2nd/10_Wiki/Topics/DevOps_and_Security/응집도 (Cohesion).md

---
id: wiki-2026-0508-응집도-cohesion
title: 응집도 (Cohesion)
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Cohesion, 응집도, Module Cohesion]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [software-design, modularity, srp]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: any
  framework: any
---

# 응집도 (Cohesion)

## 매 한 줄
> **"매 한 module, 매 한 reason to change"**. 매 module 내부 element 가 매 single purpose 의 향해 의 얼마나 의 tightly 의 related 의 measure. 매 Larry Constantine (1974) 의 7-level taxonomy. 매 SOLID 의 SRP 의 conceptual ancestor.

## 매 핵심

### 매 7 levels of cohesion (low → high)
| Level | 매 의미 | 예 |
|---|---|---|
| Coincidental | 매 unrelated | `Util.java` kitchen sink |
| Logical | 매 same category | `IOHandler` (file + net + console) |
| Temporal | 매 same time | `init()` 의 mixed setup |
| Procedural | 매 same flow | step1 → step2 → step3 |
| Communicational | 매 same data | report 의 multiple formatting |
| Sequential | 매 output → input | parse → validate → transform |
| Functional | 매 single task | `calculateTax(invoice)` |

### 매 Functional cohesion (target)
- 매 module 의 모든 element 의 한 well-defined task 의 contribute.
- 매 reusable, testable, maintainable.
- 매 SRP 의 satisfied.

### 매 측정 (informal)
- 매 module 의 description 의 "and" / "or" 의 포함? → 매 likely low cohesion.
- 매 reason to change 의 한 가지? → 매 high cohesion.
- 매 LCOM (Lack of Cohesion of Methods) — Chidamber-Kemerer metric.

## 💻 패턴

### 매 Functional cohesion (best)
```python
class TaxCalculator:
    def calculate(self, amount: Decimal, jurisdiction: str) -> Decimal:
        rate = self._lookup_rate(jurisdiction)
        return amount * rate
```

### 매 Coincidental → split
```python
# Before — 매 coincidental
class Utils:
    def format_phone(p): ...
    def calculate_tax(a): ...
    def send_email(e): ...

# After — 매 functional
class PhoneFormatter: ...
class TaxCalculator: ...
class EmailSender: ...
```

### 매 Temporal → procedural / functional
```python
# Before — 매 temporal "everything at startup"
def init():
    connect_db()
    load_config()
    warm_cache()
    start_metrics()

# After — 매 explicit phases
class Bootstrap:
    def configure(self): load_config()
    def connect(self): connect_db()
    def warm(self): warm_cache()
    def observe(self): start_metrics()
```

### 매 Logical → polymorphism
```python
# Before — 매 logical "all I/O"
class IOHandler:
    def read_file(self, p): ...
    def read_socket(self, s): ...
    def read_stdin(self): ...

# After — 매 functional via interface
class Reader(Protocol):
    def read(self) -> bytes: ...

class FileReader(Reader): ...
class SocketReader(Reader): ...
class StdinReader(Reader): ...
```

### 매 LCOM 측정 (Java)
```java
// 매 ckjm tool / SonarQube 의 LCOM4 metric
// LCOM4 = 1 → 매 perfectly cohesive
// LCOM4 > 1 → 매 split candidates
```

### 매 Communicational cohesion 예
```python
class InvoiceReport:
    def __init__(self, invoice: Invoice):
        self.invoice = invoice  # 매 shared data

    def to_pdf(self): ...
    def to_csv(self): ...
    def to_json(self): ...
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| 매 class 의 method 의 의 unrelated | 매 Split into multiple classes |
| 매 method 의 의 multiple responsibilities | 매 Extract method, then class |
| 매 utility class 의 grow | 매 Domain-specific helper 의 의 split |
| 매 god class 의 LCOM 의 high | 매 Refactor by responsibility |

**기본값**: 매 functional cohesion 의 추구. 매 SRP — 매 한 class, 매 한 reason to change.

## 🔗 Graph
- 부모: [[응집도와 결합도 (Cohesion and Coupling)]]
- 응용: [[SOLID]] · [[Single Responsibility Principle (SRP)|Single Responsibility Principle]]
- Adjacent: [[결합도]]

## 🤖 LLM 활용
**언제**: 매 class / module 의 design review. 매 god class refactoring 의 priority.
**언제 X**: 매 trivial DTO / value object — 매 cohesion 의 의 trivially high.

## ❌ 안티패턴
- **God class**: 매 매 모든 의 처리 → 매 lowest cohesion.
- **Utility dumping ground**: 매 `Helpers`, `Common`, `Misc` 의 class.
- **Manager class**: 매 `XxxManager` 의 vague responsibility.
- **Feature envy**: 매 method 의 다른 class 의 data 의 access — 매 wrong class 의 의 belong.

## 🧪 검증 / 중복
- Verified — Constantine & Yourdon, *Structured Design* (1979); Robert C. Martin, *Clean Architecture*.
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — 7 cohesion levels + refactor recipes |