High Cohesion — Things That Belong Together - Cohesion and Coupling | OneNoughtOne

0/290

System Design (HLD)Cohesion and Coupling

Cohesion and Coupling

LevelIntermediate

Duration60 mins

TopicCohesion and Coupling

1 / 4

High Cohesion — Things That Belong Together

The Art of Keeping Related Things Together

Imagine walking into a kitchen where the knives are stored in the bathroom, the pots are kept in the bedroom closet, and the spices are scattered across six different rooms. You could still cook a meal—eventually—but every task would require wandering through the entire house. The kitchen would be functionally destroyed despite all its components existing somewhere in the building.

This absurd scenario perfectly illustrates what happens when software lacks cohesion. Code that belongs together ends up scattered across the codebase. Related functionality is split among unrelated modules. Data and the operations on that data live in separate places, forcing developers to mentally reassemble context from disparate locations.

Cohesion measures how strongly related and focused the elements within a module are. It answers a fundamental question: Do the things inside this module belong together?

What You Will Learn

This page establishes cohesion as a cornerstone of modular design. You will learn what cohesion means precisely, why it matters for maintainability and understandability, how to recognize different levels of cohesion, and concrete techniques to achieve high cohesion in your systems.

Defining Cohesion Precisely

Cohesion, as a formal concept in software engineering, emerged from the structured design movement of the 1970s. Larry Constantine and Edward Yourdon, in their seminal work on structured design, defined cohesion as a measure of the degree to which the elements inside a module belong together.

But what does "belong together" actually mean? Elements belong together when they:

Contribute to a single, well-defined purpose — Every piece of code in the module works toward the same goal
Would naturally change together — When requirements evolve, the affected code is co-located
Share a common abstraction — The elements represent different facets of the same concept
Operate on the same data — The functions manipulate related or identical data structures

Cohesion vs. Coupling: The Twin Metrics

Cohesion and coupling are inversely related: increasing cohesion within modules tends to decrease coupling between modules. When related code is gathered together (high cohesion), it needs fewer connections to other modules (low coupling). We will explore coupling in depth in the next page, but understand that these concepts work in tandem.

The formal definition:

Cohesion is the degree to which the elements of a module belong together. A module has high cohesion if its elements are strongly related to one another and work together toward a single, well-defined purpose.

This definition contains several critical insights:

"Degree" implies cohesion exists on a spectrum, not as a binary property
"Elements" refers to functions, methods, variables, types—everything inside the module
"Belong together" is ultimately about semantic relatedness, not arbitrary grouping
"Single, well-defined purpose" suggests a clear answer when asked 'What does this module do?'

Why Cohesion Matters

High cohesion is not merely an aesthetic preference or academic exercise—it delivers concrete, measurable benefits that compound over the life of a software system. Let's examine why cohesion matters from multiple perspectives.

The Practical Benefits of High Cohesion

•Understandability — A cohesive module tells a coherent story. When you open it, you immediately understand its purpose. You don't need to hold five different concepts in your head simultaneously or wonder why unrelated functions share space.
•Maintainability — Changes to one concept tend to be localized. If you need to modify how user authentication works, a cohesive design means all authentication logic lives in one place. No hunting through scattered code.
•Testability — Cohesive modules have clear boundaries and focused responsibilities. This makes them natural units for testing. You can write comprehensive tests without complex setup or brittle dependencies.
•Reusability — A module that does one thing well can be reused wherever that thing is needed. A module that does eight unrelated things is only useful in contexts that need all eight—which is rarely anywhere else.
•Change Isolation — When requirements change (and they always do), cohesive modules contain the blast radius. Modifying a cohesive module is less likely to create unexpected side effects in unrelated parts of the system.

The cognitive load argument:

Software development is fundamentally constrained by human cognitive capacity. We can only hold a limited amount of information in working memory at once—research suggests about four to seven items for most people.

High cohesion respects this limitation. When a module is cohesive, you can understand it as a single concept rather than multiple unrelated concepts sharing space. Opening a cohesive UserAuthentication module requires loading one mental model. Opening a low-cohesion UserStuff module forces you to context-switch between authentication, profile management, notification preferences, and billing—four mental models in one file.

This isn't about developer comfort; it's about error rates. Every context switch is an opportunity for mistakes. Every piece of unrelated code you must hold in memory reduces your capacity to reason about the code you're actually changing.

The Hidden Cost of Low Cohesion

Low cohesion accumulates cost invisibly. Each low-cohesion module slightly increases the time to understand, modify, and test code. Across thousands of modules and years of development, these small taxes compound into massive productivity drains. Teams rarely notice because the degradation is gradual, but the difference between a cohesive and incohesive codebase can be 10x in development velocity.

The Cohesion Spectrum

Constantine and Yourdon identified seven levels of cohesion, ranging from worst to best. Understanding this spectrum helps you recognize where your modules fall and how to improve them.

The levels are presented from lowest (worst) to highest (best) cohesion:

The Seven Levels of Cohesion
Level	Description	Example	Quality
Coincidental	Elements have no meaningful relationship; grouped arbitrarily	A 'Utils' class with date formatting, string parsing, and file compression	❌ Worst
Logical	Elements do similar things but are unrelated; grouped by category	An 'InputHandler' that handles mouse, keyboard, and network input	❌ Poor
Temporal	Elements are related by when they execute, not what they do	An 'OnStartup' module that initializes logging, DB, cache, and UI	⚠️ Weak
Procedural	Elements must execute in sequence; related by control flow	A module that validates, transforms, then saves data	⚠️ Moderate
Communicational	Elements operate on the same data	Functions that read, validate, and format a user record	✓ Acceptable
Sequential	Output of one element is input to the next	A data pipeline: parse → validate → transform → persist	✓ Good
Functional	All elements contribute to a single, well-defined function	A 'PasswordHasher' that only handles password hashing operations	✅ Best

Understanding each level in depth:

Coincidental Cohesion (Worst) — This is the absence of cohesion. Elements are grouped because someone had to put them somewhere. The classic example is the Utils or Helpers class that becomes a dumping ground for anything without an obvious home. These modules grow unboundedly, becoming harder to understand and maintain over time. There's no organizing principle—if you need to add a new utility function, it goes here, regardless of what it does.

Logical Cohesion — Elements are grouped because they do 'similar' things, but the similarity is superficial. An InputHandler might contain code for mouse events, keyboard shortcuts, touch gestures, and network socket inputs. They're all 'input'—but the implementations and concerns are entirely different. Modifying mouse handling shouldn't require navigating keyboard code.

Temporal Cohesion — Elements are grouped because they happen at the same time. Initialization code often suffers from this: 'everything that needs to happen at startup.' The elements don't conceptually belong together; they merely share a trigger. This makes it hard to reason about any single responsibility because you're constantly wading through unrelated code.

Procedural Cohesion — Elements are grouped because they execute in sequence. A procedure that validates input, calculates results, and prints output has procedural cohesion. The elements share control flow but not necessarily purpose. This is common in scripts and transaction sequences where the sequence matters more than the abstraction.

Communicational Cohesion — Elements operate on the same data or dataset. A module with functions that all manipulate the same Customer record (reading, validating, updating, formatting) has communicational cohesion. This is stronger because the data provides a natural organizing principle—but be careful that 'operates on same data' doesn't become a loophole for including unrelated operations.

Sequential Cohesion — The output of one element feeds the input of another, forming a processing chain. Data pipelines exhibit this: raw data flows through parsing, validation, transformation, and persistence. Each stage has a clear role, and the sequence is meaningful. This is strong cohesion because the elements are bound by data flow.

Functional Cohesion (Best) — All elements contribute to exactly one well-defined task. A PasswordHasher that only exposes hash(password) and verify(password, hash) has functional cohesion. Every piece of code in the module exists to serve password hashing. Changes to email formatting cannot possibly affect this module because email formatting isn't here.

Aim for Functional, Accept Communicational

In practice, purely functional cohesion is sometimes too granular. Many well-designed modules exhibit communicational cohesion—functions that operate on the same data entity. This is healthy. The key is to avoid coincidental, logical, and temporal cohesion, which indicate structural problems.

Recognizing Cohesion Levels in Real Code

Let's examine concrete code examples to develop intuition for identifying cohesion levels. The ability to quickly recognize low cohesion is a key skill for architectural review and continuous improvement.

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
// This class is a dumping ground for anything without an obvious home
class Utils {
    // String manipulation
    static formatPhoneNumber(phone: string): string { /* ... */ }
    static truncateText(text: string, maxLength: number): string { /* ... */ }
    
    // Date operations - completely unrelated to strings
    static formatDate(date: Date): string { /* ... */ }
    static addBusinessDays(date: Date, days: number): Date { /* ... */ }
    
    // Mathematical calculations - nothing to do with dates or strings
    static calculateCompoundInterest(principal: number, rate: number, time: number): number { /* ... */ }
    static convertCurrency(amount: number, rate: number): number { /* ... */ }
    
    // File operations - now we're dealing with I/O
    static readConfig(path: string): Config { /* ... */ }
    static writeLog(message: string): void { /* ... */ }
    
    // Validation - different concern entirely
    static isValidEmail(email: string): boolean { /* ... */ }
    static isValidCreditCard(number: string): boolean { /* ... */ }
}
 
// Problem: No organizing principle. Will grow unboundedly.
// Every developer adds whatever they need here because it's "convenient."

The Utils class above has coincidental cohesion—the worst kind. There's no reason these functions are together except convenience. This creates several problems:

The class will grow forever because anything can be added
Finding a specific function requires scanning through unrelated code
Changes to email validation could accidentally affect currency conversion
Testing requires either testing everything or extracting pieces

The fix is to decompose by responsibility: PhoneFormatter, DateCalculator, FinancialMath, ConfigReader, Logger, EmailValidator, CreditCardValidator. Each becomes a focused module with functional cohesion.

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
async function initializeApplication(): Promise<void> {
    // Logging setup
    configureLogging();
    setLogLevel('info');
    attachLogHandlers();
    
    // Database initialization
    await connectToDatabase();
    await runMigrations();
    await seedTestData();
    
    // Cache warming
    await loadCacheFromRedis();
    await precomputeFrequentQueries();
    
    // Feature flags
    await fetchFeatureFlags();
    await validateFlagConsistency();
    
    // External service connections
    await connectToPaymentGateway();
    await initializeEmailService();
    await setupAnalyticsTracking();
    
    // UI initialization
    renderSplashScreen();
    loadUserPreferences();
    initializeTheme();
}
 
// Problem: These things happen together (at startup) but are unrelated.
// Modifying cache logic requires navigating through database, payment, and UI code.

The initialization function above has temporal cohesion—elements grouped because they execute at the same time. This is seductive because it feels organized ('all startup code in one place'), but it creates maintenance problems:

Changing logging shouldn't require understanding database code
Testing cache initialization requires all other systems or mocking them
Adding a new startup task means modifying this central, heavily-trafficked function

A better approach: Create separate initializers (LoggingInitializer, DatabaseInitializer, CacheInitializer) with functional cohesion, then compose them in a startup orchestrator that only manages sequencing.

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
/**
 * PasswordHasher: A cohesive module with one purpose.
 * Every line exists to serve password hashing and verification.
 */
class PasswordHasher {
    private readonly algorithm: string = 'bcrypt';
    private readonly saltRounds: number = 12;
    private readonly minPasswordLength: number = 8;
    
    constructor(private readonly options?: HashingOptions) {
        if (options?.saltRounds) {
            this.saltRounds = options.saltRounds;
        }
    }
    
    /**
     * Hash a plaintext password for storage.
     */
    async hash(plaintext: string): Promise<string> {
        this.validatePasswordStrength(plaintext);
        const salt = await this.generateSalt();
        return this.computeHash(plaintext, salt);
    }
    
    /**
     * Verify a plaintext password against a stored hash.
     */
    async verify(plaintext: string, storedHash: string): Promise<boolean> {
        return this.compareHashes(plaintext, storedHash);
    }
    
    /**
     * Check if a stored hash needs rehashing (algorithm upgrade).
     */
    needsRehash(storedHash: string): boolean {
        return this.detectOutdatedAlgorithm(storedHash);
    }
    
    // Private implementation details - all serving password hashing
    private validatePasswordStrength(password: string): void { /* ... */ }
    private async generateSalt(): Promise<string> { /* ... */ }
    private async computeHash(password: string, salt: string): Promise<string> { /* ... */ }
    private async compareHashes(plaintext: string, hash: string): Promise<boolean> { /* ... */ }
    private detectOutdatedAlgorithm(hash: string): boolean { /* ... */ }
}
 
// Every method, every property, every line serves password hashing.
// You can answer "What does this class do?" in one sentence.

The Litmus Test for Cohesion

If you can describe what a module does in one sentence without using 'and' or 'or', it likely has high cohesion. 'This module handles password hashing' ✅. 'This module handles password hashing and email notifications and user preferences' ❌.

Techniques for Achieving High Cohesion

High cohesion doesn't happen by accident. It requires intentional design decisions and ongoing vigilance. Here are proven techniques for achieving and maintaining cohesive modules.

Practical Techniques for High Cohesion

•Start with behavior, not data — Define what operations the module performs before deciding what data it holds. This naturally groups behavior that belongs together. A module that 'manages user authentication' is more cohesive than a module that 'holds user-related data.'
•Apply the Single Responsibility Principle — SRP and cohesion are deeply related. If a module has one reason to change, its elements are focused on that one responsibility. Use 'reason to change' as a cohesion indicator.
•Split on domain boundaries — Real-world concepts provide natural cohesion boundaries. 'Order', 'Payment', 'Shipment', 'Inventory' are distinct concepts; mixing them in one module creates incohesion. Model your code after the domain.
•Resist convenience grouping — The 'I'll just add it here for now' pattern destroys cohesion. Require justification for why each element belongs. If it doesn't clearly fit, it belongs elsewhere.
•Refactor ruthlessly — Cohesion degrades as systems evolve. Regularly review modules and extract elements that no longer fit. Refactoring for cohesion should be routine, not exceptional.
•Use interfaces to reveal cohesion — A module's public interface should be cohesive even if internals are complex. If your interface methods seem unrelated, your module likely lacks cohesion.

The extraction pattern:

When you notice low cohesion, the remedy is extraction. Identify elements that belong together (by data, by purpose, by domain concept) and extract them into their own module. This is a mechanical process:

Identify the group — Which elements form a natural unit?
Define the boundary — What's the public interface for this group?
Extract — Move elements to a new module with that interface
Update references — Existing code now depends on the new module
Verify — The original module should now be more cohesive; the new module should have high cohesion

Repeat until all modules are cohesive. The sign that you're done: every module has a clear, single-sentence purpose.

The 'What Does This Do?' Test

Regularly ask 'What does this module do?' For every function, ask 'Why is this here instead of somewhere else?' If answers are vague or require lists, investigate. Clear answers indicate cohesion; confused answers indicate problems.

Cohesion at Different Scales

Cohesion applies at every level of software architecture, from individual functions to entire services. Understanding how cohesion manifests at each scale helps you design coherent systems top to bottom.

Cohesion Across Architectural Scales
Scale	What Cohesion Means	Warning Signs of Low Cohesion
Function/Method	The function does one thing at one abstraction level	Function name contains 'And', function body spans multiple concerns, long parameter lists
Class/Module	All methods relate to a single responsibility or concept	Class name is vague (Manager, Handler, Util), methods operate on unrelated data
Package/Namespace	All modules in the package serve a common purpose	Package is a dumping ground, modules don't interact, unclear naming
Service/Microservice	The service owns one bounded context or capability	Service handles unrelated business capabilities, requires many external dependencies
System/Application	The system solves one business problem end-to-end	Application feels like multiple products stitched together, inconsistent UX

Function-level cohesion:

A cohesive function operates at one abstraction level and accomplishes one task. It doesn't mix high-level orchestration with low-level implementation details. Consider:

// Low cohesion: Multiple abstraction levels
function processOrder(orderId: string) {
    const bytes = fs.readFileSync('/config/db.json');  // Low-level I/O
    const config = JSON.parse(bytes.toString());       // Low-level parsing
    const db = new Database(config.host, config.port); // Infrastructure
    const order = db.query('SELECT * FROM orders WHERE id = ?', orderId);
    order.status = 'processing';                       // Business logic
    sendEmail(order.email, 'Your order is processing'); // Side effect
    return order;
}

// High cohesion: One abstraction level
function processOrder(order: Order): ProcessingResult {
    validateOrderState(order);
    const result = applyProcessingRules(order);
    return result;
}

The cohesive version delegates details to other functions that each have their own cohesive focus.

Service-level cohesion (Microservices):

In microservice architectures, cohesion becomes critical at the service boundary. A cohesive service owns one bounded context from Domain-Driven Design—a distinct area of the business with its own ubiquitous language and models.

✅ Cohesive Service Boundaries:
- InventoryService: Manages stock levels, reservations, reordering
- PaymentService: Handles transactions, refunds, payment methods
- ShippingService: Calculates rates, tracks packages, manages carriers

❌ Low-Cohesion 'Order Service' (Anti-pattern):
- Creates orders (✓ appropriate)
- Processes payments (✗ should be separate service)
- Updates inventory (✗ should be separate service)  
- Sends notifications (✗ should be separate service)
- Generates invoices (✗ should be separate service)

The cohesive services can evolve independently. The low-cohesion service becomes a monolith wearing microservice clothing.

Cohesion Cascades Upward

If your functions lack cohesion, your classes will too. If classes are incohesive, packages will be messy. If packages are poorly bounded, services will sprawl. Start at the function level and maintain discipline upward. Cohesion at lower levels makes higher-level cohesion possible.

Common Cohesion Anti-Patterns

Certain structural patterns predictably lead to poor cohesion. Recognizing these anti-patterns helps you avoid them during design and identify them during review.

Cohesion Anti-Patterns to Avoid

•The God Class — One enormous class that knows everything and does everything. It's central to the system, tightly coupled to everything, and impossible to modify safely. Often named Manager, Controller, System, or just App.
•The Utils/Helpers Dumping Ground — A module that exists because 'we didn't know where else to put this.' It grows without bound and becomes a grab-bag of unrelated functions. Typically has the worst cohesion in any codebase.
•The Kitchen Sink Service — A microservice that handles too many unrelated capabilities because splitting it seemed inconvenient. Shows up as a service with many unrelated API endpoints and constant deployment conflicts.
•The Initializer/Bootstrapper — A module that 'sets everything up' by grouping temporally related but semantically unrelated operations. Opens you to initialization order bugs and makes startup opaque.
•The Feature Toggle Module — A central place for all feature flags that every module depends on. Creates coupling to an incohesive module and makes flags hard to reason about.
•The Transaction Script — A procedural module that orchestrates a workflow by embedding all steps inline. Related to temporal cohesion; should be decomposed into focused collaborating components.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
class UserManager {
  // Authentication
  login() { }
  logout() { }
  resetPassword() { }
  
  // Profile
  updateProfile() { }
  uploadAvatar() { }
  
  // Preferences  
  getPreferences() { }
  updateTheme() { }
  
  // Notifications
  sendEmail() { }
  sendPush() { }
  
  // Analytics
  trackEvent() { }
  getMetrics() { }
  
  // ... 50 more methods
}

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
// Each class has one reason to exist
 
class Authenticator {
  login() { }
  logout() { }
  resetPassword() { }
}
 
class ProfileService {
  updateProfile() { }
  uploadAvatar() { }
}
 
class PreferencesStore {
  getPreferences() { }
  updateTheme() { }
}
 
class NotificationSender {
  sendEmail() { }
  sendPush() { }
}
 
class AnalyticsTracker {
  trackEvent() { }
  getMetrics() { }
}

The Gravity of Incohesion

Low-cohesion modules exert gravitational pull. Once a Utils class exists, it attracts more unrelated code. Once a God Class grows, developers add more to it because it seems convenient. Breaking these patterns requires active intervention—they don't self-correct.

Summary: Embracing High Cohesion

We've explored cohesion as a fundamental quality of well-designed software modules. Let's consolidate the key insights:

Key Takeaways

•Cohesion measures how well elements of a module belong together — High cohesion means all elements contribute to a single, well-defined purpose.
•Cohesion exists on a spectrum from coincidental (worst) to functional (best) — Aim for functional or communicational cohesion; avoid coincidental, logical, and temporal cohesion.
•High cohesion enables understandability, maintainability, testability, and reusability — These benefits compound over the life of a system.
•Use the one-sentence test — If you can't describe what a module does in one sentence without 'and' or 'or', it likely lacks cohesion.
•Cohesion applies at every scale — From functions to services, the same principles apply. Discipline at lower levels enables cohesion at higher levels.
•Anti-patterns like God Classes and Utils dumping grounds destroy cohesion — Recognize and actively break these patterns through extraction and decomposition.

What's next:

Cohesion is one half of the equation. The other half is coupling—the degree of interdependence between modules. High cohesion within modules and low coupling between modules form the twin pillars of modular design.

In the next page, we explore coupling: why minimal dependencies matter, how coupling manifests in different forms, and how to achieve loose coupling that enables independent evolution of system components.

Page Complete

You now understand cohesion as a fundamental quality metric for software modules. You can identify different levels of cohesion, recognize anti-patterns that destroy it, and apply techniques to achieve high cohesion. Next, we'll explore its twin concept: coupling.