System Design (LLD)ISP and Dependency Management

ISP and Dependency Management

LevelIntermediate

Duration90 mins

TopicISP and Dependency Management

1 / 4

Reducing Transitive Dependencies

The Hidden Cost of Every Import Statement

Every time you add an import or include statement to your code, you're signing a contract—not just with the module you're importing, but with everything that module transitively depends upon. This seemingly innocuous action can silently weave your component into a vast web of interconnected dependencies, each one capable of breaking your build, introducing security vulnerabilities, or forcing upgrades you never anticipated.

Transitive dependencies are the dependencies of your dependencies. When module A depends on module B, and module B depends on module C, then A has a transitive dependency on C—whether you intended it or not, whether you're aware of it or not. In large systems, this creates dependency graphs with thousands of nodes, where a single change ripples through the entire organization.

What You Will Learn

By the end of this page, you will understand exactly how fat interfaces propagate transitive dependencies, why this creates fragile systems, and how applying ISP systematically can reduce your dependency footprint by orders of magnitude. You'll gain practical techniques used by principal engineers to break dependency chains in enterprise systems.

Understanding Transitive Dependencies

Before we can reduce transitive dependencies, we must deeply understand what they are, how they arise, and why they become problematic as systems scale.

Definition: A transitive dependency exists when component A depends on component B, and B depends on component C. Even though A never directly references C, A is still affected by changes to C—this is transitivity.

The Dependency Graph Reality:

When you write import UserService from './user-service', you're not just importing UserService. You're also importing:

Everything UserService imports
Everything those imports import
Recursively, until you reach the leaves of your dependency tree

In a typical enterprise application, a single high-level module might transitively depend on 500+ other modules. A single change anywhere in that tree can trigger recompilation, retesting, and redeployment of your module.

transitive-dependency-example.ts
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
// Level 1: Your code
// payment-processor.ts
import { OrderService } from './order-service';
 
class PaymentProcessor {
    constructor(private orderService: OrderService) {}
    
    processPayment(orderId: string): void {
        const order = this.orderService.getOrder(orderId);
        // Process payment...
    }
}
 
// Level 2: OrderService (your direct dependency)
// order-service.ts
import { UserService } from './user-service';
import { InventoryService } from './inventory-service';
import { NotificationService } from './notification-service';
import { AuditLogger } from './audit-logger';
import { PromotionEngine } from './promotion-engine';
 
export class OrderService {
    constructor(
        private userService: UserService,
        private inventoryService: InventoryService,
        private notificationService: NotificationService,
        private auditLogger: AuditLogger,
        private promotionEngine: PromotionEngine
    ) {}
    
    getOrder(orderId: string): Order { /* ... */ }
    createOrder(/* ... */): Order { /* ... */ }
    updateInventory(/* ... */): void { /* ... */ }
    notifyUser(/* ... */): void { /* ... */ }
    calculateDiscounts(/* ... */): Discount[] { /* ... */ }
}
 
// Level 3+: Each of those services has its own dependencies...
// UserService → DatabaseConnection, CacheService, AuthProvider, ...
// InventoryService → WarehouseAPI, SupplierIntegration, ...
// NotificationService → EmailProvider, SMSGateway, PushService, ...
 
// Your PaymentProcessor now transitively depends on:
// - Database drivers
// - Email services
// - SMS gateways
// - Warehouse APIs
// - Third-party auth providers
// - And dozens more...

The Compilation Cascade

In statically-typed languages, changing any transitive dependency requires recompilation of everything that depends on it. In a 10,000-module codebase, a single interface change in a core utility can trigger recompilation of 8,000 modules—turning a 30-second change into a 30-minute build.

How Fat Interfaces Multiply Dependencies

Fat interfaces are dependency multipliers. When an interface contains methods that serve different concerns, it forces all clients to depend on all the types and dependencies required by all those methods—even if a client only uses one method.

The Mathematics of Interface Bloat:

Consider an interface with N methods, where each method introduces M unique type dependencies. A fat interface containing all N methods exposes clients to N × M potential dependencies. If the interface is segregated into N single-method interfaces, each client only depends on 1 × M dependencies—the ones they actually use.

This isn't just arithmetic. In practice, fat interfaces create combinatorial explosion of the dependency graph.

fat-interface-dependency-explosion.ts
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
// BAD: Fat interface forces all clients to depend on ALL types
interface IOrderManagement {
    // Order queries - depends on: Order, OrderFilter, PagedResult
    getOrder(id: string): Order;
    queryOrders(filter: OrderFilter): PagedResult<Order>;
    
    // Order mutations - depends on: CreateOrderRequest, UpdateOrderRequest
    createOrder(request: CreateOrderRequest): Order;
    updateOrder(id: string, request: UpdateOrderRequest): Order;
    
    // Inventory operations - depends on: InventoryItem, WarehouseLocation
    checkInventory(productId: string): InventoryItem;
    reserveInventory(orderId: string, location: WarehouseLocation): void;
    
    // Shipping - depends on: ShippingCarrier, TrackingInfo, Address
    calculateShipping(carrier: ShippingCarrier, address: Address): Money;
    getTracking(orderId: string): TrackingInfo;
    
    // Billing - depends on: PaymentMethod, Invoice, TaxCalculation
    processPayment(orderId: string, payment: PaymentMethod): Invoice;
    calculateTax(order: Order): TaxCalculation;
    
    // Notifications - depends on: NotificationChannel, NotificationTemplate
    sendConfirmation(orderId: string, channel: NotificationChannel): void;
    sendShippingUpdate(orderId: string, template: NotificationTemplate): void;
}
 
// A simple order display component only needs getOrder()
// But it now transitively depends on:
// - WarehouseLocation (from inventory)
// - ShippingCarrier (from shipping)  
// - PaymentMethod (from billing)
// - NotificationChannel (from notifications)
// - Plus everything THOSE types depend on
 
class OrderDisplayComponent {
    // We only call ONE method
    constructor(private orderManagement: IOrderManagement) {}
    
    showOrder(id: string): void {
        const order = this.orderManagement.getOrder(id);
        this.render(order);
    }
}
 
// If ShippingCarrier changes, OrderDisplayComponent must recompile
// even though it has absolutely nothing to do with shipping!

Dependency Impact Analysis: Fat vs Segregated Interfaces
Metric	Fat Interface	Segregated Interfaces	Impact
Types exposed to each client	12 domain types	2-3 relevant types only	~80% reduction
Recompilation triggers	Any of 12 types changes	Only when relevant types change	~85% fewer builds
Test dependencies	All 12 types must be stubbed	Only used types stubbed	~75% less test setup
Deployment coupling	All clients deploy together	Independent deployment per client	100% deployment isolation
Breaking change surface	Any method signature change	Only changes to used methods	~90% reduction

The Dependency Ratio

A healthy module has a dependency ratio (transitive deps / direct deps) below 10:1. Fat interfaces commonly create ratios of 50:1 or higher. When you see a simple component pulling in hundreds of transitive dependencies, look for fat interfaces in the dependency chain.

The Mechanics of Dependency Propagation

Understanding exactly how dependencies propagate through interfaces is essential for breaking dependency chains. There are three primary propagation vectors: type signatures, implementation requirements, and behavioral contracts.

Vector 1: Type Signature Dependencies

Every type mentioned in a method signature creates a compile-time dependency. If IOrderManagement.calculateShipping() returns Money and accepts ShippingCarrier, then every client of IOrderManagement must have access to both Money and ShippingCarrier definitions—even clients that never call calculateShipping().

Vector 2: Implementation Requirements

When implementing a fat interface, the implementing class must satisfy all methods. This forces the implementation to depend on all the infrastructure needed by all methods. A class implementing IOrderManagement must have database access (for queries), warehouse API access (for inventory), payment gateway access (for billing), and notification services (for alerts)—even if different clients only need one of these capabilities.

Vector 3: Behavioral Contract Dependencies

Less obvious but equally important: clients often must understand the behavioral contracts of methods they don't call because those behaviors affect the state that methods they do call depend on. If reserveInventory() can change order status, clients of getOrder() must understand inventory reservation semantics to interpret order state correctly.

dependency-propagation-vectors.ts
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
// VECTOR 1: Type Signature Dependencies
// Every type in the interface creates a hard compile-time dependency
 
interface IDocumentManager {
    // Anyone depending on IDocumentManager must know what these types are:
    createDocument(content: DocumentContent): Document;
    //             ↑ depends on DocumentContent  ↑ depends on Document
    
    convertToPdf(doc: Document): PdfDocument;
    //                           ↑ depends on PdfDocument
    
    signDocument(doc: Document, cert: SigningCertificate): SignedDocument;
    //                               ↑ depends on SigningCertificate
    //                                                    ↑ depends on SignedDocument
    
    archiveDocument(doc: Document, policy: ArchivalPolicy): ArchiveReceipt;
    //                                   ↑ depends on ArchivalPolicy
    //                                                      ↑ depends on ArchiveReceipt
}
 
// Even if a client only reads documents, they depend on:
// - SigningCertificate (cryptography libraries)
// - ArchivalPolicy (compliance frameworks)
// - PdfDocument (PDF processing libraries)
 
 
// VECTOR 2: Implementation Requirements
// The implementation must satisfy ALL methods, pulling in ALL dependencies
 
class DocumentManagerImpl implements IDocumentManager {
    constructor(
        private storage: StorageService,      // For createDocument
        private pdfConverter: PdfConverter,    // For convertToPdf
        private cryptoService: CryptoService,  // For signDocument
        private archiveClient: ArchiveClient,  // For archiveDocument
    ) {}
    
    // Even if no client ever calls signDocument(),
    // this class must still inject CryptoService!
    // That means the entire dependency tree of CryptoService
    // is pulled into this implementation.
}
 
 
// VECTOR 3: Behavioral Contract Dependencies
// Understanding how methods interact creates implicit dependencies
 
interface IInventoryManager {
    getStock(productId: string): StockLevel;
    reserveStock(productId: string, quantity: number): Reservation;
    releaseReservation(reservationId: string): void;
    adjustStock(productId: string, delta: number): StockLevel;
}
 
// A client calling getStock() must understand:
// - How reserveStock() affects reported stock levels
// - Whether adjustStock() can make stock negative
// - How long reservations last before auto-release
// 
// These behavioral dependencies aren't in the types, but they're real.

Signs of Dependency Propagation Problems

•Import lists exceeding 20 lines — When a file imports dozens of types it doesn't directly use, it's suffering from transitive type dependencies through fat interfaces.
•Mock objects requiring 50+ method stubs — Large mock setup indicates fat interfaces forcing test dependencies on unused functionality.
•'Diamond dependency' conflicts — When transitive dependencies create version conflicts (e.g., two paths to the same library at different versions).
•Build times scaling non-linearly — Small changes trigger massive recompilation because of deep transitive dependency trees.
•Deployment packages that dwarf actual code — Runtime dependencies ballooning because implementations must satisfy unused interface methods.

ISP as a Dependency Firewall

When properly applied, ISP creates dependency firewalls—barriers that prevent unnecessary dependencies from propagating to clients. Each segregated interface becomes a checkpoint: dependencies can only cross if explicitly required.

The Firewall Principle:

A well-segregated interface exposes only the types required by its specific operation. Clients of that interface are protected from all other types in the system. Changes outside the interface's concern cannot propagate to its clients.

This creates compartmentalization. Instead of one massive dependency blob where everything affects everything, you get isolated dependency clusters where changes are contained.

before-isp.ts
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
// WITHOUT ISP: No dependency firewalls
// All clients depend on all types
 
interface IOrderSystem {
    // Query operations
    getOrder(id: string): Order;
    
    // Shipping operations
    ship(id: string, carrier: ShippingCarrier): TrackingInfo;
    
    // Payment operations
    charge(id: string, method: PaymentMethod): Receipt;
    
    // Reporting operations
    generateReport(params: ReportParams): Report;
}
 
// Client only needs order queries:
class OrderViewer {
    constructor(
        private orders: IOrderSystem  // Full dependency exposure!
    ) {}
}
 
// OrderViewer transitively depends on:
// - ShippingCarrier
// - TrackingInfo
// - PaymentMethod
// - Receipt
// - ReportParams
// - Report
// Plus all THEIR dependencies...

after-isp.ts
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
// WITH ISP: Dependency firewalls established
 
// Each interface is a firewall
interface IOrderQueries {
    getOrder(id: string): Order;
}
 
interface IOrderShipping {
    ship(id: string, carrier: ShippingCarrier): TrackingInfo;
}
 
interface IOrderPayments {
    charge(id: string, method: PaymentMethod): Receipt;
}
 
interface IOrderReporting {
    generateReport(params: ReportParams): Report;
}
 
// Client only gets query dependencies:
class OrderViewer {
    constructor(
        private orders: IOrderQueries  // Firewall active!
    ) {}
}
 
// OrderViewer ONLY depends on:
// - Order
// 
// Shipping, payment, and reporting 
// dependencies are BLOCKED by the firewall.

Dependency Containment Ratio

After applying ISP, measure your containment ratio: (blocked dependencies) / (total dependencies the fat interface would have exposed). A well-segregated system achieves containment ratios above 80%—meaning 80% of potential dependencies are blocked by interface firewalls.

Creating Effective Firewalls:

Not all interface splits create effective firewalls. To maximize containment:

Group by dependency cluster — Methods that share the same type dependencies belong together. createOrder() and updateOrder() likely share OrderRequest types; keep them in one interface.
Separate by infrastructure — Methods requiring different infrastructure (database, network, filesystem) should be in different interfaces. This prevents infrastructure dependencies from crossing boundaries.
Isolate external systems — Methods that integrate with external services should be in dedicated interfaces. If the shipping API changes, only IOrderShipping clients need updates.
Consider stability — Put stable operations in separate interfaces from volatile ones. If payment methods change frequently but order queries are stable, separate them.

Measuring Transitive Dependency Reduction

What gets measured gets managed. To systematically reduce transitive dependencies, you need quantitative metrics. Here are the key measurements used by principal engineers to track and reduce dependency bloat.

Metric 1: Dependency Fan-Out

For each interface, count the unique types in all method signatures. A fat interface might have 20+ types; after segregation, each focused interface should have 3-5.

Metric 2: Transitive Closure Size

The transitive closure is the total number of modules reachable from a given module through the dependency graph. Tools like madge (JavaScript), jdepend (Java), or snakefood (Python) can compute this.

Metric 3: Recompilation Impact Score

For each interface change, measure how many modules require recompilation. Track this over time—it should decrease as you apply ISP.

Metric 4: Deployment Independence

Count how many services can be deployed independently. Before ISP: if core interfaces change, all services redeploy. After ISP: only affected clients redeploy.

dependency-metrics.ts
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
// Tools and techniques for measuring dependency reduction
 
// 1. Calculate direct type dependencies per interface
function calculateTypeFanOut(interfaceDeclaration: InterfaceNode): number {
    const types = new Set<string>();
    
    for (const method of interfaceDeclaration.methods) {
        // Count parameter types
        for (const param of method.parameters) {
            types.add(param.type.name);
        }
        // Count return type
        types.add(method.returnType.name);
    }
    
    return types.size;
}
 
// Before ISP: IOrderManagement has 12 unique types
// After ISP:  IOrderQueries has 3, IOrderShipping has 4, etc.
 
 
// 2. Measure transitive closure growth
interface DependencyMetrics {
    module: string;
    directDependencies: number;
    transitiveDependencies: number;
    closureRatio: number;  // transitive / direct
}
 
function analyzeModule(module: Module): DependencyMetrics {
    const directDeps = module.imports.length;
    const transitiveDeps = computeTransitiveClosure(module).size;
    
    return {
        module: module.name,
        directDependencies: directDeps,
        transitiveDependencies: transitiveDeps,
        closureRatio: transitiveDeps / directDeps
    };
}
 
// BEFORE ISP:
// PaymentProcessor: { direct: 1, transitive: 147, ratio: 147 }
// 
// AFTER ISP:
// PaymentProcessor: { direct: 1, transitive: 12, ratio: 12 }
 
 
// 3. Track recompilation impact
interface ChangeImpact {
    changedModule: string;
    affectedModules: string[];
    recompilationCount: number;
    percentageOfCodebase: number;
}
 
function assessChangeImpact(changedModule: Module): ChangeImpact {
    const affected = findAllDependents(changedModule);
    return {
        changedModule: changedModule.name,
        affectedModules: affected.map(m => m.name),
        recompilationCount: affected.length,
        percentageOfCodebase: (affected.length / totalModules) * 100
    };
}
 
// Tracking over time:
// Week 1: Average impact = 340 modules (34% of codebase)
// Week 4: Average impact = 89 modules (8.9% of codebase)
// Week 8: Average impact = 23 modules (2.3% of codebase)

Real-World Metrics: Before and After ISP Refactoring
System	Metric	Before ISP	After ISP	Improvement
E-commerce Platform	Avg. transitive deps per module	234	41	82% reduction
E-commerce Platform	Build time (incremental)	8.4 min	47 sec	91% faster
Payment Gateway	Closure ratio	67:1	8:1	88% reduction
Payment Gateway	Independent deployable units	1	7	7x increase
CMS Backend	Recompilation on core change	847 files	112 files	87% contained
CMS Backend	Test setup time	12.3 sec	1.8 sec	85% faster

Strategies for Breaking Dependency Chains

Reducing transitive dependencies requires systematic strategies. Here are the proven techniques used to break dependency chains in production systems.

Strategy 1: Interface Extraction

Identify the methods each client actually uses. Extract those methods into client-specific interfaces. The original fat interface can extend all the new interfaces for backward compatibility during migration.

Strategy 2: Dependency Inversion at Boundaries

At module boundaries, invert dependencies so that high-level modules own the interfaces they depend on. This prevents low-level implementation details from propagating upward.

Strategy 3: Type Narrowing

When a method returns a large object but clients only need a subset, create a narrower return type. Instead of returning Order with 50 fields, return OrderSummary with 5 fields.

Strategy 4: Adapter Interfaces

Create adapter interfaces that translate between internal rich types and external minimal types. Clients depend on the minimal adapter interface, not the full internal types.

dependency-breaking-strategies.ts
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
83
84
85
86
87
// STRATEGY 1: Interface Extraction
// Start with fat interface, extract based on usage analysis
 
// Original fat interface
interface IUserService {
    getUser(id: string): User;
    updateUser(id: string, data: UpdateUserRequest): User;
    deleteUser(id: string): void;
    authenticateUser(credentials: Credentials): AuthToken;
    resetPassword(userId: string, request: PasswordResetRequest): void;
    getUserPermissions(userId: string): Permission[];
    generateReport(params: UserReportParams): UserReport;
}
 
// Step 1: Analyze client usage
// - ProfileView only calls getUser()
// - AuthController calls authenticateUser(), resetPassword()
// - AdminDashboard calls all methods
 
// Step 2: Extract client-specific interfaces
interface IUserReader {
    getUser(id: string): User;
}
 
interface IUserAuthenticator {
    authenticateUser(credentials: Credentials): AuthToken;
    resetPassword(userId: string, request: PasswordResetRequest): void;
}
 
interface IUserAdmin extends IUserReader, IUserAuthenticator {
    updateUser(id: string, data: UpdateUserRequest): User;
    deleteUser(id: string): void;
    getUserPermissions(userId: string): Permission[];
    generateReport(params: UserReportParams): UserReport;
}
 
// STRATEGY 3: Type Narrowing
// Reduce exposed types by returning minimal projections
 
// Before: Returns full Order with all dependencies
interface IOrderQueries {
    getOrder(id: string): Order;  // Order has 50+ fields including nested objects
}
 
// After: Returns only what's needed
interface IOrderSummaryQueries {
    getOrderSummary(id: string): OrderSummary;
}
 
interface OrderSummary {
    id: string;
    status: OrderStatus;
    totalAmount: Money;
    createdAt: Date;
    // Only 4 fields instead of 50
    // No nested objects that pull in more dependencies
}
 
 
// STRATEGY 4: Adapter Interfaces
// Create a minimal surface for external consumers
 
// Internal rich type (many dependencies)
interface InternalOrder {
    id: string;
    customer: Customer;          // Pulls in Customer dependencies
    items: OrderItem[];          // Pulls in Product dependencies
    shipping: ShippingDetails;   // Pulls in Address, Carrier dependencies
    payment: PaymentInfo;        // Pulls in PaymentMethod dependencies
    // ... more fields
}
 
// Adapter interface for external consumers
interface IExternalOrderAdapter {
    getOrderForDisplay(id: string): OrderDisplayDTO;
}
 
interface OrderDisplayDTO {
    id: string;
    customerName: string;        // String, not Customer object
    itemCount: number;           // Number, not OrderItem[]
    formattedTotal: string;      // String, not Money object
    statusLabel: string;         // String, not enum
}
 
// External consumers depend only on strings and primitives
// Zero transitive dependencies from domain model!

Dependency Breaking Decision Tree

•Identify the pain point — Which client is suffering from dependency bloat? Measure their transitive dependency count.
•Analyze actual usage — What methods does this client actually call? Use static analysis or runtime tracing.
•Check for type narrowing opportunities — Can return types be simplified to DTOs or projections?
•Extract minimal interface — Create an interface containing only the methods and types the client needs.
•Implement adapter if needed — If types can't be narrowed, create an adapter that translates to minimal types.
•Migrate client — Update the client to depend on the new minimal interface.
•Measure improvement — Compare transitive dependency count before and after.

Case Study: Breaking a Dependency Monolith

Let's walk through a real-world scenario: a payment platform where a single IPaymentService interface had become a dependency monolith, forcing every consumer to pull in the entire payment ecosystem.

Initial State:

IPaymentService with 34 methods
47 different types in method signatures
Average consumer only used 3-4 methods
But every consumer had 312 transitive dependencies
Build time for any payment change: 18 minutes
Every microservice in the platform depended on payment types

payment-service-before.ts
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
// THE PROBLEM: Payment service interface before ISP
// 34 methods, 47 types, every consumer gets everything
 
interface IPaymentService {
    // Card processing (depends on: Card, CardToken, CardNetwork, ...)
    tokenizeCard(card: CardDetails): CardToken;
    chargeCard(token: CardToken, amount: Money): ChargeResult;
    voidCharge(chargeId: string): VoidResult;
    captureCharge(chargeId: string, amount?: Money): CaptureResult;
    refundCharge(chargeId: string, amount?: Money): RefundResult;
    
    // Bank transfers (depends on: BankAccount, RoutingNumber, ACHDetails, ...)
    initiateACHTransfer(source: BankAccount, dest: BankAccount, amount: Money): ACHTransfer;
    getACHStatus(transferId: string): ACHStatus;
    
    // Digital wallets (depends on: WalletProvider, WalletToken, ApplePaySession, ...)
    createWalletSession(provider: WalletProvider): WalletSession;
    chargeWallet(session: WalletSession, amount: Money): ChargeResult;
    
    // Subscriptions (depends on: Subscription, BillingPlan, SubscriptionEvent, ...)
    createSubscription(customer: Customer, plan: BillingPlan): Subscription;
    updateSubscription(subId: string, plan: BillingPlan): Subscription;
    cancelSubscription(subId: string): CancelResult;
    
    // Invoicing (depends on: Invoice, LineItem, TaxCalculation, ...)
    createInvoice(customer: Customer, items: LineItem[]): Invoice;
    sendInvoice(invoiceId: string): SendResult;
    recordPayment(invoiceId: string, payment: PaymentRecord): Invoice;
    
    // Payouts (depends on: Payout, PayoutMethod, PayoutSchedule, ...)
    createPayout(recipient: Recipient, method: PayoutMethod): Payout;
    schedulePayout(payout: Payout, schedule: PayoutSchedule): ScheduledPayout;
    
    // Fraud (depends on: FraudSignal, RiskScore, FraudRule, ...)
    assessFraudRisk(transaction: Transaction): RiskAssessment;
    reportFraud(transactionId: string, signal: FraudSignal): FraudReport;
    
    // Reporting (depends on: Report, ReportParams, DateRange, ...)
    generateTransactionReport(params: ReportParams): Report;
    getPaymentAnalytics(dateRange: DateRange): Analytics;
    
    // ... 16 more methods
}
 
// IMPACT: A simple checkout service that only needs chargeCard()
// must import and potentially instantiate mocks for:
// - BankAccount, ACHTransfer, ACHStatus
// - WalletProvider, WalletSession
// - Subscription, BillingPlan
// - Invoice, LineItem, TaxCalculation
// - Payout, PayoutMethod, PayoutSchedule
// - FraudSignal, RiskScore
// - Report, Analytics
// ...and dozens more
 
// Compilation: 18 minutes (any payment change)
// Test setup: 4,000 lines of mock configuration
// Deployment: All 23 services must deploy together

payment-service-after.ts
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
// THE SOLUTION: Segregated payment interfaces
// Each consumer only gets what they need
 
// Core card processing - most common use case
interface ICardProcessor {
    tokenizeCard(card: CardDetails): CardToken;
    chargeCard(token: CardToken, amount: Money): ChargeResult;
    voidCharge(chargeId: string): VoidResult;
    captureCharge(chargeId: string, amount?: Money): CaptureResult;
    refundCharge(chargeId: string, amount?: Money): RefundResult;
}
// Types: CardDetails, CardToken, Money, ChargeResult, VoidResult, CaptureResult, RefundResult
// Total: 7 types
 
// Bank transfers - isolated from card processing
interface IBankTransferService {
    initiateACHTransfer(source: BankAccount, dest: BankAccount, amount: Money): ACHTransfer;
    getACHStatus(transferId: string): ACHStatus;
}
// Types: BankAccount, Money, ACHTransfer, ACHStatus
// Total: 4 types
 
// Digital wallets - separate authentication flow
interface IWalletService {
    createWalletSession(provider: WalletProvider): WalletSession;
    chargeWallet(session: WalletSession, amount: Money): ChargeResult;
}
// Types: WalletProvider, WalletSession, Money, ChargeResult
// Total: 4 types
 
// Subscription management - different domain
interface ISubscriptionManager {
    createSubscription(customerId: string, planId: string): Subscription;
    updateSubscription(subId: string, planId: string): Subscription;
    cancelSubscription(subId: string): CancelResult;
}
// Types: Subscription, CancelResult
// Total: 2 types (identifiers are strings, not objects!)
 
// Invoicing - B2B focused
interface IInvoicingService {
    createInvoice(customerId: string, items: InvoiceLineItem[]): Invoice;
    sendInvoice(invoiceId: string): SendResult;
    recordPayment(invoiceId: string, payment: PaymentRecord): Invoice;
}
// Types: InvoiceLineItem, Invoice, SendResult, PaymentRecord
// Total: 4 types
 
// Payouts - Marketplace/platform focused
interface IPayoutService {
    createPayout(recipientId: string, method: PayoutMethod): Payout;
    schedulePayout(payoutId: string, schedule: PayoutSchedule): ScheduledPayout;
}
// Types: PayoutMethod, Payout, PayoutSchedule, ScheduledPayout
// Total: 4 types
 
// Fraud prevention - Security focused
interface IFraudService {
    assessFraudRisk(transactionId: string): RiskAssessment;
    reportFraud(transactionId: string, signal: FraudSignal): FraudReport;
}
// Types: RiskAssessment, FraudSignal, FraudReport
// Total: 3 types
 
// RESULT: Checkout service with ICardProcessor only
// 
// Types needed: 7 (down from 47)
// Transitive dependencies: 23 (down from 312)
// Compilation time: 47 seconds (down from 18 minutes)
// Test setup: 200 lines (down from 4,000)
// Independent deployments: Yes!

Before vs After ISP Refactoring Results
Metric	Before	After	Improvement
Interface methods	34	7 interfaces, 3-5 methods each	Clear boundaries
Types per consumer (avg)	47	4-7	85% reduction
Transitive dependencies	312	18-31	90% reduction
Build time (incremental)	18 min	45 sec	96% faster
Test mock setup	4,000 lines	150-300 lines	93% reduction
Independent deploy units	1	8	8x increase
Change containment	0%	92%	Changes isolated

Summary: Reducing Transitive Dependencies with ISP

Transitive dependencies are the hidden tax on every import statement. Fat interfaces compound this tax, turning a single dependency into hundreds of transitive ones. ISP is your primary tool for reducing this burden.

Key Takeaways

•Transitive dependencies propagate through interfaces — Every type in an interface signature creates a compile-time dependency chain.
•Fat interfaces are dependency multipliers — A single fat interface can create 50:1 or higher transitive-to-direct dependency ratios.
•Three propagation vectors exist — Type signatures, implementation requirements, and behavioral contracts all create dependency coupling.
•ISP creates dependency firewalls — Segregated interfaces block unwanted dependencies from propagating to clients.
•Measurement drives improvement — Track fan-out, closure size, recompilation impact, and deployment independence.
•Four strategies break dependency chains — Interface extraction, dependency inversion, type narrowing, and adapter interfaces.
•Results are dramatic — Proper ISP application routinely achieves 80-95% reduction in transitive dependencies.

What's Next:

Now that you understand how ISP reduces transitive dependencies, we'll explore interface-based decoupling in depth—how to design interfaces that create true architectural boundaries, enabling independent development and deployment of system components.

Page Complete

You now understand the mechanics of transitive dependency propagation and how ISP creates dependency firewalls. Apply these techniques to transform tightly-coupled dependency monoliths into loosely-coupled, independently deployable components.

1 / 4

Loading learning content...

System Design (LLD)ISP and Dependency Management

ISP and Dependency Management

LevelIntermediate

Duration90 mins

TopicISP and Dependency Management

1 / 4

Reducing Transitive Dependencies

The Hidden Cost of Every Import Statement

What You Will Learn

Understanding Transitive Dependencies

Before we can reduce transitive dependencies, we must deeply understand what they are, how they arise, and why they become problematic as systems scale.

The Dependency Graph Reality:

When you write import UserService from './user-service', you're not just importing UserService. You're also importing:

Everything UserService imports
Everything those imports import
Recursively, until you reach the leaves of your dependency tree

transitive-dependency-example.ts
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
// Level 1: Your code
// payment-processor.ts
import { OrderService } from './order-service';
 
class PaymentProcessor {
    constructor(private orderService: OrderService) {}
    
    processPayment(orderId: string): void {
        const order = this.orderService.getOrder(orderId);
        // Process payment...
    }
}
 
// Level 2: OrderService (your direct dependency)
// order-service.ts
import { UserService } from './user-service';
import { InventoryService } from './inventory-service';
import { NotificationService } from './notification-service';
import { AuditLogger } from './audit-logger';
import { PromotionEngine } from './promotion-engine';
 
export class OrderService {
    constructor(
        private userService: UserService,
        private inventoryService: InventoryService,
        private notificationService: NotificationService,
        private auditLogger: AuditLogger,
        private promotionEngine: PromotionEngine
    ) {}
    
    getOrder(orderId: string): Order { /* ... */ }
    createOrder(/* ... */): Order { /* ... */ }
    updateInventory(/* ... */): void { /* ... */ }
    notifyUser(/* ... */): void { /* ... */ }
    calculateDiscounts(/* ... */): Discount[] { /* ... */ }
}
 
// Level 3+: Each of those services has its own dependencies...
// UserService → DatabaseConnection, CacheService, AuthProvider, ...
// InventoryService → WarehouseAPI, SupplierIntegration, ...
// NotificationService → EmailProvider, SMSGateway, PushService, ...
 
// Your PaymentProcessor now transitively depends on:
// - Database drivers
// - Email services
// - SMS gateways
// - Warehouse APIs
// - Third-party auth providers
// - And dozens more...

The Compilation Cascade

How Fat Interfaces Multiply Dependencies

The Mathematics of Interface Bloat:

This isn't just arithmetic. In practice, fat interfaces create combinatorial explosion of the dependency graph.

fat-interface-dependency-explosion.ts
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
// BAD: Fat interface forces all clients to depend on ALL types
interface IOrderManagement {
    // Order queries - depends on: Order, OrderFilter, PagedResult
    getOrder(id: string): Order;
    queryOrders(filter: OrderFilter): PagedResult<Order>;
    
    // Order mutations - depends on: CreateOrderRequest, UpdateOrderRequest
    createOrder(request: CreateOrderRequest): Order;
    updateOrder(id: string, request: UpdateOrderRequest): Order;
    
    // Inventory operations - depends on: InventoryItem, WarehouseLocation
    checkInventory(productId: string): InventoryItem;
    reserveInventory(orderId: string, location: WarehouseLocation): void;
    
    // Shipping - depends on: ShippingCarrier, TrackingInfo, Address
    calculateShipping(carrier: ShippingCarrier, address: Address): Money;
    getTracking(orderId: string): TrackingInfo;
    
    // Billing - depends on: PaymentMethod, Invoice, TaxCalculation
    processPayment(orderId: string, payment: PaymentMethod): Invoice;
    calculateTax(order: Order): TaxCalculation;
    
    // Notifications - depends on: NotificationChannel, NotificationTemplate
    sendConfirmation(orderId: string, channel: NotificationChannel): void;
    sendShippingUpdate(orderId: string, template: NotificationTemplate): void;
}
 
// A simple order display component only needs getOrder()
// But it now transitively depends on:
// - WarehouseLocation (from inventory)
// - ShippingCarrier (from shipping)  
// - PaymentMethod (from billing)
// - NotificationChannel (from notifications)
// - Plus everything THOSE types depend on
 
class OrderDisplayComponent {
    // We only call ONE method
    constructor(private orderManagement: IOrderManagement) {}
    
    showOrder(id: string): void {
        const order = this.orderManagement.getOrder(id);
        this.render(order);
    }
}
 
// If ShippingCarrier changes, OrderDisplayComponent must recompile
// even though it has absolutely nothing to do with shipping!

Dependency Impact Analysis: Fat vs Segregated Interfaces
Metric	Fat Interface	Segregated Interfaces	Impact
Types exposed to each client	12 domain types	2-3 relevant types only	~80% reduction
Recompilation triggers	Any of 12 types changes	Only when relevant types change	~85% fewer builds
Test dependencies	All 12 types must be stubbed	Only used types stubbed	~75% less test setup
Deployment coupling	All clients deploy together	Independent deployment per client	100% deployment isolation
Breaking change surface	Any method signature change	Only changes to used methods	~90% reduction

The Dependency Ratio

The Mechanics of Dependency Propagation

Vector 1: Type Signature Dependencies

Vector 2: Implementation Requirements

Vector 3: Behavioral Contract Dependencies

dependency-propagation-vectors.ts
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
// VECTOR 1: Type Signature Dependencies
// Every type in the interface creates a hard compile-time dependency
 
interface IDocumentManager {
    // Anyone depending on IDocumentManager must know what these types are:
    createDocument(content: DocumentContent): Document;
    //             ↑ depends on DocumentContent  ↑ depends on Document
    
    convertToPdf(doc: Document): PdfDocument;
    //                           ↑ depends on PdfDocument
    
    signDocument(doc: Document, cert: SigningCertificate): SignedDocument;
    //                               ↑ depends on SigningCertificate
    //                                                    ↑ depends on SignedDocument
    
    archiveDocument(doc: Document, policy: ArchivalPolicy): ArchiveReceipt;
    //                                   ↑ depends on ArchivalPolicy
    //                                                      ↑ depends on ArchiveReceipt
}
 
// Even if a client only reads documents, they depend on:
// - SigningCertificate (cryptography libraries)
// - ArchivalPolicy (compliance frameworks)
// - PdfDocument (PDF processing libraries)
 
 
// VECTOR 2: Implementation Requirements
// The implementation must satisfy ALL methods, pulling in ALL dependencies
 
class DocumentManagerImpl implements IDocumentManager {
    constructor(
        private storage: StorageService,      // For createDocument
        private pdfConverter: PdfConverter,    // For convertToPdf
        private cryptoService: CryptoService,  // For signDocument
        private archiveClient: ArchiveClient,  // For archiveDocument
    ) {}
    
    // Even if no client ever calls signDocument(),
    // this class must still inject CryptoService!
    // That means the entire dependency tree of CryptoService
    // is pulled into this implementation.
}
 
 
// VECTOR 3: Behavioral Contract Dependencies
// Understanding how methods interact creates implicit dependencies
 
interface IInventoryManager {
    getStock(productId: string): StockLevel;
    reserveStock(productId: string, quantity: number): Reservation;
    releaseReservation(reservationId: string): void;
    adjustStock(productId: string, delta: number): StockLevel;
}
 
// A client calling getStock() must understand:
// - How reserveStock() affects reported stock levels
// - Whether adjustStock() can make stock negative
// - How long reservations last before auto-release
// 
// These behavioral dependencies aren't in the types, but they're real.

Signs of Dependency Propagation Problems

•Import lists exceeding 20 lines — When a file imports dozens of types it doesn't directly use, it's suffering from transitive type dependencies through fat interfaces.
•Mock objects requiring 50+ method stubs — Large mock setup indicates fat interfaces forcing test dependencies on unused functionality.
•'Diamond dependency' conflicts — When transitive dependencies create version conflicts (e.g., two paths to the same library at different versions).
•Build times scaling non-linearly — Small changes trigger massive recompilation because of deep transitive dependency trees.
•Deployment packages that dwarf actual code — Runtime dependencies ballooning because implementations must satisfy unused interface methods.

ISP as a Dependency Firewall

The Firewall Principle:

This creates compartmentalization. Instead of one massive dependency blob where everything affects everything, you get isolated dependency clusters where changes are contained.

before-isp.ts
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
// WITHOUT ISP: No dependency firewalls
// All clients depend on all types
 
interface IOrderSystem {
    // Query operations
    getOrder(id: string): Order;
    
    // Shipping operations
    ship(id: string, carrier: ShippingCarrier): TrackingInfo;
    
    // Payment operations
    charge(id: string, method: PaymentMethod): Receipt;
    
    // Reporting operations
    generateReport(params: ReportParams): Report;
}
 
// Client only needs order queries:
class OrderViewer {
    constructor(
        private orders: IOrderSystem  // Full dependency exposure!
    ) {}
}
 
// OrderViewer transitively depends on:
// - ShippingCarrier
// - TrackingInfo
// - PaymentMethod
// - Receipt
// - ReportParams
// - Report
// Plus all THEIR dependencies...

after-isp.ts
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
// WITH ISP: Dependency firewalls established
 
// Each interface is a firewall
interface IOrderQueries {
    getOrder(id: string): Order;
}
 
interface IOrderShipping {
    ship(id: string, carrier: ShippingCarrier): TrackingInfo;
}
 
interface IOrderPayments {
    charge(id: string, method: PaymentMethod): Receipt;
}
 
interface IOrderReporting {
    generateReport(params: ReportParams): Report;
}
 
// Client only gets query dependencies:
class OrderViewer {
    constructor(
        private orders: IOrderQueries  // Firewall active!
    ) {}
}
 
// OrderViewer ONLY depends on:
// - Order
// 
// Shipping, payment, and reporting 
// dependencies are BLOCKED by the firewall.

Dependency Containment Ratio

Creating Effective Firewalls:

Not all interface splits create effective firewalls. To maximize containment:

Group by dependency cluster — Methods that share the same type dependencies belong together. createOrder() and updateOrder() likely share OrderRequest types; keep them in one interface.
Separate by infrastructure — Methods requiring different infrastructure (database, network, filesystem) should be in different interfaces. This prevents infrastructure dependencies from crossing boundaries.
Isolate external systems — Methods that integrate with external services should be in dedicated interfaces. If the shipping API changes, only IOrderShipping clients need updates.
Consider stability — Put stable operations in separate interfaces from volatile ones. If payment methods change frequently but order queries are stable, separate them.

Measuring Transitive Dependency Reduction

Metric 1: Dependency Fan-Out

For each interface, count the unique types in all method signatures. A fat interface might have 20+ types; after segregation, each focused interface should have 3-5.

Metric 2: Transitive Closure Size

Metric 3: Recompilation Impact Score

For each interface change, measure how many modules require recompilation. Track this over time—it should decrease as you apply ISP.

Metric 4: Deployment Independence

Count how many services can be deployed independently. Before ISP: if core interfaces change, all services redeploy. After ISP: only affected clients redeploy.

dependency-metrics.ts
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
// Tools and techniques for measuring dependency reduction
 
// 1. Calculate direct type dependencies per interface
function calculateTypeFanOut(interfaceDeclaration: InterfaceNode): number {
    const types = new Set<string>();
    
    for (const method of interfaceDeclaration.methods) {
        // Count parameter types
        for (const param of method.parameters) {
            types.add(param.type.name);
        }
        // Count return type
        types.add(method.returnType.name);
    }
    
    return types.size;
}
 
// Before ISP: IOrderManagement has 12 unique types
// After ISP:  IOrderQueries has 3, IOrderShipping has 4, etc.
 
 
// 2. Measure transitive closure growth
interface DependencyMetrics {
    module: string;
    directDependencies: number;
    transitiveDependencies: number;
    closureRatio: number;  // transitive / direct
}
 
function analyzeModule(module: Module): DependencyMetrics {
    const directDeps = module.imports.length;
    const transitiveDeps = computeTransitiveClosure(module).size;
    
    return {
        module: module.name,
        directDependencies: directDeps,
        transitiveDependencies: transitiveDeps,
        closureRatio: transitiveDeps / directDeps
    };
}
 
// BEFORE ISP:
// PaymentProcessor: { direct: 1, transitive: 147, ratio: 147 }
// 
// AFTER ISP:
// PaymentProcessor: { direct: 1, transitive: 12, ratio: 12 }
 
 
// 3. Track recompilation impact
interface ChangeImpact {
    changedModule: string;
    affectedModules: string[];
    recompilationCount: number;
    percentageOfCodebase: number;
}
 
function assessChangeImpact(changedModule: Module): ChangeImpact {
    const affected = findAllDependents(changedModule);
    return {
        changedModule: changedModule.name,
        affectedModules: affected.map(m => m.name),
        recompilationCount: affected.length,
        percentageOfCodebase: (affected.length / totalModules) * 100
    };
}
 
// Tracking over time:
// Week 1: Average impact = 340 modules (34% of codebase)
// Week 4: Average impact = 89 modules (8.9% of codebase)
// Week 8: Average impact = 23 modules (2.3% of codebase)

Real-World Metrics: Before and After ISP Refactoring
System	Metric	Before ISP	After ISP	Improvement
E-commerce Platform	Avg. transitive deps per module	234	41	82% reduction
E-commerce Platform	Build time (incremental)	8.4 min	47 sec	91% faster
Payment Gateway	Closure ratio	67:1	8:1	88% reduction
Payment Gateway	Independent deployable units	1	7	7x increase
CMS Backend	Recompilation on core change	847 files	112 files	87% contained
CMS Backend	Test setup time	12.3 sec	1.8 sec	85% faster

Strategies for Breaking Dependency Chains

Reducing transitive dependencies requires systematic strategies. Here are the proven techniques used to break dependency chains in production systems.

Strategy 1: Interface Extraction

Strategy 2: Dependency Inversion at Boundaries

At module boundaries, invert dependencies so that high-level modules own the interfaces they depend on. This prevents low-level implementation details from propagating upward.

Strategy 3: Type Narrowing

When a method returns a large object but clients only need a subset, create a narrower return type. Instead of returning Order with 50 fields, return OrderSummary with 5 fields.

Strategy 4: Adapter Interfaces

Create adapter interfaces that translate between internal rich types and external minimal types. Clients depend on the minimal adapter interface, not the full internal types.

dependency-breaking-strategies.ts
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
83
84
85
86
87
// STRATEGY 1: Interface Extraction
// Start with fat interface, extract based on usage analysis
 
// Original fat interface
interface IUserService {
    getUser(id: string): User;
    updateUser(id: string, data: UpdateUserRequest): User;
    deleteUser(id: string): void;
    authenticateUser(credentials: Credentials): AuthToken;
    resetPassword(userId: string, request: PasswordResetRequest): void;
    getUserPermissions(userId: string): Permission[];
    generateReport(params: UserReportParams): UserReport;
}
 
// Step 1: Analyze client usage
// - ProfileView only calls getUser()
// - AuthController calls authenticateUser(), resetPassword()
// - AdminDashboard calls all methods
 
// Step 2: Extract client-specific interfaces
interface IUserReader {
    getUser(id: string): User;
}
 
interface IUserAuthenticator {
    authenticateUser(credentials: Credentials): AuthToken;
    resetPassword(userId: string, request: PasswordResetRequest): void;
}
 
interface IUserAdmin extends IUserReader, IUserAuthenticator {
    updateUser(id: string, data: UpdateUserRequest): User;
    deleteUser(id: string): void;
    getUserPermissions(userId: string): Permission[];
    generateReport(params: UserReportParams): UserReport;
}
 
// STRATEGY 3: Type Narrowing
// Reduce exposed types by returning minimal projections
 
// Before: Returns full Order with all dependencies
interface IOrderQueries {
    getOrder(id: string): Order;  // Order has 50+ fields including nested objects
}
 
// After: Returns only what's needed
interface IOrderSummaryQueries {
    getOrderSummary(id: string): OrderSummary;
}
 
interface OrderSummary {
    id: string;
    status: OrderStatus;
    totalAmount: Money;
    createdAt: Date;
    // Only 4 fields instead of 50
    // No nested objects that pull in more dependencies
}
 
 
// STRATEGY 4: Adapter Interfaces
// Create a minimal surface for external consumers
 
// Internal rich type (many dependencies)
interface InternalOrder {
    id: string;
    customer: Customer;          // Pulls in Customer dependencies
    items: OrderItem[];          // Pulls in Product dependencies
    shipping: ShippingDetails;   // Pulls in Address, Carrier dependencies
    payment: PaymentInfo;        // Pulls in PaymentMethod dependencies
    // ... more fields
}
 
// Adapter interface for external consumers
interface IExternalOrderAdapter {
    getOrderForDisplay(id: string): OrderDisplayDTO;
}
 
interface OrderDisplayDTO {
    id: string;
    customerName: string;        // String, not Customer object
    itemCount: number;           // Number, not OrderItem[]
    formattedTotal: string;      // String, not Money object
    statusLabel: string;         // String, not enum
}
 
// External consumers depend only on strings and primitives
// Zero transitive dependencies from domain model!

Dependency Breaking Decision Tree

•Identify the pain point — Which client is suffering from dependency bloat? Measure their transitive dependency count.
•Analyze actual usage — What methods does this client actually call? Use static analysis or runtime tracing.
•Check for type narrowing opportunities — Can return types be simplified to DTOs or projections?
•Extract minimal interface — Create an interface containing only the methods and types the client needs.
•Implement adapter if needed — If types can't be narrowed, create an adapter that translates to minimal types.
•Migrate client — Update the client to depend on the new minimal interface.
•Measure improvement — Compare transitive dependency count before and after.

Case Study: Breaking a Dependency Monolith

Initial State:

IPaymentService with 34 methods
47 different types in method signatures
Average consumer only used 3-4 methods
But every consumer had 312 transitive dependencies
Build time for any payment change: 18 minutes
Every microservice in the platform depended on payment types

payment-service-before.ts
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
// THE PROBLEM: Payment service interface before ISP
// 34 methods, 47 types, every consumer gets everything
 
interface IPaymentService {
    // Card processing (depends on: Card, CardToken, CardNetwork, ...)
    tokenizeCard(card: CardDetails): CardToken;
    chargeCard(token: CardToken, amount: Money): ChargeResult;
    voidCharge(chargeId: string): VoidResult;
    captureCharge(chargeId: string, amount?: Money): CaptureResult;
    refundCharge(chargeId: string, amount?: Money): RefundResult;
    
    // Bank transfers (depends on: BankAccount, RoutingNumber, ACHDetails, ...)
    initiateACHTransfer(source: BankAccount, dest: BankAccount, amount: Money): ACHTransfer;
    getACHStatus(transferId: string): ACHStatus;
    
    // Digital wallets (depends on: WalletProvider, WalletToken, ApplePaySession, ...)
    createWalletSession(provider: WalletProvider): WalletSession;
    chargeWallet(session: WalletSession, amount: Money): ChargeResult;
    
    // Subscriptions (depends on: Subscription, BillingPlan, SubscriptionEvent, ...)
    createSubscription(customer: Customer, plan: BillingPlan): Subscription;
    updateSubscription(subId: string, plan: BillingPlan): Subscription;
    cancelSubscription(subId: string): CancelResult;
    
    // Invoicing (depends on: Invoice, LineItem, TaxCalculation, ...)
    createInvoice(customer: Customer, items: LineItem[]): Invoice;
    sendInvoice(invoiceId: string): SendResult;
    recordPayment(invoiceId: string, payment: PaymentRecord): Invoice;
    
    // Payouts (depends on: Payout, PayoutMethod, PayoutSchedule, ...)
    createPayout(recipient: Recipient, method: PayoutMethod): Payout;
    schedulePayout(payout: Payout, schedule: PayoutSchedule): ScheduledPayout;
    
    // Fraud (depends on: FraudSignal, RiskScore, FraudRule, ...)
    assessFraudRisk(transaction: Transaction): RiskAssessment;
    reportFraud(transactionId: string, signal: FraudSignal): FraudReport;
    
    // Reporting (depends on: Report, ReportParams, DateRange, ...)
    generateTransactionReport(params: ReportParams): Report;
    getPaymentAnalytics(dateRange: DateRange): Analytics;
    
    // ... 16 more methods
}
 
// IMPACT: A simple checkout service that only needs chargeCard()
// must import and potentially instantiate mocks for:
// - BankAccount, ACHTransfer, ACHStatus
// - WalletProvider, WalletSession
// - Subscription, BillingPlan
// - Invoice, LineItem, TaxCalculation
// - Payout, PayoutMethod, PayoutSchedule
// - FraudSignal, RiskScore
// - Report, Analytics
// ...and dozens more
 
// Compilation: 18 minutes (any payment change)
// Test setup: 4,000 lines of mock configuration
// Deployment: All 23 services must deploy together

payment-service-after.ts
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
// THE SOLUTION: Segregated payment interfaces
// Each consumer only gets what they need
 
// Core card processing - most common use case
interface ICardProcessor {
    tokenizeCard(card: CardDetails): CardToken;
    chargeCard(token: CardToken, amount: Money): ChargeResult;
    voidCharge(chargeId: string): VoidResult;
    captureCharge(chargeId: string, amount?: Money): CaptureResult;
    refundCharge(chargeId: string, amount?: Money): RefundResult;
}
// Types: CardDetails, CardToken, Money, ChargeResult, VoidResult, CaptureResult, RefundResult
// Total: 7 types
 
// Bank transfers - isolated from card processing
interface IBankTransferService {
    initiateACHTransfer(source: BankAccount, dest: BankAccount, amount: Money): ACHTransfer;
    getACHStatus(transferId: string): ACHStatus;
}
// Types: BankAccount, Money, ACHTransfer, ACHStatus
// Total: 4 types
 
// Digital wallets - separate authentication flow
interface IWalletService {
    createWalletSession(provider: WalletProvider): WalletSession;
    chargeWallet(session: WalletSession, amount: Money): ChargeResult;
}
// Types: WalletProvider, WalletSession, Money, ChargeResult
// Total: 4 types
 
// Subscription management - different domain
interface ISubscriptionManager {
    createSubscription(customerId: string, planId: string): Subscription;
    updateSubscription(subId: string, planId: string): Subscription;
    cancelSubscription(subId: string): CancelResult;
}
// Types: Subscription, CancelResult
// Total: 2 types (identifiers are strings, not objects!)
 
// Invoicing - B2B focused
interface IInvoicingService {
    createInvoice(customerId: string, items: InvoiceLineItem[]): Invoice;
    sendInvoice(invoiceId: string): SendResult;
    recordPayment(invoiceId: string, payment: PaymentRecord): Invoice;
}
// Types: InvoiceLineItem, Invoice, SendResult, PaymentRecord
// Total: 4 types
 
// Payouts - Marketplace/platform focused
interface IPayoutService {
    createPayout(recipientId: string, method: PayoutMethod): Payout;
    schedulePayout(payoutId: string, schedule: PayoutSchedule): ScheduledPayout;
}
// Types: PayoutMethod, Payout, PayoutSchedule, ScheduledPayout
// Total: 4 types
 
// Fraud prevention - Security focused
interface IFraudService {
    assessFraudRisk(transactionId: string): RiskAssessment;
    reportFraud(transactionId: string, signal: FraudSignal): FraudReport;
}
// Types: RiskAssessment, FraudSignal, FraudReport
// Total: 3 types
 
// RESULT: Checkout service with ICardProcessor only
// 
// Types needed: 7 (down from 47)
// Transitive dependencies: 23 (down from 312)
// Compilation time: 47 seconds (down from 18 minutes)
// Test setup: 200 lines (down from 4,000)
// Independent deployments: Yes!

Before vs After ISP Refactoring Results
Metric	Before	After	Improvement
Interface methods	34	7 interfaces, 3-5 methods each	Clear boundaries
Types per consumer (avg)	47	4-7	85% reduction
Transitive dependencies	312	18-31	90% reduction
Build time (incremental)	18 min	45 sec	96% faster
Test mock setup	4,000 lines	150-300 lines	93% reduction
Independent deploy units	1	8	8x increase
Change containment	0%	92%	Changes isolated

Summary: Reducing Transitive Dependencies with ISP

Key Takeaways

•Transitive dependencies propagate through interfaces — Every type in an interface signature creates a compile-time dependency chain.
•Fat interfaces are dependency multipliers — A single fat interface can create 50:1 or higher transitive-to-direct dependency ratios.
•Three propagation vectors exist — Type signatures, implementation requirements, and behavioral contracts all create dependency coupling.
•ISP creates dependency firewalls — Segregated interfaces block unwanted dependencies from propagating to clients.
•Measurement drives improvement — Track fan-out, closure size, recompilation impact, and deployment independence.
•Four strategies break dependency chains — Interface extraction, dependency inversion, type narrowing, and adapter interfaces.
•Results are dramatic — Proper ISP application routinely achieves 80-95% reduction in transitive dependencies.

What's Next:

Page Complete

1 / 4