simplicity-principles（简单原则）

Simplicity Principles

Write code that is simple, necessary, and unsurprising.

Three Core Principles

1. KISS - Keep It Simple, Stupid

Simple solutions are better than clever ones

What Simple Means

• Readable by developers of varying skill levels
• Fewer moving parts and abstractions
• Direct and obvious implementation
• Easy to debug and test
• Minimal cognitive load

Elixir Examples

# COMPLEX - Over-engineered
defmodule PaymentCalculator do
  use GenServer

  def start_link(_), do: GenServer.start_link(__MODULE__, %{}, name: __MODULE__)
  def calculate(items), do: GenServer.call(__MODULE__, {:calculate, items})

  def handle_call({:calculate, items}, _from, state) do
    result = Enum.reduce(items, Money.new(:USD, 0), &Money.add(&2, &1.price))
    {:reply, result, state}
  end
end

# SIMPLE - Just a function
defmodule PaymentCalculator do
  def calculate(items) do
    Enum.reduce(items, Money.new(:USD, 0), &Money.add(&2, &1.price))
  end
end
# Use GenServer only when you need state/concurrency

# COMPLEX - Unnecessary abstraction
defmodule UserQuery do
  defmacro by_status(status) do
    quote do
      from u in User, where: u.status == ^unquote(status)
    end
  end
end

# SIMPLE - Direct query
def active_users do
  from u in User, where: u.status == "active"
end

def inactive_users do
  from u in User, where: u.status == "inactive"
end
# Macros only when you need metaprogramming

TypeScript Examples

// COMPLEX - Over-abstraction
class UserDataManager {
  private dataSource: DataSource;
  private cache: Cache;
  private transformer: DataTransformer;

  async getUser(id: string): Promise<User> {
    const cached = await this.cache.get(id);
    if (cached) return this.transformer.transform(cached);

    const raw = await this.dataSource.fetch(id);
    await this.cache.set(id, raw);
    return this.transformer.transform(raw);
  }
}

// SIMPLE - Direct approach
async function getUser(id: string): Promise<User> {
  const response = await fetch(`/api/users/${id}`);
  return response.json();
}
// Add cache/transform only when performance demands it

// COMPLEX - Clever but confusing
const isValid = (x: number) => !!(x >= 0 && x <= 100);
const process = (items: number[]) =>
  items.filter(isValid).reduce((a, b) => a + b, 0);

// SIMPLE - Clear intent
function calculateTotal(scores: number[]): number {
  const validScores = scores.filter(score => score >= 0 && score <= 100);
  return validScores.reduce((sum, score) => sum + score, 0);
}

KISS Guidelines

• Prefer functions over classes (unless you need state)
• Prefer explicit over implicit
• Prefer boring over clever
• Prefer standard library over custom solutions
• Prefer clear names over short names
• Prefer straightforward logic over "elegant" one-liners

When NOT to KISS

• Performance-critical code (after profiling proves need)
• Preventing code duplication (after third instance)
• Enforcing constraints (types, validations)

2. YAGNI - You Aren't Gonna Need It

Don't implement features until you actually need them

Signs You're Violating YAGNI

• "We might need this someday"
• "Let me add flexibility for future use cases"
• "I'll build a generic solution"
• "This could be configurable"

Elixir Examples (YAGNI)

# YAGNI VIOLATION - Premature abstraction
defmodule NotificationService do
  def send(notification, opts \\ []) do
    provider = opts[:provider] || :default
    priority = opts[:priority] || :normal
    retry_strategy = opts[:retry_strategy] || :exponential
    callback = opts[:callback]

    # Complex routing logic for features we don't use yet
  end
end

# GOOD - Build what you need now
defmodule NotificationService do
  def send_email(to, subject, body) do
    Email.send(to, subject, body)
  end

  # Add SMS when we actually need it
  # Add priorities when we have the requirement
  # Add retries when we see failures
end

# YAGNI VIOLATION - Unused flexibility
defmodule User do
  schema "users" do
    field :email, :string
    field :settings, :map  # "For future configuration"
    field :metadata, :map  # "For anything we might need"
    field :flags, {:array, :string}  # "For feature flags"

    # None of these are used yet!
  end
end

# GOOD - Add fields when needed
defmodule User do
  schema "users" do
    field :email, :string
    # Add fields when requirements emerge
  end
end

TypeScript Examples (YAGNI)

// YAGNI VIOLATION - Premature generalization
interface DataFetcher<T, P extends object = object> {
  fetch(params: P): Promise<T>;
  cache?(params: P): Promise<void>;
  invalidate?(key: string): Promise<void>;
  prefetch?(params: P[]): Promise<void>;
  // We don't use cache, invalidate, or prefetch yet!
}

// GOOD - Start simple
interface DataFetcher<T> {
  fetch(params: object): Promise<T>;
  // Add methods when we need caching
}

// YAGNI VIOLATION - Configurable everything
interface ButtonProps {
  onClick: () => void;
  variant?: 'primary' | 'secondary' | 'tertiary' | 'ghost' | 'outline';
  size?: 'xs' | 'sm' | 'md' | 'lg' | 'xl';
  shape?: 'rounded' | 'square' | 'pill';
  shadow?: 'none' | 'sm' | 'md' | 'lg';
  animation?: 'fade' | 'slide' | 'bounce';
  // Design only uses 2 variants and 1 size!
}

// GOOD - Implement what designs require
interface ButtonProps {
  onClick: () => void;
  variant?: 'primary' | 'secondary';  // Only what we use
  // Add options when design requires them
}

YAGNI Guidelines

• Implement features when you have a concrete use case, not a hypothetical one
• Delete unused code immediately (it's in git)
• Start with hardcoded values, extract constants when they vary
• Build for today's requirements, refactor for tomorrow's
• Question every "nice to have" and "might need"

Exceptions to YAGNI

• Security features (implement defense in depth upfront)
• Data migrations (plan schema carefully)
• Public APIs (harder to change later)
• Accessibility (build in from start)

3. Principle of Least Astonishment (POLA)

Code should behave the way users expect it to behave

What Makes Code Astonishing

• Unexpected side effects
• Inconsistent naming
• Breaking conventions
• Hidden behavior
• Surprising return values

Elixir Examples (POLA)

# ASTONISHING - Updates AND returns different thing
def update_user(user, attrs) do
  Repo.update!(User.changeset(user, attrs))
  UserCache.invalidate(user.id)  # Side effect!
  :ok  # Returns :ok instead of updated user!?
end

# EXPECTED - Clear behavior
def update_user(user, attrs) do
  Repo.update(User.changeset(user, attrs))
  # Caller handles cache invalidation explicitly
end

# ASTONISHING - Function name lies
def get_user(id) do
  case Repo.get(User, id) do
    nil ->
      attrs = %{id: id, email: "#{id}@example.com"}
      Repo.insert!(User.changeset(attrs))  # Created user in a getter!
    user -> user
  end
end

# EXPECTED - Name matches behavior
def get_or_create_user(id) do
  case Repo.get(User, id) do
    nil -> create_default_user(id)
    user -> user
  end
end

TypeScript Examples (POLA)

// ASTONISHING - Function mutates input
function processTask(gig: Task): Task {
  gig.status = 'processed';  // Mutates input!
  gig.processedAt = new Date();
  return gig;
}

// EXPECTED - Pure function
function processTask(gig: Task): Task {
  return {
    ...gig,
    status: 'processed',
    processedAt: new Date(),
  };
}

// ASTONISHING - Inconsistent return types
async function getUser(id: string): Promise<User | null | undefined> {
  // Returns null sometimes, undefined other times, no pattern
}

// EXPECTED - Consistent return
async function getUser(id: string): Promise<User | null> {
  // Always null when not found
}

// ASTONISHING - Breaking conventions
interface Props {
  onPress?: () => void;      // React convention: onX
  clickHandler?: () => void;  // Different convention in same interface!
  onTapGesture?: () => void;  // Yet another name for same thing!
}

// EXPECTED - Consistent conventions
interface Props {
  onPress?: () => void;
  onLongPress?: () => void;
  onDoublePress?: () => void;
}

POLA Guidelines

• Follow framework conventions (Phoenix, React, Relay)
• Use clear, descriptive names that match behavior
• Return what the function name promises
• Keep side effects explicit or avoid them
• Be consistent within the codebase
• Match platform conventions (iOS, Android, Web)
• Honor principle of least surprise in APIs

Examples of Good POLA in YourApp

• Command handlers return {:ok, result} or {:error, reason} (consistent)
• React components with onPress not onClick (platform convention)
• Ecto changesets don't touch database (pure validation)
• GraphQL mutations clearly named: createTask, updateTask, deleteTask

Application Checklist

Before implementing

• Is this the simplest solution? (KISS)
• Do we actually need this now? (YAGNI)
• Will this behavior surprise users? (POLA)

During implementation

• Prefer straightforward over clever
• Implement only what's required
• Follow established conventions
• Name things accurately
• Make side effects explicit

During code review

• Is there a simpler approach?
• Are we building speculative features?
• Does the API behave as expected?
• Are conventions followed?

Red Flags

KISS Violations

• "Let me show you this clever trick..."
• More than 3 levels of abstraction
• Requires 10-minute explanation
• Uses advanced language features unnecessarily

YAGNI Violations

• "We might need this later..."
• Unused parameters/options
• Configurable everything
• "Generic framework" for 2 use cases

POLA Violations

• "Well, technically it does..."
• Inconsistent naming
• Hidden side effects
• Surprising error conditions

Integration with Existing Skills

Works with

• solid-principles: Simple implementations of SOLID patterns
• boy-scout-rule: Simplify when improving code
• test-driven-development: Simple code is easier to test
• elixir-code-quality-enforcer: Credo flags complexity
• typescript-code-quality-enforcer: Linting enforces consistency

Remember

"Simplicity is the ultimate sophistication." - Leonardo da Vinci

• Prefer boring, proven solutions over novel approaches
• Build incrementally based on actual requirements
• Follow conventions so code behaves as expected
• Delete speculative code immediately
• Simple!= simplistic (handle errors, edge cases properly)

When in doubt, choose the simpler path