2nd/10_Wiki/Topics/AI_and_ML/Test-Driven_Development.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-test-driven-development	Test-Driven Development	10_Wiki/Topics	verified	self	TDD Red-Green-Refactor test first	none	A	0.9	applied	tdd testing software-engineering ai-aided		2026-05-10	pending	language framework python pytest

Test-Driven Development

매 한 줄

"매 Red → Green → Refactor 매 fail-first feedback loop". Kent Beck (Extreme Programming, ~2003) origin. 2026 매 AI-pair (Claude Code, Cursor) 의 default mode — agent의 test 먼저 generate 후 implementation 의 iterate.

매 핵심

매 Cycle

1. Red: 매 failing test 의 write (no impl yet).
2. Green: 매 minimum code 의 test 의 pass.
3. Refactor: 매 clean up, 매 test still green.

매 Levels

• Unit: pure function, 1 class.
• Integration: 매 DB / network boundary.
• Acceptance / E2E: 매 user flow.

매 vs BDD

• TDD: developer-facing, 매 method-level.
• BDD: stakeholder-facing, 매 Given-When-Then, Cucumber/Behave.

매 응용

1. Greenfield library — 매 API design 의 test 의 drive.
2. Bug fix — 매 reproducing test 먼저 write.
3. AI agent — 매 spec → test → impl 매 deterministic loop.

💻 패턴

Pytest red-green-refactor

# tests/test_pricing.py
import pytest
from pricing import discount_price

def test_no_discount_for_zero_qty():
    assert discount_price(100, qty=0) == 100

def test_10pct_discount_at_qty_10():
    assert discount_price(100, qty=10) == 90.0

# pricing.py — minimum impl after red
def discount_price(price: float, qty: int) -> float:
    if qty >= 10:
        return price * 0.9
    return price

Parametrize

@pytest.mark.parametrize("qty,expected", [
    (0, 100), (1, 100), (9, 100),
    (10, 90), (50, 90), (100, 80),
])
def test_discount_table(qty, expected):
    assert discount_price(100, qty=qty) == expected

Fixture + integration test

@pytest.fixture
def db():
    conn = sqlite3.connect(":memory:")
    conn.execute("CREATE TABLE orders (id INT, total REAL)")
    yield conn
    conn.close()

def test_save_order(db):
    save_order(db, id=1, total=99.0)
    row = db.execute("SELECT total FROM orders WHERE id=1").fetchone()
    assert row[0] == 99.0

Property-based (Hypothesis)

from hypothesis import given, strategies as st

@given(st.floats(min_value=0, max_value=1e6), st.integers(min_value=0, max_value=1000))
def test_discount_never_exceeds_price(price, qty):
    assert discount_price(price, qty) <= price

TypeScript / Vitest

import { test, expect } from 'vitest';
import { discountPrice } from './pricing';

test('10% discount at qty 10', () => {
  expect(discountPrice(100, 10)).toBeCloseTo(90);
});

AI-aided TDD loop (2026)

# .claude/commands/tdd.md
# 1. Write failing test for: $ARGUMENTS
# 2. Run pytest (expect red)
# 3. Implement minimum code to pass
# 4. Run pytest (expect green)
# 5. Refactor; ensure still green
# 6. Open PR

# Claude Code automated loop pseudocode
while not all_green:
    test = agent.generate_test(spec)
    write_file(test)
    run_pytest()  # expect FAIL
    impl = agent.generate_impl_to_pass(test)
    write_file(impl)
    if run_pytest().passed:
        agent.refactor()

매 결정 기준

상황	Approach
매 pure function	TDD strict
매 UI tweak	Snapshot + visual diff (TDD overkill)
매 spike / prototype	Skip TDD, 매 throw away
매 bug fix	매 reproducing test 먼저
매 AI agent task	Test-first prompt — 매 verifiable

기본값: 매 TDD for business logic + libraries. 매 visual / experimental code 매 skip.

🔗 Graph

• 변형: BDD · Property-Based Testing · Mutation Testing
• 응용: Refactoring_Best_Practices
• Adjacent: 121_pytest_기본기 · Hypothesis

🤖 LLM 활용

언제: 매 test generation (Claude Code "write tests for X"). 매 stub from spec. 매 reproducing-test from bug report. 언제 X: 매 LLM의 hallucinate API → 매 test의 wrong. 매 always run + verify red→green.

❌ 안티패턴

• Test after: 매 not TDD; 매 confirms current code, 매 design 의 X drive.
• Mock everything: 매 test의 verify nothing real.
• Brittle tests: 매 implementation detail 의 lock — refactor breaks.
• Skipping refactor step: 매 code rot.
• AI-generated test without running: 매 false-green.

🧪 검증 / 중복

• Verified (Kent Beck "TDD by Example" 2003; Anthropic Claude Code TDD docs 2026).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — TDD red-green-refactor + property-based + AI agent loop