Definition — Talk Only to Friends - Law Of Demeter | OneNoughtOne

0/246

System Design (HLD)Other Essential Design Principles

Law of Demeter

LevelIntermediate

Duration60 mins

TopicOther Essential Design Principles

1 / 4

Definition — Talk Only to Friends

The Stranger Danger Problem in Code

Consider walking into a friend's house and, instead of asking your friend to make you a cup of coffee, you wander into their kitchen, open their cabinets, find their coffee maker, rummage through their drawers for filters, and start brewing yourself. In the physical world, this behavior is considered bizarre, intrusive, and a violation of social norms.

Yet this is exactly how most software is written.

Objects routinely reach through other objects, navigate deep into their internal structures, manipulate things they don't own, and make assumptions about implementation details they have no business knowing. The result? Tightly coupled systems that resist change, propagate failures in unexpected ways, and become maintenance nightmares.

The Law of Demeter addresses this problem directly. It's a design guideline that formalizes what we intuitively understand in social contexts: you should only interact with things you directly know about, not the friends of your friends.

What You Will Learn

By the end of this page, you will understand the formal definition of the Law of Demeter, the intuitive mental model behind it, why it was formulated, and how it fundamentally shapes the way objects should interact in well-designed systems. You'll learn to recognize LoD as not merely a coding style preference but as a profound architectural principle with far-reaching implications for maintainability, testability, and system evolution.

The Origin and Formal Definition

The Law of Demeter, also known as the Principle of Least Knowledge, was formulated in 1987 by Ian Holland at Northeastern University while working on the Demeter Project—a research initiative focused on aspect-oriented programming and adaptive software. The name "Demeter" comes from the Greek goddess of agriculture, reflecting the project's focus on "growing" software in an adaptive manner.

The principle emerged from Holland's observation that systems with limited, focused object interactions were significantly easier to modify, debug, and extend than systems where objects freely navigated complex object graphs.

The Formal Statement

The Law of Demeter for Methods states that a method M of object O should only call methods belonging to:

O itself — the object's own methods
M's parameters — objects passed directly to the method
Objects created within M — objects instantiated by the method
O's direct component objects — objects held in O's instance variables
Global objects accessible in M's scope — though this is rarely recommended

In essence: An object should only talk to its immediate friends, never to strangers.

Understanding "Friends" vs "Strangers":

Friends are objects that a method legitimately knows about: its own object, its parameters, objects it creates, and direct collaborators held as instance fields.
Strangers are objects obtained through friends—the results of method calls on friends, especially those that return internal objects that the caller then manipulates.

The critical insight is that strangers are not your responsibility. When you interact with a stranger, you're making assumptions about its interface, its behavior, and crucially, its continued existence and structure. These assumptions create hidden dependencies that propagate throughout the codebase.

lod-violation-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
// ❌ VIOLATING the Law of Demeter
// This method interacts with "strangers" — objects it doesn't directly know
 
class OrderProcessor {
    processOrder(customer: Customer): void {
        // We get the wallet (stranger) through the customer (friend)
        // Then we reach into the wallet to get the credit card (stranger's friend)
        // Then we reach into the credit card to get the limit (stranger's friend's detail)
        const creditLimit = customer.getWallet().getCreditCard().getLimit();
        
        if (creditLimit > this.orderTotal) {
            // We're navigating even deeper into unknown territory
            customer.getWallet().getCreditCard().charge(this.orderTotal);
            
            // And again, reaching through multiple objects
            customer.getAddress().getCity().getShippingZone().calculateDeliveryTime();
        }
    }
}
 
// What's wrong here?
// 1. OrderProcessor knows about Wallet, CreditCard, Address, City, ShippingZone
// 2. If ANY of these classes change their structure, OrderProcessor breaks
// 3. OrderProcessor cannot be tested without all these collaborators
// 4. The coupling tree is deep and fragile

lod-compliant-example.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
// ✅ FOLLOWING the Law of Demeter
// This method only talks to its direct friends
 
class OrderProcessor {
    processOrder(customer: Customer): void {
        // Ask the customer directly — they handle their own internals
        if (customer.canAfford(this.orderTotal)) {
            customer.charge(this.orderTotal);
            
            // Ask the customer about delivery — they know their own address
            const deliveryTime = customer.estimateDeliveryTime();
        }
    }
}
 
// What's better here?
// 1. OrderProcessor only knows about Customer
// 2. Customer's internal structure can change freely
// 3. OrderProcessor can be tested with a simple Customer mock
// 4. The coupling is shallow and resilient

The Mental Model — Trust Boundaries

While the formal definition provides precise rules, the Law of Demeter is best understood through a mental model of trust boundaries. Every object in your system has a boundary—a clear demarcation between what it knows about and what it doesn't.

When you violate LoD, you're reaching across trust boundaries into territory you shouldn't know about. You're making yourself responsible for things that aren't your concern, and you're binding yourself to implementation details that may change.

The "Don't Ask, Tell" Perspective:

LoD is intimately connected to the principle of "Tell, Don't Ask" (which we'll cover later in this chapter). When you violate LoD, you're typically asking an object for its internals so you can make decisions yourself. When you follow LoD, you tell the object what you need and let it figure out how to accomplish it.

This shift from interrogation to delegation is fundamental to object-oriented design. Objects should be autonomous agents that handle their own responsibilities, not passive data containers that expose their guts for external manipulation.

LoD Violation Pattern

•Asking for data: "Give me your wallet"
•Navigating structure: "Let me reach into you"
•Making decisions externally: "I'll figure out what to do"
•Coupling to internals: "I know how you're organized"
•Fragile dependencies: "Hope you don't change"

LoD Compliant Pattern

•Telling, not asking: "Please charge this amount"
•Respecting boundaries: "You handle your internals"
•Delegating decisions: "You know best how to do this"
•Coupling to interface: "I only know your public face"
•Resilient dependencies: "Change freely inside"

The Newspaper Metaphor

Imagine you want to know if your neighbor received their newspaper today. Violating LoD would be: walking into their house, finding their bedroom, locating their nightstand, checking if a newspaper is on it. Following LoD would be: asking your neighbor directly, "Did you get your newspaper today?" You don't need to know their house layout, and if they rearrange their furniture, your question still works.

The Five Allowed Interactions

Let's examine each of the five categories of "friends" that the Law of Demeter permits you to interact with. Understanding these categories precisely is essential for correctly applying the principle.

Category 1: The Object Itself (this/self)

•A method can always call other methods on its own object
•This is the most fundamental and obvious case
•this.calculateTotal() — always allowed
•The object is always its own most intimate friend

self-interaction.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
class ShoppingCart {
    private items: CartItem[] = [];
    
    calculateTotal(): number {
        // ✅ Calling another method on 'this' — always allowed
        return this.items.reduce((sum, item) => sum + this.calculateItemPrice(item), 0);
    }
    
    private calculateItemPrice(item: CartItem): number {
        // ✅ Using 'this' to access discount logic
        return item.basePrice * this.getDiscountRate();
    }
    
    private getDiscountRate(): number {
        // ✅ Self-interaction is always permissible
        return this.items.length > 5 ? 0.9 : 1.0;
    }
}

Category 2: Method Parameters

•Objects passed directly to a method are legitimate friends
•The caller has explicitly introduced you to these objects
•They're part of the method's contract
•But you can only interact with these objects, not objects you get through them

parameter-interaction.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
class PaymentProcessor {
    // 'paymentMethod' is a parameter — it's a friend
    processPayment(order: Order, paymentMethod: PaymentMethod): void {
        // ✅ Calling methods on the parameter 'paymentMethod' — allowed
        if (paymentMethod.isValid()) {
            // ✅ Calling methods on the parameter 'order' — allowed
            const amount = order.getTotal();
            paymentMethod.charge(amount);
        }
        
        // ❌ But this would violate LoD:
        // order.getCustomer().getEmail().getDomain().validate();
        // We're reaching through 'order' to objects we weren't introduced to
    }
}

Category 3: Objects Created Within the Method

•If you create an object, you implicitly know about it
•Factory methods, instantiations, builder results — all fair game
•You created them, so you're responsible for them
•This is why factories and builders don't violate LoD

created-object-interaction.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
class ReportGenerator {
    generateReport(data: ReportData): void {
        // ✅ We create the formatter — it's our friend
        const formatter = new ReportFormatter();
        
        // ✅ We create the output — it's our friend  
        const output = new StringBuilder();
        
        // ✅ Interacting with objects we created is always allowed
        output.append(formatter.formatHeader(data.getTitle()));
        output.append(formatter.formatBody(data.getContent()));
        
        // ✅ We create the file writer — it's our friend
        const writer = new FileWriter("/reports/output.txt");
        writer.write(output.toString());
        writer.close();
    }
}

Category 4: Direct Component Objects (Instance Fields)

•Objects held as instance variables are part of your object's composition
•They're your permanent collaborators — you've committed to knowing them
•This is fundamental to object composition
•But again, only interact with them, not objects you get through them

component-interaction.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
class EmailService {
    // These are direct components — permanent friends
    private readonly templateEngine: TemplateEngine;
    private readonly smtpClient: SmtpClient;
    private readonly logger: Logger;
    
    constructor(
        templateEngine: TemplateEngine,
        smtpClient: SmtpClient,
        logger: Logger
    ) {
        this.templateEngine = templateEngine;
        this.smtpClient = smtpClient;
        this.logger = logger;
    }
    
    sendWelcomeEmail(user: User): void {
        // ✅ Interacting with our component objects — allowed
        this.logger.info("Sending welcome email");
        
        const body = this.templateEngine.render("welcome", { name: user.getName() });
        
        // ✅ Using user (parameter) and smtpClient (component)
        this.smtpClient.send(user.getEmail(), "Welcome!", body);
    }
}

Category 5: Global Objects (Use Sparingly)

•Technically allowed, but widely considered problematic
•Singletons, static service locators, global registries
•They create hidden dependencies and hurt testability
•Prefer dependency injection over global access

Category 5 Is Controversial

While the original Law of Demeter permits interacting with globally accessible objects, modern design wisdom discourages this practice. Globals create hidden dependencies, make testing difficult, and introduce implicit coupling. In contemporary practice, dependency injection is preferred over global access.

The "One Dot" Rule — A Common Simplification

You may have heard the Law of Demeter summarized as the "One Dot Rule" or "No Method Chaining": only use one dot in a method call chain.

This simplification is:

Useful as a quick heuristic for spotting potential violations
Incomplete because it doesn't capture the full principle
Sometimes Wrong because some chains are perfectly fine

Let's understand when this heuristic helps and when it misleads.

one-dot-heuristic.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
// The "One Dot Rule" says these are suspicious:
 
// ❌ Clear LoD violations (multiple dots reaching through objects)
customer.getWallet().getCreditCard().charge(amount);
order.getShippingAddress().getCity().getTimeZone();
employee.getDepartment().getManager().getEmail();
 
// ✓ These would pass the "One Dot Rule"
customer.charge(amount);
order.getDeliveryTimeZone();
employee.getManagerEmail();
 
// BUT: Not all multi-dot expressions violate LoD!
 
// ✅ Fluent interfaces return 'this' — NOT an LoD violation
queryBuilder
    .select("name", "email")
    .from("users")
    .where("active", true)
    .orderBy("created_at")
    .limit(10);
 
// ✅ Building up a result from your own creation — NOT an LoD violation
new StringBuilder()
    .append("Hello, ")
    .append(name)
    .append("!")
    .toString();
 
// ✅ Navigating a collection you received — context-dependent
// If 'items' is a parameter or component, this may be fine
items.filter(i => i.isActive()).map(i => i.getName());

The Real Question to Ask

Instead of counting dots, ask: "Am I reaching through an object to get to something inside it that I shouldn't know about?" If you're navigating into the internal structure of another object to access its collaborators, you're likely violating LoD. If you're working with an API designed for chaining (like fluent builders) or processing collections, you may be fine.

When Multiple Dots Are Acceptable
Pattern	Why It's OK	Example
Fluent Interfaces	Each method returns 'this' — no navigation through internals	builder.withName('x').withAge(20).build()
Method Chaining on Created Objects	You created it, so you own it	new Date().toISOString().substring(0, 10)
Stream/Collection Processing	You're transforming data you own	list.filter().map().sort()
Optional Unwrapping	Intentional null-safe navigation	Optional.of(x).map(fn).orElse(default)
DSL/Query Builders	Designed for chained configuration	query.select().from().where()

Why LoD Matters — The Deeper Rationale

The Law of Demeter isn't just a stylistic preference or a rule for rule's sake. It addresses fundamental problems in software design that surface as systems grow and evolve. Understanding the why behind LoD makes its application more intuitive.

Coupling Reduction

•Direct coupling is when A depends on B
•Transitive coupling is when A depends on B, which exposes C, making A indirectly dependent on C
•LoD violations create deep transitive coupling — A depends on B's structure, C's structure, D's structure...
•Every object in the chain becomes a dependency
•Changing any object in the chain can break the caller

coupling-illustration.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
// LoD violation: OrderService is coupled to 5 classes
class OrderService {
    processOrder(order: Order): void {
        // This single line creates dependencies on:
        // 1. Order
        // 2. Customer (returned by getCustomer)
        // 3. Wallet (returned by getWallet)
        // 4. CreditCard (returned by getCreditCard)
        // 5. Bank (returned by getBank)
        order.getCustomer().getWallet().getCreditCard().getBank().validate();
    }
}
 
// If Bank changes its API: OrderService breaks
// If CreditCard changes how it references Bank: OrderService breaks
// If Wallet changes how it stores CreditCard: OrderService breaks
// If Customer changes how it stores Wallet: OrderService breaks
 
// LoD-compliant: OrderService is coupled to 1 class
class OrderService {
    processOrder(order: Order): void {
        // Only depends on Order — Order handles the rest internally
        order.validatePayment();
    }
}
 
// Changes to Customer, Wallet, CreditCard, Bank don't affect OrderService
// Only Order knows about its internal payment validation logic

Encapsulation Enforcement

•LoD violations break encapsulation by exposing internal structure
•When you navigate a.getB().getC(), you're exposing that A contains B, and B contains C
•This internal structure becomes public contract — you can't change it without breaking clients
•LoD protects encapsulation by keeping internal structure private
•Objects remain free to reorganize their internals

Testability Improvement

•LoD violations require deep mock structures for testing
•You must mock A, which returns mock B, which returns mock C...
•Tests become complex, brittle, and tightly coupled to implementation
•LoD-compliant code has shallow dependencies
•Tests are simpler: mock the direct collaborator, done

testing-impact.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
// Testing LoD-violating code is painful
describe('OrderService with LoD violation', () => {
    it('processes order', () => {
        // We need to create entire mock chains
        const mockBank = { validate: jest.fn() };
        const mockCreditCard = { getBank: () => mockBank };
        const mockWallet = { getCreditCard: () => mockCreditCard };
        const mockCustomer = { getWallet: () => mockWallet };
        const mockOrder = { getCustomer: () => mockCustomer };
        
        orderService.processOrder(mockOrder);
        
        expect(mockBank.validate).toHaveBeenCalled();
    });
});
 
// Testing LoD-compliant code is simple
describe('OrderService with LoD compliance', () => {
    it('processes order', () => {
        // Just mock the direct collaborator
        const mockOrder = { validatePayment: jest.fn() };
        
        orderService.processOrder(mockOrder);
        
        expect(mockOrder.validatePayment).toHaveBeenCalled();
    });
});

The Ripple Effect Principle

When you violate LoD, any change along the access chain ripples back to the caller. A small internal refactoring in a distant class can break code that never should have known about that class in the first place. LoD creates firebreaks that contain change impact.

Recognizing Violations in Practice

Developing an intuition for LoD violations requires practice. Here are common patterns that signal potential problems, along with questions to ask yourself when reviewing code.

Red Flags in Code

•Long method chains: a.getB().getC().getD().doSomething() — reaching deep into object graphs
•Getter-heavy code: Methods that exist solely to expose internal state for external manipulation
•Return-type navigation: Calling methods on return values to navigate to other objects
•Null check chains: if (a.getB() != null && a.getB().getC() != null...) — defensive navigation
•Feature envy: A method that uses more features of another class than its own
•Data transfer orchestration: Code that extracts data from A, passes it to B, which gives it to C

Questions to Ask Yourself

•Am I reaching through this object? If I call a method and then immediately call a method on its return value to access something else, I'm likely reaching through.
•Should I know about this? If the objects I'm accessing are implementation details of another object, I probably shouldn't know about them.
•What would break if this internal structure changed? If reorganizing another class's internals would break my code, I have an LoD violation.
•Can I ask instead of navigating? If I can ask the object I have to give me what I need directly, I should.
•Am I testing internal structure or behavior? If my tests verify that A.getB().getC() returns D, I'm testing structure, not behavior.

violation-patterns.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
// Common violation patterns to recognize
 
// Pattern 1: "Train Wreck" — long chains of getters
const city = order.getCustomer().getAddress().getCity(); // ❌
 
// Pattern 2: "Interrogation" — extracting data for external decisions
if (customer.getAccount().getBalance() > amount) { // ❌
    customer.getAccount().debit(amount);
}
 
// Pattern 3: "Middle Man Bypass" — skipping encapsulation
invoice.getLineItems().get(0).getProduct().getSupplier().notify(); // ❌
 
// Pattern 4: "Hidden Cost" — null checks reveal deep coupling
if (user != null && user.getProfile() != null 
    && user.getProfile().getSettings() != null) { // ❌
    applySettings(user.getProfile().getSettings());
}
 
// Pattern 5: "Feature Envy" — method knows too much about others
class OrderPrinter {
    print(order: Order): void {
        // This method knows way too much about Order's internals
        const customer = order.getCustomer();
        const address = customer.getAddress();
        const lines = order.getLineItems();
        lines.forEach(line => {
            const product = line.getProduct();
            const price = product.getPrice();
            // ... this is Order's job, not OrderPrinter's
        });
    }
}

The Connection to "Tell, Don't Ask"

The Law of Demeter is deeply connected to the "Tell, Don't Ask" principle (which we'll explore in depth later in this chapter). Understanding this connection clarifies both principles.

When you violate LoD, you're almost always "asking" instead of "telling":

You ask an object for its collaborator
You ask that collaborator for its collaborator
You ask until you find the thing you actually want
Then you tell that distant thing to do something

When you follow LoD, you're "telling" your immediate friend:

You tell your friend what you need accomplished
Your friend figures out how to accomplish it
Your friend may tell its friends, but that's its business
You don't care about the implementation chain

tell-dont-ask-connection.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
// "Asking" pattern — violates both LoD and Tell-Don't-Ask
class PaymentService {
    processRefund(order: Order): void {
        // Ask for the payment method
        const payment = order.getPayment();
        // Ask for the gateway
        const gateway = payment.getGateway();
        // Ask for the original transaction
        const transaction = payment.getTransaction();
        // Finally tell the gateway to do something
        gateway.refund(transaction.getId(), transaction.getAmount());
    }
}
 
// "Telling" pattern — follows both LoD and Tell-Don't-Ask
class PaymentService {
    processRefund(order: Order): void {
        // Tell the order to handle the refund
        // Order knows how its payment works internally
        order.processRefund();
    }
}
 
// Inside Order, the same principle applies
class Order {
    processRefund(): void {
        // Tell payment to handle refund — Order doesn't reach into Payment
        this.payment.refund();
    }
}
 
// Inside Payment, the same principle continues
class Payment {
    refund(): void {
        // Tell gateway to refund — Payment handles its own internals
        this.gateway.refund(this.transaction.getId(), this.transaction.getAmount());
    }
}

Push Behavior to Data

When you follow LoD, you naturally push behavior closer to the data it operates on. Instead of extracting data from objects and manipulating it externally, you ask objects to perform operations themselves. This leads to richer domain models where objects have meaningful behavior, not just getters and setters.

Summary — The Law of Demeter Defined

We've established the foundational understanding of the Law of Demeter. Let's consolidate what we've learned before moving on to explore how LoD reduces coupling in practice.

Key Takeaways

•The Law of Demeter states that a method should only interact with its immediate friends: itself, its parameters, objects it creates, and its direct component objects.
•"Talk only to friends" captures the essence: don't reach through friends to interact with strangers.
•The "One Dot Rule" is a useful but imperfect heuristic — fluent interfaces and builders are exceptions, not violations.
•LoD reduces coupling by preventing transitive dependencies that spread through your codebase.
•LoD enforces encapsulation by keeping internal structure hidden from external code.
•LoD improves testability by requiring only direct collaborator mocks, not deep mock chains.
•LoD connects to "Tell, Don't Ask" — following LoD naturally leads to telling objects what to do rather than interrogating their internals.

Looking Ahead

Now that we understand what the Law of Demeter is and why it matters, the next page will dive deep into how LoD reduces coupling in practice. We'll examine concrete scenarios where LoD violations cause real problems, and how applying LoD creates more resilient, maintainable systems.