the essence of role-based interface design: interfaces are discovered from client needs, not extracted from implementations.

The Direction of Discovery

In role-based design, the 'arrows' of discovery point from clients toward providers. We don't start with an implementation and ask 'What interface should I extract?' We start with clients and ask 'What role do I need to fill this collaboration?' This client-first thinking prevents fat interfaces from ever forming.

Roles in Real-World Systems

Role-based interface design isn't just a theoretical nicety—it's how well-designed real-world systems are structured. Let's examine how this pattern appears across different domains.

Example 1: The File System Abstraction

A file on disk can play many roles. Different clients need different subsets of its capabilities:

file-system-roles.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
// Different clients need different views of a "file"
 
// Role: Something whose content can be read
interface Readable {
    read(buffer: Buffer, offset: number, length: number): number;
}
 
// Role: Something whose content can be written
interface Writable {
    write(buffer: Buffer, offset: number, length: number): number;
}
 
// Role: Something that can be positioned
interface Seekable {
    seek(position: number, whence: SeekMode): number;
    tell(): number;
}
 
// Role: Something that can be closed/released
interface Closeable {
    close(): void;
}
 
// Role: Something with metadata
interface FileMetadata {
    readonly size: number;
    readonly createdAt: Date;
    readonly modifiedAt: Date;
    readonly permissions: FilePermissions;
}
 
// A file descriptor in the OS implements multiple roles
class FileDescriptor implements 
    Readable, 
    Writable, 
    Seekable, 
    Closeable, 
    FileMetadata {
    // Full implementation...
}
 
// A log writer only needs Writable + Closeable
class LogWriter {
    constructor(
        private output: Writable & Closeable
    ) {}
    
    log(message: string): void {
        const buffer = Buffer.from(message + '
');
        this.output.write(buffer, 0, buffer.length);
    }
    
    finish(): void {
        this.output.close();
    }
}
 
// A file searcher only needs Readable + Seekable
class FileSearcher {
    constructor(
        private file: Readable & Seekable
    ) {}
    
    findPattern(pattern: Buffer): number | null {
        // Can read and seek, doesn't need to write or know metadata
    }
}
 
// A backup system needs FileMetadata + Readable
class BackupManager {
    constructor(
        private source: FileMetadata & Readable
    ) {}
    
    shouldBackup(): boolean {
        // Uses metadata.modifiedAt
        return this.source.modifiedAt > this.lastBackupTime;
    }
}

Example 2: E-Commerce Order System

An Order object participates in many different business processes. Each process views the order through a different role:

order-roles.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
// Role: Something that can be priced
interface Priceable {
    readonly subtotal: Money;
    readonly tax: Money;
    readonly total: Money;
    applyDiscount(discount: Discount): void;
}
 
// Role: Something that can be shipped
interface Shippable {
    readonly shippingAddress: Address;
    readonly weight: Weight;
    readonly dimensions: Dimensions;
    calculateShippingCost(carrier: Carrier): Money;
}
 
// Role: Something that can be paid for
interface Payable {
    readonly paymentDue: Money;
    recordPayment(payment: Payment): void;
    readonly isFullyPaid: boolean;
}
 
// Role: Something with line items
interface LineItemContainer {
    readonly items: ReadonlyArray<LineItem>;
    addItem(product: Product, quantity: number): void;
    removeItem(itemId: string): void;
}
 
// Role: Something with order lifecycle
interface OrderLifecycle {
    readonly status: OrderStatus;
    submit(): void;
    cancel(reason: string): void;
    refund(): void;
}
 
// Role: Something that can be tracked by customer
interface CustomerTrackable {
    readonly orderId: string;
    readonly estimatedDelivery: Date;
    readonly trackingNumber: string | null;
    readonly statusHistory: ReadonlyArray<StatusEvent>;
}
 
// The Order aggregate implements all these roles
class Order implements 
    Priceable, 
    Shippable, 
    Payable, 
    LineItemContainer, 
    OrderLifecycle, 
    CustomerTrackable {
    // Full implementation...
}
 
// Payment processor only sees Payable
class PaymentProcessor {
    process(payable: Payable, paymentMethod: PaymentMethod): PaymentResult {
        // Works with any payable thing, not just Orders
        // Could also process invoices, subscriptions, etc.
    }
}
 
// Shipping calculator only sees Shippable
class ShippingCalculator {
    calculateOptions(shippable: Shippable): ShippingOption[] {
        // Works with any shippable thing
        // Doesn't know or care about payment status
    }
}
 
// Customer portal only sees CustomerTrackable
class CustomerOrderView {
    constructor(private order: CustomerTrackable) {}
    
    render(): CustomerViewDTO {
        // Can only access tracking info
        // Cannot cancel, modify, or see internal order details
    }
}

Roles Cross-Cut Types

Notice that roles like 'Priceable' or 'Shippable' can apply to many different types, not just Order. An Invoice might be Priceable. A Product might be Shippable. By defining interfaces around roles rather than implementations, we enable polymorphism across unrelated types that share behaviors—one of the most powerful aspects of role-based design.

The Linguistic Power of Roles

Role-based interfaces do more than improve coupling—they enhance communication. Well-named role interfaces form a ubiquitous language that makes code self-documenting and discussions more precise.

Compare these two method signatures:

Implementation-Centric

•sendNotification(userService: IUserService)
•What can userService do? Everything a user service does.
•Why are we passing a user service to send notifications?
•Would a mock UserService work here?
•Meaning is obscured by scope

Role-Centric

•sendNotification(notifier: Notifier)
•What can notifier do? Notify users.
•Why? Because we need to send notifications.
•Any object that can notify will work.
•Meaning is expressed in the type

The naming of roles carries semantic weight. Consider these role names and what they communicate:

Role Name	Semantic Meaning
`Authenticator`	Validates identity, establishes sessions
`Authorizer`	Checks permissions, enforces access control
`Encryptor`	Transforms data into secure, unreadable form
`Validator`	Checks data against rules, reports violations
`Transformer`	Converts data from one form to another
`Publisher`	Broadcasts events to interested parties
`Subscriber`	Receives and reacts to published events
`Repository`	Provides collection-like access to persistent objects
`Factory`	Creates complex objects without exposing construction logic
`Observer`	Watches for changes, reacts when notified

Each name is a promise about behavior, not a description of implementation. This is why role-based interfaces are often named with "-er" or "-or" suffixes—they describe what something does, not what something is.

Role Naming Heuristics

Good role names typically: (1) Use agent nouns (Validator, not IValidation), (2) Describe capability, not structure (Readable, not IHasReadMethod), (3) Are context-appropriate (UserAuthenticator vs just Authenticator when disambiguation is needed), and (4) Avoid implementation hints (Notifier, not EmailNotifier—that's for the class name).

Role Composition — The Multiplicative Power

One of the most powerful consequences of role-based interface design is role composition. When interfaces are small and focused, clients can combine them to express exactly what they need—no more, no less.

This is enabled by intersection types (in TypeScript), multiple interface inheritance (in Java/C#), or protocols (in Swift). The ability to compose roles multiplicatively expands our design vocabulary.

role-composition.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
// Individual roles
interface Identifiable {
    readonly id: string;
}
 
interface Timestamped {
    readonly createdAt: Date;
    readonly updatedAt: Date;
}
 
interface Versionable {
    readonly version: number;
    incrementVersion(): void;
}
 
interface SoftDeletable {
    readonly isDeleted: boolean;
    softDelete(): void;
    restore(): void;
}
 
interface Auditable {
    readonly createdBy: UserId;
    readonly lastModifiedBy: UserId;
    recordModification(userId: UserId): void;
}
 
// Composed roles for specific needs
type Trackable = Identifiable & Timestamped;
type VersionedEntity = Identifiable & Versionable;
type AuditedEntity = Identifiable & Timestamped & Auditable;
type FullyManaged = Identifiable & Timestamped & Versionable & SoftDeletable & Auditable;
 
// Repository that works with any identifiable, timestamped entity
interface Repository<T extends Trackable> {
    findById(id: string): T | null;
    findByCreatedAfter(date: Date): T[];
    save(entity: T): void;
}
 
// Audit service works with anything auditable
class AuditService {
    recordChange<T extends AuditedEntity>(entity: T, userId: UserId): void {
        entity.recordModification(userId);
        this.auditLog.append({
            entityId: entity.id,
            timestamp: new Date(),
            user: userId,
            previousVersion: entity.version
        });
    }
}
 
// Soft delete handler works with anything soft-deletable and identifiable
class SoftDeleteHandler {
    delete<T extends Identifiable & SoftDeletable>(entity: T): void {
        entity.softDelete();
        this.eventBus.publish(new EntitySoftDeleted(entity.id));
    }
    
    restore<T extends Identifiable & SoftDeletable>(entity: T): void {
        entity.restore();
        this.eventBus.publish(new EntityRestored(entity.id));
    }
}

The mathematics of role composition:

With 5 orthogonal role interfaces (Identifiable, Timestamped, Versionable, SoftDeletable, Auditable), we can express 32 different combinations (2^5). Each combination represents a distinct set of capabilities that a client might require.

With a single fat interface that combines all these capabilities, we have 1 option: all or nothing.

This multiplicative power is why role-based design scales so well. As systems grow, the number of ways to compose existing roles grows exponentially, while the number of new interfaces needed grows only linearly.

Role Composition Scales Better
Approach	10 Capabilities	20 Capabilities	30 Capabilities
Fat Interfaces	1 way to use	1 way to use	1 way to use
Role Composition	1,024 combinations	~1 million combinations	~1 billion combinations

The Combinatorial Advantage

Role-based interfaces buy us exponential flexibility with linear effort. Each new role interface multiplies the expressiveness of our type vocabulary. This is why experienced designers prefer many small interfaces over few large ones—the composition possibilities compound.

Roles Enable True Polymorphism

Polymorphism, at its core, means "many forms"—the ability to treat different objects uniformly based on shared behavior. Role-based interfaces enable genuine polymorphism across objects that may have nothing else in common.

Consider: What do these have in common?

A File on disk
An HttpResponse from a server
A DatabaseResultSet from a query
A GeneratorFunction that yields values
A MockDataSource in a test

They're completely different types with different implementations. But they can all be Readable—something that provides data when asked. This shared role enables polymorphic code:

true-polymorphism.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
// Role interface for anything that can provide data
interface DataSource {
    read(): AsyncIterable<Buffer>;
    readonly metadata: SourceMetadata;
}
 
// All these unrelated types implement the same role
class FileSource implements DataSource { /* ... */ }
class HttpResponseSource implements DataSource { /* ... */ }
class DatabaseSource implements DataSource { /* ... */ }
class InMemorySource implements DataSource { /* ... */ }
class S3BucketSource implements DataSource { /* ... */ }
 
// Polymorphic processor works with ANY data source
class DataProcessor {
    // This single method works with 5+ completely different types!
    async process(source: DataSource): Promise<ProcessedResult> {
        const results: ProcessedChunk[] = [];
        
        for await (const chunk of source.read()) {
            results.push(this.processChunk(chunk));
        }
        
        return {
            metadata: source.metadata,
            chunks: results,
            processedAt: new Date()
        };
    }
}
 
// Usage: same processor, different sources
const processor = new DataProcessor();
 
// Process a local file
const fileResult = await processor.process(new FileSource('/data/input.csv'));
 
// Process an HTTP response
const httpResult = await processor.process(new HttpResponseSource(apiUrl));
 
// Process database results
const dbResult = await processor.process(new DatabaseSource(query));
 
// Process test data in unit tests
const testResult = await processor.process(new InMemorySource(mockData));
 
// Process data from S3
const s3Result = await processor.process(new S3BucketSource(bucketName, key));

This is polymorphism working as intended:

One algorithm (DataProcessor.process()) works with many types
Types don't need a common base class—they share a role
New sources can be added without modifying the processor
Testing is trivial: provide an InMemorySource with controlled data

Without role-based interfaces, we'd be forced into awkward alternatives:

Giant switch statements checking types at runtime
Adapter classes wrapping every type into a common base
Tight coupling where the processor knows about every source type
Duplicated logic with separate processing methods per type

Role-based interfaces give us polymorphism without inheritance, composition without boilerplate.

Duck Typing Done Right

Role-based interfaces formalize the intuition behind 'duck typing': if it walks like a duck and quacks like a duck, we can treat it as a duck. Role interfaces make the duck's expected behavior explicit and type-safe, capturing the benefits of duck typing without the runtime surprises.

Roles vs. Base Classes — The Critical Distinction

Some developers wonder: "Why not just use abstract base classes for shared behavior?" The answer reveals a critical distinction between role-based design and inheritance-based design.

Base classes express what something is. A Dog extends Mammal because a dog is a mammal. This is an ontological claim about identity and shared characteristics.

Interfaces express what something can do. A Dog implements Runnable because a dog can run. This is a behavioral claim about capabilities that may be shared with completely unrelated types.

The problems with using base classes for roles:

Base Classes vs. Role Interfaces
Aspect	Abstract Base Class	Role Interface
Relationship type	IS-A (identity)	CAN-DO (capability)
Inheritance count	Single (one base)	Multiple (many roles)
Coupling	Tight (inherits implementation)	Loose (only contract)
State sharing	Yes (can inherit fields)	No (only behavior contract)
Runtime cost	Virtual dispatch + state	Virtual dispatch only
Testing	May need to handle base class behavior	Mock only the role's methods
Evolution	Dangerous (changes affect all children)	Safe (implementations are independent)

roles-vs-base-classes.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
// PROBLEMATIC: Using base class for role
abstract class Serializable {
    abstract serialize(): string;
    abstract deserialize(data: string): void;
}
 
class User extends Serializable { /* ... */ }
class Order extends Serializable { /* ... */ }
 
// Problem: What if Order needs to extend DomainEntity?
class DomainEntity {
    readonly id: string;
    readonly createdAt: Date;
}
 
// CAN'T DO THIS in single-inheritance languages:
// class Order extends DomainEntity, Serializable { }
 
// SOLUTION: Role interfaces
interface Serializable {
    serialize(): string;
    deserialize(data: string): void;
}
 
interface DomainEntity {
    readonly id: string;
    readonly createdAt: Date;
}
 
// Now Order can be both!
class Order implements Serializable, DomainEntity {
    readonly id: string;
    readonly createdAt: Date;
    
    serialize(): string { /* ... */ }
    deserialize(data: string): void { /* ... */ }
}
 
// And we can combine them freely
class User implements Serializable, DomainEntity { /* ... */ }
class Configuration implements Serializable { /* ... */ } // No DomainEntity needed
class AuditLog implements DomainEntity { /* ... */ } // No Serializable needed

The Role Interface Rule

If you find yourself creating an abstract base class with no fields—only abstract methods—you almost certainly want an interface instead. Base classes are for sharing implementation; interfaces are for sharing contracts. Conflating them leads to inflexible hierarchies.

Summary: Thinking in Roles

We've explored a fundamental paradigm shift in interface design—from implementation-centric to role-centric thinking. This isn't merely a technique; it's a change in perspective that transforms how we model systems.

Key Takeaways

•Interfaces should represent roles, not implementations — Design interfaces from the client's perspective, asking 'What role do I need?' not 'What does this class do?'
•Roles are context-specific and client-defined — The same class can play different roles in different contexts; roles emerge from client needs, not provider capabilities.
•Role interfaces enable semantic naming — Names like Authenticator, Validator, and Repository communicate intent and make code self-documenting.
•Role composition multiplies expressiveness — Small, orthogonal interfaces can be combined to express precisely what each client needs—nothing more.
•Roles enable true polymorphism — Unrelated types that share a role can be treated uniformly, enabling algorithms that work across diverse implementations.
•Roles are not base classes — Interfaces express capabilities (CAN-DO); base classes express identity (IS-A). Confusing them leads to rigid hierarchies.

What's next:

Now that we understand role-based thinking, we'll explore how to design client-specific interfaces—interfaces tailored to what each particular client actually uses. This is the practical application of ISP: ensuring that no client is forced to depend on methods it doesn't need.

Page Complete

You now understand the foundational concept of role-based interface design. This mental model—thinking of interfaces as roles rather than implementation contracts—is the key insight that makes ISP not just a rule to follow, but a natural consequence of good design thinking. Next, we'll see how to apply this by designing interfaces around specific client needs.

1 / 4

Loading learning content...

System Design (LLD)Role-Based Interface Design

Role-Based Interface Design

LevelIntermediate

Duration60 mins

TopicRole-Based Interface Design

1 / 4

Interfaces Representing Roles

The Role Revolution in Interface Design

What You Will Learn

The Conventional View of Interfaces

Designing a class with all its functionality
Extracting an interface that exposes that functionality
Using the interface wherever we want to abstract away the concrete class

This approach seems logical—after all, we're "programming to interfaces, not implementations," right? But there's a critical flaw in this reasoning.

conventional-interface-design.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
// The Conventional Approach: Interface mirrors implementation
 
// Step 1: Design the class with all its functionality
class UserService {
    // Authentication methods
    login(email: string, password: string): User { /* ... */ }
    logout(userId: string): void { /* ... */ }
    resetPassword(email: string): void { /* ... */ }
    changePassword(userId: string, newPassword: string): void { /* ... */ }
    
    // Profile management methods
    getProfile(userId: string): UserProfile { /* ... */ }
    updateProfile(userId: string, profile: UserProfile): void { /* ... */ }
    uploadAvatar(userId: string, image: Buffer): string { /* ... */ }
    
    // User administration methods
    listAllUsers(): User[] { /* ... */ }
    banUser(userId: string): void { /* ... */ }
    deleteUser(userId: string): void { /* ... */ }
    promoteToAdmin(userId: string): void { /* ... */ }
    
    // Notification methods
    sendWelcomeEmail(userId: string): void { /* ... */ }
    sendPasswordResetEmail(email: string): void { /* ... */ }
    notifyLoginFromNewDevice(userId: string, device: DeviceInfo): void { /* ... */ }
}
 
// Step 2: Extract an interface that "abstracts" this class
interface IUserService {
    login(email: string, password: string): User;
    logout(userId: string): void;
    resetPassword(email: string): void;
    changePassword(userId: string, newPassword: string): void;
    getProfile(userId: string): UserProfile;
    updateProfile(userId: string, profile: UserProfile): void;
    uploadAvatar(userId: string, image: Buffer): string;
    listAllUsers(): User[];
    banUser(userId: string): void;
    deleteUser(userId: string): void;
    promoteToAdmin(userId: string): void;
    sendWelcomeEmail(userId: string): void;
    sendPasswordResetEmail(email: string): void;
    notifyLoginFromNewDevice(userId: string, device: DeviceInfo): void;
}
 
// Step 3: Use the interface everywhere
class LoginController {
    constructor(private userService: IUserService) {}
    
    handleLogin(email: string, password: string) {
        // Uses only 2 of 14 methods!
        const user = this.userService.login(email, password);
        this.userService.notifyLoginFromNewDevice(user.id, this.getDeviceInfo());
        return user;
    }
}

The problem is now visible: The LoginController depends on IUserService, but it only uses 2 of the 14 methods. This means:

Unnecessary coupling: LoginController is coupled to concepts it doesn't care about (profile management, admin operations, avatar uploads)
Testing overhead: Mocking IUserService for LoginController tests requires stubbing all 14 methods
Change ripple effects: Adding a method to handle user subscriptions forces recompilation of LoginController, even though it has nothing to do with subscriptions

The fundamental mistake? We designed the interface around the implementation, not around the clients' needs.

The Implementation-Mirroring Trap

Interfaces as Roles — The Paradigm Shift

A role is a coherent set of related responsibilities that a client expects from a collaborator. Roles are:

Context-specific: The same class might play different roles in different contexts
Client-defined: Roles are discovered by examining what clients actually need, not what implementations provide
Minimal: A role includes only what's necessary for that specific collaboration
Semantic: Roles carry meaning about why something is needed, not just what is provided

Let's revisit our example with role-based thinking:

role-based-interface-design.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
// Role-Based Approach: Interfaces defined by what clients need
 
// Role: Something that can authenticate users
interface Authenticator {
    login(email: string, password: string): User;
    logout(userId: string): void;
}
 
// Role: Something that can manage passwords
interface PasswordManager {
    resetPassword(email: string): void;
    changePassword(userId: string, newPassword: string): void;
}
 
// Role: Something that provides user profile access
interface ProfileProvider {
    getProfile(userId: string): UserProfile;
    updateProfile(userId: string, profile: UserProfile): void;
}
 
// Role: Something that can manage avatars
interface AvatarManager {
    uploadAvatar(userId: string, image: Buffer): string;
}
 
// Role: Something with administrative powers over users
interface UserAdministrator {
    listAllUsers(): User[];
    banUser(userId: string): void;
    deleteUser(userId: string): void;
    promoteToAdmin(userId: string): void;
}
 
// Role: Something that can send user notifications
interface UserNotifier {
    sendWelcomeEmail(userId: string): void;
    sendPasswordResetEmail(email: string): void;
    notifyLoginFromNewDevice(userId: string, device: DeviceInfo): void;
}
 
// The implementation can fulfill multiple roles
class UserService implements 
    Authenticator, 
    PasswordManager, 
    ProfileProvider, 
    AvatarManager, 
    UserAdministrator, 
    UserNotifier {
    // Implementation of all methods...
}
 
// Now LoginController depends only on the roles it needs!
class LoginController {
    constructor(
        private authenticator: Authenticator,
        private notifier: UserNotifier
    ) {}
    
    handleLogin(email: string, password: string) {
        const user = this.authenticator.login(email, password);
        this.notifier.notifyLoginFromNewDevice(user.id, this.getDeviceInfo());
        return user;
    }
}

Notice the transformation:

Each interface represents a coherent role with a clear name that expresses its purpose
Interfaces are small and focused — typically 1-4 methods
Clients depend only on the roles they need — LoginController takes an Authenticator and UserNotifier
The class fulfills multiple roles by implementing multiple interfaces
Testing is simpler — mock only the 2 interfaces with 5 total methods, not 14

This is the essence of role-based interface design: interfaces are discovered from client needs, not extracted from implementations.

The Direction of Discovery

Roles in Real-World Systems

Role-based interface design isn't just a theoretical nicety—it's how well-designed real-world systems are structured. Let's examine how this pattern appears across different domains.

Example 1: The File System Abstraction

A file on disk can play many roles. Different clients need different subsets of its capabilities:

file-system-roles.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
// Different clients need different views of a "file"
 
// Role: Something whose content can be read
interface Readable {
    read(buffer: Buffer, offset: number, length: number): number;
}
 
// Role: Something whose content can be written
interface Writable {
    write(buffer: Buffer, offset: number, length: number): number;
}
 
// Role: Something that can be positioned
interface Seekable {
    seek(position: number, whence: SeekMode): number;
    tell(): number;
}
 
// Role: Something that can be closed/released
interface Closeable {
    close(): void;
}
 
// Role: Something with metadata
interface FileMetadata {
    readonly size: number;
    readonly createdAt: Date;
    readonly modifiedAt: Date;
    readonly permissions: FilePermissions;
}
 
// A file descriptor in the OS implements multiple roles
class FileDescriptor implements 
    Readable, 
    Writable, 
    Seekable, 
    Closeable, 
    FileMetadata {
    // Full implementation...
}
 
// A log writer only needs Writable + Closeable
class LogWriter {
    constructor(
        private output: Writable & Closeable
    ) {}
    
    log(message: string): void {
        const buffer = Buffer.from(message + '
');
        this.output.write(buffer, 0, buffer.length);
    }
    
    finish(): void {
        this.output.close();
    }
}
 
// A file searcher only needs Readable + Seekable
class FileSearcher {
    constructor(
        private file: Readable & Seekable
    ) {}
    
    findPattern(pattern: Buffer): number | null {
        // Can read and seek, doesn't need to write or know metadata
    }
}
 
// A backup system needs FileMetadata + Readable
class BackupManager {
    constructor(
        private source: FileMetadata & Readable
    ) {}
    
    shouldBackup(): boolean {
        // Uses metadata.modifiedAt
        return this.source.modifiedAt > this.lastBackupTime;
    }
}

Example 2: E-Commerce Order System

An Order object participates in many different business processes. Each process views the order through a different role:

order-roles.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
// Role: Something that can be priced
interface Priceable {
    readonly subtotal: Money;
    readonly tax: Money;
    readonly total: Money;
    applyDiscount(discount: Discount): void;
}
 
// Role: Something that can be shipped
interface Shippable {
    readonly shippingAddress: Address;
    readonly weight: Weight;
    readonly dimensions: Dimensions;
    calculateShippingCost(carrier: Carrier): Money;
}
 
// Role: Something that can be paid for
interface Payable {
    readonly paymentDue: Money;
    recordPayment(payment: Payment): void;
    readonly isFullyPaid: boolean;
}
 
// Role: Something with line items
interface LineItemContainer {
    readonly items: ReadonlyArray<LineItem>;
    addItem(product: Product, quantity: number): void;
    removeItem(itemId: string): void;
}
 
// Role: Something with order lifecycle
interface OrderLifecycle {
    readonly status: OrderStatus;
    submit(): void;
    cancel(reason: string): void;
    refund(): void;
}
 
// Role: Something that can be tracked by customer
interface CustomerTrackable {
    readonly orderId: string;
    readonly estimatedDelivery: Date;
    readonly trackingNumber: string | null;
    readonly statusHistory: ReadonlyArray<StatusEvent>;
}
 
// The Order aggregate implements all these roles
class Order implements 
    Priceable, 
    Shippable, 
    Payable, 
    LineItemContainer, 
    OrderLifecycle, 
    CustomerTrackable {
    // Full implementation...
}
 
// Payment processor only sees Payable
class PaymentProcessor {
    process(payable: Payable, paymentMethod: PaymentMethod): PaymentResult {
        // Works with any payable thing, not just Orders
        // Could also process invoices, subscriptions, etc.
    }
}
 
// Shipping calculator only sees Shippable
class ShippingCalculator {
    calculateOptions(shippable: Shippable): ShippingOption[] {
        // Works with any shippable thing
        // Doesn't know or care about payment status
    }
}
 
// Customer portal only sees CustomerTrackable
class CustomerOrderView {
    constructor(private order: CustomerTrackable) {}
    
    render(): CustomerViewDTO {
        // Can only access tracking info
        // Cannot cancel, modify, or see internal order details
    }
}

Roles Cross-Cut Types

The Linguistic Power of Roles

Compare these two method signatures:

Implementation-Centric

•sendNotification(userService: IUserService)
•What can userService do? Everything a user service does.
•Why are we passing a user service to send notifications?
•Would a mock UserService work here?
•Meaning is obscured by scope

Role-Centric

•sendNotification(notifier: Notifier)
•What can notifier do? Notify users.
•Why? Because we need to send notifications.
•Any object that can notify will work.
•Meaning is expressed in the type

The naming of roles carries semantic weight. Consider these role names and what they communicate:

Role Name	Semantic Meaning
`Authenticator`	Validates identity, establishes sessions
`Authorizer`	Checks permissions, enforces access control
`Encryptor`	Transforms data into secure, unreadable form
`Validator`	Checks data against rules, reports violations
`Transformer`	Converts data from one form to another
`Publisher`	Broadcasts events to interested parties
`Subscriber`	Receives and reacts to published events
`Repository`	Provides collection-like access to persistent objects
`Factory`	Creates complex objects without exposing construction logic
`Observer`	Watches for changes, reacts when notified

Role Naming Heuristics

Role Composition — The Multiplicative Power

role-composition.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
// Individual roles
interface Identifiable {
    readonly id: string;
}
 
interface Timestamped {
    readonly createdAt: Date;
    readonly updatedAt: Date;
}
 
interface Versionable {
    readonly version: number;
    incrementVersion(): void;
}
 
interface SoftDeletable {
    readonly isDeleted: boolean;
    softDelete(): void;
    restore(): void;
}
 
interface Auditable {
    readonly createdBy: UserId;
    readonly lastModifiedBy: UserId;
    recordModification(userId: UserId): void;
}
 
// Composed roles for specific needs
type Trackable = Identifiable & Timestamped;
type VersionedEntity = Identifiable & Versionable;
type AuditedEntity = Identifiable & Timestamped & Auditable;
type FullyManaged = Identifiable & Timestamped & Versionable & SoftDeletable & Auditable;
 
// Repository that works with any identifiable, timestamped entity
interface Repository<T extends Trackable> {
    findById(id: string): T | null;
    findByCreatedAfter(date: Date): T[];
    save(entity: T): void;
}
 
// Audit service works with anything auditable
class AuditService {
    recordChange<T extends AuditedEntity>(entity: T, userId: UserId): void {
        entity.recordModification(userId);
        this.auditLog.append({
            entityId: entity.id,
            timestamp: new Date(),
            user: userId,
            previousVersion: entity.version
        });
    }
}
 
// Soft delete handler works with anything soft-deletable and identifiable
class SoftDeleteHandler {
    delete<T extends Identifiable & SoftDeletable>(entity: T): void {
        entity.softDelete();
        this.eventBus.publish(new EntitySoftDeleted(entity.id));
    }
    
    restore<T extends Identifiable & SoftDeletable>(entity: T): void {
        entity.restore();
        this.eventBus.publish(new EntityRestored(entity.id));
    }
}

The mathematics of role composition:

With a single fat interface that combines all these capabilities, we have 1 option: all or nothing.

Role Composition Scales Better
Approach	10 Capabilities	20 Capabilities	30 Capabilities
Fat Interfaces	1 way to use	1 way to use	1 way to use
Role Composition	1,024 combinations	~1 million combinations	~1 billion combinations

The Combinatorial Advantage

Roles Enable True Polymorphism

Consider: What do these have in common?

A File on disk
An HttpResponse from a server
A DatabaseResultSet from a query
A GeneratorFunction that yields values
A MockDataSource in a test

They're completely different types with different implementations. But they can all be Readable—something that provides data when asked. This shared role enables polymorphic code:

true-polymorphism.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
// Role interface for anything that can provide data
interface DataSource {
    read(): AsyncIterable<Buffer>;
    readonly metadata: SourceMetadata;
}
 
// All these unrelated types implement the same role
class FileSource implements DataSource { /* ... */ }
class HttpResponseSource implements DataSource { /* ... */ }
class DatabaseSource implements DataSource { /* ... */ }
class InMemorySource implements DataSource { /* ... */ }
class S3BucketSource implements DataSource { /* ... */ }
 
// Polymorphic processor works with ANY data source
class DataProcessor {
    // This single method works with 5+ completely different types!
    async process(source: DataSource): Promise<ProcessedResult> {
        const results: ProcessedChunk[] = [];
        
        for await (const chunk of source.read()) {
            results.push(this.processChunk(chunk));
        }
        
        return {
            metadata: source.metadata,
            chunks: results,
            processedAt: new Date()
        };
    }
}
 
// Usage: same processor, different sources
const processor = new DataProcessor();
 
// Process a local file
const fileResult = await processor.process(new FileSource('/data/input.csv'));
 
// Process an HTTP response
const httpResult = await processor.process(new HttpResponseSource(apiUrl));
 
// Process database results
const dbResult = await processor.process(new DatabaseSource(query));
 
// Process test data in unit tests
const testResult = await processor.process(new InMemorySource(mockData));
 
// Process data from S3
const s3Result = await processor.process(new S3BucketSource(bucketName, key));

This is polymorphism working as intended:

One algorithm (DataProcessor.process()) works with many types
Types don't need a common base class—they share a role
New sources can be added without modifying the processor
Testing is trivial: provide an InMemorySource with controlled data

Without role-based interfaces, we'd be forced into awkward alternatives:

Giant switch statements checking types at runtime
Adapter classes wrapping every type into a common base
Tight coupling where the processor knows about every source type
Duplicated logic with separate processing methods per type

Role-based interfaces give us polymorphism without inheritance, composition without boilerplate.

Duck Typing Done Right

Roles vs. Base Classes — The Critical Distinction

Some developers wonder: "Why not just use abstract base classes for shared behavior?" The answer reveals a critical distinction between role-based design and inheritance-based design.

Base classes express what something is. A Dog extends Mammal because a dog is a mammal. This is an ontological claim about identity and shared characteristics.

The problems with using base classes for roles:

Base Classes vs. Role Interfaces
Aspect	Abstract Base Class	Role Interface
Relationship type	IS-A (identity)	CAN-DO (capability)
Inheritance count	Single (one base)	Multiple (many roles)
Coupling	Tight (inherits implementation)	Loose (only contract)
State sharing	Yes (can inherit fields)	No (only behavior contract)
Runtime cost	Virtual dispatch + state	Virtual dispatch only
Testing	May need to handle base class behavior	Mock only the role's methods
Evolution	Dangerous (changes affect all children)	Safe (implementations are independent)

roles-vs-base-classes.ts
TypeScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
// PROBLEMATIC: Using base class for role
abstract class Serializable {
    abstract serialize(): string;
    abstract deserialize(data: string): void;
}
 
class User extends Serializable { /* ... */ }
class Order extends Serializable { /* ... */ }
 
// Problem: What if Order needs to extend DomainEntity?
class DomainEntity {
    readonly id: string;
    readonly createdAt: Date;
}
 
// CAN'T DO THIS in single-inheritance languages:
// class Order extends DomainEntity, Serializable { }
 
// SOLUTION: Role interfaces
interface Serializable {
    serialize(): string;
    deserialize(data: string): void;
}
 
interface DomainEntity {
    readonly id: string;
    readonly createdAt: Date;
}
 
// Now Order can be both!
class Order implements Serializable, DomainEntity {
    readonly id: string;
    readonly createdAt: Date;
    
    serialize(): string { /* ... */ }
    deserialize(data: string): void { /* ... */ }
}
 
// And we can combine them freely
class User implements Serializable, DomainEntity { /* ... */ }
class Configuration implements Serializable { /* ... */ } // No DomainEntity needed
class AuditLog implements DomainEntity { /* ... */ } // No Serializable needed

The Role Interface Rule

Summary: Thinking in Roles

Key Takeaways

•Interfaces should represent roles, not implementations — Design interfaces from the client's perspective, asking 'What role do I need?' not 'What does this class do?'
•Roles are context-specific and client-defined — The same class can play different roles in different contexts; roles emerge from client needs, not provider capabilities.
•Role interfaces enable semantic naming — Names like Authenticator, Validator, and Repository communicate intent and make code self-documenting.
•Role composition multiplies expressiveness — Small, orthogonal interfaces can be combined to express precisely what each client needs—nothing more.
•Roles enable true polymorphism — Unrelated types that share a role can be treated uniformly, enabling algorithms that work across diverse implementations.
•Roles are not base classes — Interfaces express capabilities (CAN-DO); base classes express identity (IS-A). Confusing them leads to rigid hierarchies.

What's next:

Page Complete

1 / 4