Design Principles

Design Principles Reference

SOLID Principles

Single Responsibility (SRP)

A module/class should have one reason to change, meaning one actor or stakeholder it serves.

Test: "If I describe this module's job, do I use the word 'and'?" If yes, consider splitting.

Violation pattern:

UserService → handles registration AND sends emails AND generates reports

Fix: Split into RegistrationService, EmailNotifier, UserReportGenerator.

Nuance: SRP is about change-axes, not about doing only one thing. A JsonParser does many steps but changes for one reason (JSON format changes).

Open-Closed (OCP)

Modules should be extendable without modifying existing code.

Primary mechanism: Define behavior behind interfaces. New behavior = new implementation, not editing existing code.

# Instead of:
if type == "pdf": ...
elif type == "csv": ...
elif type == "xlsx": ...    # Must edit to add new format

# Define:
type Exporter interface { Export(data) }
# Then: PdfExporter, CsvExporter, XlsxExporter each implement it
# New format = new struct, no edits to existing code

When to apply: When you can anticipate a clear extension axis. Don't pre-engineer OCP for hypothetical changes.

Liskov Substitution (LSP)

Subtypes must honor the behavioral contract of their parent type. Any code using a base type must work correctly with any subtype.

Classic violation: Square extending Rectangle — setting width on a Square also changes height, breaking Rectangle's contract.

Practical test: If subtype overrides a method and changes behavior in ways callers don't expect, LSP is violated.

Interface Segregation (ISP)

Prefer focused interfaces over broad ones. Clients should not depend on methods they don't use.

Violation:

type DataStore interface {
    Read(id) → Item
    Write(item)
    Delete(id)
    BulkImport(items)
    RunMigration()
    GenerateReport()
}

Fix: Split by consumer need:

type Reader interface { Read(id) → Item }
type Writer interface { Write(item); Delete(id) }
type Admin  interface { BulkImport(items); RunMigration() }

Go idiom: Accept interfaces, return structs. Define interfaces at the call site (consumer), not the implementation site.

Dependency Inversion (DIP)

High-level policy should not depend on low-level detail. Both should depend on abstractions.

In practice: Use cases define the interfaces they need (ports). Infrastructure implements them (adapters). The wiring happens at the composition root.

# usecases/ports.go — owned by use case layer
type OrderRepository interface { Save(order Order) error }

# adapters/postgres/order_repo.go — implements the interface
type PostgresOrderRepo struct { db *sql.DB }
func (r *PostgresOrderRepo) Save(order Order) error { ... }

# main.go — wires it
repo := postgres.NewOrderRepo(db)
useCase := usecases.NewCreateOrder(repo)

Dependency Management

Dependency Injection Approaches

Composition Root Pattern

• ONE place in the application creates all concrete types and wires dependencies
• Usually main() or an app bootstrap function
• No other code should use new ConcreteType() for injectable dependencies
• This is where you choose Postgres vs. SQLite, real email vs. mock, etc.

Dependency Direction Rules

• Dependencies flow from outer (infra) to inner (domain), never the reverse
• If a domain type needs to call infra, define an interface in the domain layer
• Circular dependencies between modules indicate a design problem. Fix options:
  1. Extract shared types into a new module both depend on
  2. Use events/callbacks instead of direct calls
  3. Merge the modules (they may be one concern)

Module Coupling and Cohesion

Coupling Spectrum (best → worst)

1. No coupling — modules are completely independent
2. Data coupling — modules share only simple data (primitives, DTOs)
3. Stamp coupling — modules share composite data structures
4. Control coupling — one module passes a flag that controls another's behavior
5. Content coupling — one module reaches into another's internals

Goal: Stay at data coupling or stamp coupling. If you see control or content coupling, refactor.

Cohesion Spectrum (best → worst)

1. Functional — all elements contribute to a single well-defined task
2. Sequential — output of one element feeds into the next
3. Communicational — elements operate on the same data
4. Procedural — elements follow an execution order but aren't related
5. Logical — elements do similar things but are unrelated (e.g., a "utils" grab-bag)
6. Coincidental — elements are grouped arbitrarily

Goal: Aim for functional or sequential cohesion. If a module has logical or coincidental cohesion, split it.

Detecting Bad Boundaries

• Shotgun surgery: One feature change requires edits in many modules
• Divergent change: One module changes for many different reasons
• Feature envy: A class uses more methods from another class than its own
• Shared mutable state: Multiple modules read/write the same data structure

Interface Design

API Surface Principles

• Minimal surface: Export only what external consumers need
• Semantic naming: Interface names describe capability, not implementation (Storer not DatabaseClient)
• Error contracts: Define what errors an interface can return and what they mean
• Idempotency: State whether operations are idempotent in the contract

Go-Specific Interface Guidance

// Define interfaces where they're USED, not where they're implemented
// In the consumer package:
type UserFinder interface {
    FindByID(ctx context.Context, id string) (*User, error)
}

// Accept the narrowest interface possible
func NewOrderService(users UserFinder) *OrderService { ... }

General Interface Guidelines

• Prefer composition of small interfaces over inheritance of large ones
• An interface with >5 methods is a smell — consider splitting
• If every implementation has the same method body, it should be a default/base implementation, not an interface method
• Interfaces should represent capabilities, not objects (Readable, Closable not File)