software-design-principles（软件设计原则）

Software Design Principles

Professional software design patterns and principles for writing maintainable, well-structured code.

Critical Rules

🚨 Fail-fast over silent fallbacks. Never use fallback chains (value?? backup?? 'unknown'). If data should exist, validate and throw a clear error.

🚨 Strive for maximum type-safety. No any. No as. Type escape hatches defeat TypeScript's purpose. There's always a type-safe solution.

🚨 Make illegal states unrepresentable. Use discriminated unions, not optional fields. If a state combination shouldn't exist, make the type system forbid it.

🚨 Inject dependencies, don't instantiate. No new SomeService() inside methods. Pass dependencies through constructors.

🚨 Intention-revealing names only. Never use data, utils, helpers, handler, processor. Name things for what they do in the domain.

🚨 No code comments. Comments are a failure to express intent in code. If you need a comment to explain what code does, the code isn't clear enough—refactor it.

🚨 Use Zod for runtime validation. In TypeScript, use Zod schemas for parsing external data, API responses, and user input. Type inference from schemas keeps types and validation in sync.

When This Applies

• Writing new code (these are defaults, not just refactoring goals)
• Refactoring existing code
• Code reviews and design reviews
• During TDD REFACTOR phase
• When analyzing coupling and cohesion

Core Philosophy

Well-designed, maintainable code is far more important than getting things done quickly. Every design decision should favor:

• Clarity over cleverness
• Explicit over implicit
• Fail-fast over silent fallbacks
• Loose coupling over tight integration
• Intention-revealing over generic

Code Without Comments

Never write comments - write expressive code instead.

Object Calisthenics

Apply object calisthenics principles:

The Nine Rules

1. One level of indentation per method

- In practice, I will tolerate upto 3

1. Don't use the ELSE keyword

- Use early returns instead

1. Wrap all primitives and strings

- Create value objects - Encapsulate validation logic - Make domain concepts explicit

1. First class collections

- Classes with collections should contain nothing else

1. One dot per line
2. Don't abbreviate

- Use full, descriptive names

1. Keep all entities small

- Small classes (< 150 lines) - Small methods (< 10 lines) - Small packages/modules - Easier to understand and maintain

1. Avoid getters/setters/properties on entities

- Tell, don't ask - Objects should do work, not expose data

When to Apply

• During refactoring:
• During code review:

Feature Envy Detection

Method uses another class's data more than its own? Move it there.

// ❌ FEATURE ENVY - obsessed with Order's data
class InvoiceGenerator {
  generate(order: Order): Invoice {
    const total = order.getItems().map(i => i.getPrice() * i.getQuantity()).reduce((a,b) => a+b, 0)
    return new Invoice(total + total * order.getTaxRate() + order.calculateShipping())
  }
}

// ✅ Move logic to the class it envies
class Order {
  calculateTotal(): number { /* uses this.items, this.taxRate */ }
}
class InvoiceGenerator {
  generate(order: Order): Invoice { return new Invoice(order.calculateTotal()) }
}

Detection: Count external vs own references. More external? Feature envy.

Dependency Inversion Principle

Don't instantiate dependencies inside methods. Inject them.

// ❌ TIGHT COUPLING
class OrderProcessor {
  process(order: Order): void {
    const validator = new OrderValidator()  // Hard to test/change
    const emailer = new EmailService()      // Hidden dependency
  }
}

// ✅ LOOSE COUPLING
class OrderProcessor {
  constructor(private validator: OrderValidator, private emailer: EmailService) {}
  process(order: Order): void {
    this.validator.isValid(order)  // Injected, mockable
    this.emailer.send(...)         // Explicit dependency
  }
}

Scan for: new X() inside methods, static method calls. Extract to constructor.

Fail-Fast Error Handling

NEVER use fallback chains:

value ?? backup ?? default ?? 'unknown'  // ❌

Validate and throw clear errors instead:

// ❌ SILENT FAILURE - hides problems
return content.eventType ?? content.className ?? 'Unknown'

// ✅ FAIL FAST - immediate, debuggable
if (!content.eventType) {
  throw new Error(`Expected 'eventType', got undefined. Keys: [${Object.keys(content)}]`)
}
return content.eventType

Error format: Expected [X]. Got [Y]. Context: [debugging info]

Naming Conventions

Principle: Use business domain terminology and intention-revealing names. Never use generic programmer jargon.

Forbidden Generic Names

NEVER use these names:

• data
• utils
• helpers
• common
• shared
• manager
• handler
• processor

These names are meaningless - they tell you nothing about what the code actually does.

Intention-Revealing Names

Instead of generic names, use specific domain language:

// ❌ GENERIC - meaningless
class DataProcessor {
  processData(data: any): any {
    const utils = new DataUtils()
    return utils.transform(data)
  }
}

// ✓ INTENTION-REVEALING - clear purpose
class OrderTotalCalculator {
  calculateTotal(order: Order): Money {
    return taxCalculator.applyTax(order.subtotal, order.taxRate)
  }
}

Naming Checklist

For classes:

• Does the name reveal what the class is responsible for?
• Is it a noun (or noun phrase) from the domain?
• Would a domain expert recognize this term?

For methods:

• Does the name reveal what the method does?
• Is it a verb (or verb phrase)?
• Does it describe the business operation?

For variables:

• Does the name reveal what the variable contains?
• Is it specific to this context?
• Could someone understand it without reading the code?

Refactoring Generic Names

When you encounter generic names:

1. Understand the purpose: What is this really doing?
2. Ask domain experts: What would they call this?
3. Extract domain concept: Is there a domain term for this?
4. Rename comprehensively: Update all references

Type-Driven Design

Principle: Follow Scott Wlaschin's type-driven approach to domain modeling. Express domain concepts using the type system.

Make Illegal States Unrepresentable

Use types to encode business rules:

// ❌ PRIMITIVE OBSESSION - illegal states possible
interface Order {
  status: string  // Could be any string
  shippedDate: Date | null  // Could be set when status != 'shipped'
}

// ✓ TYPE-SAFE - illegal states impossible
type UnconfirmedOrder = { type: 'unconfirmed', items: Item[] }
type ConfirmedOrder = { type: 'confirmed', items: Item[], confirmationNumber: string }
type ShippedOrder = { type: 'shipped', items: Item[], confirmationNumber: string, shippedDate: Date }

type Order = UnconfirmedOrder | ConfirmedOrder | ShippedOrder

Avoid Type Escape Hatches

STRICTLY FORBIDDEN without explicit user approval:

• any type
• as type assertions (as unknown as, as any, as SomeType)
• @ts-ignore / @ts-expect-error

There is always a better type-safe solution. These make code unsafe and defeat TypeScript's purpose.

Use the Type System for Validation

// ✓ TYPE-SAFE - validates at compile time
type PositiveNumber = number & { __brand: 'positive' }

function createPositive(value: number): PositiveNumber {
  if (value <= 0) {
    throw new Error(`Expected positive number, got ${value}`)
  }
  return value as PositiveNumber
}

// Can only be called with validated positive numbers
function calculateDiscount(price: PositiveNumber, rate: number): Money {
  // price is guaranteed positive by type system
}

Prefer Immutability

Principle: Default to immutable data. Mutation is a source of bugs—unexpected changes, race conditions, and difficult debugging.

The Problem: Mutable State

// MUTABLE - hard to reason about
function processOrder(order: Order): void {
  order.status = 'processing'  // Mutates input!
  order.items.push(freeGift)   // Side effect!
}

// Caller has no idea their object changed
const myOrder = getOrder()
processOrder(myOrder)
// myOrder is now different - surprise!

The Solution: Return New Values

// IMMUTABLE - predictable
function processOrder(order: Order): Order {
  return {
    ...order,
    status: 'processing',
    items: [...order.items, freeGift]
  }
}

// Caller controls what happens
const myOrder = getOrder()
const processedOrder = processOrder(myOrder)
// myOrder unchanged, processedOrder is new

Application Rules

• Prefer const over let
• Prefer spread (...) over mutation
• Prefer map/filter/reduce over forEach with mutation
• If you must mutate, make it explicit and contained

YAGNI - You Aren't Gonna Need It

Principle: Don't build features until they're actually needed. Speculative code is waste—it costs time to write, time to maintain, and is often wrong when requirements become clear.

The Problem: Speculative Generalization

// YAGNI VIOLATION - over-engineered for "future" needs
interface PaymentProcessor {
  process(payment: Payment): Result
  refund(payment: Payment): Result
  partialRefund(payment: Payment, amount: Money): Result
  schedulePayment(payment: Payment, date: Date): Result
  recurringPayment(payment: Payment, schedule: Schedule): Result
  // ... 10 more methods "we might need"
}

// Only ONE method is actually used today

Application Rules

• Build the simplest thing that works
• Add capabilities when requirements demand them, not before
• "But we might need it" is not a requirement

When Tempted to Cut Corners

STOP if you're about to:

• Use ?? chains → fail fast with clear error instead
• Use any or as → fix the types, not the symptoms
• Use new X() inside a method → inject through constructor
• Name something data, utils, handler → use domain language
• Add a getter → ask if the object should do the work instead
• Skip refactor because "it works" → refactor IS part of the work
• Write a comment → make the code self-explanatory
• Mutate a parameter → return a new value
• Build "for later" → build what you need now