o2o-site-AEO/solution/backend/tests/test_copy_api.py
Mina Choi 11d30bb3d1 [chore] solution,admin,ontology: 코드 주석을 한 줄로 — 히스토리 주석 삭제
여러 줄 주석이 설명보다 경위(예전·실측·지적)를 적고 있어 읽는 사람이 결론을 찾기 어려웠다.

- ts·tsx·js·mjs·css·py 478개: 여러 줄 주석은 첫 문장 한 줄로, 과거형·날짜 문장은 삭제
- 주석 위치는 TypeScript 파서·파이썬 tokenize/ast 로 찾는다 — 문자열 안의 # · /* 는 건드리지 않는다
- eslint·ts·noqa·type: ignore 같은 지시 주석은 그대로 둔다

파이썬 275개 정리 전후 AST 동일, TS 298개 주석 뺀 토큰 동일(빈 JSX 주석 10곳만 차이).
site·frontend·admin tsc, site vitest 105 passed

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-28 16:05:19 +09:00

306 lines
16 KiB
Python

"""소개문·FAQ 생성 — ★ LLM 은 사실을 만들지 않는다."""
import uuid
from sqlalchemy import text
from common.enums import ErrorType, FactStatus, JobStatus, SourceType
from crud.job_crud import JobQueue
from services.external import gemini_text
from worker.handlers import build_handler
from worker.runner import Worker
def _copy(intro="조용한 숙소입니다.", faqs=None, rejected=None):
return gemini_text.GeneratedCopy(
intro=intro,
intro_fact_keys=["check_in_time"],
meta_description="양양 하조대 펜션",
faqs=faqs if faqs is not None else [
gemini_text.GeneratedFaq("체크인은 몇 시인가요?", "15시입니다.", ["check_in_time"]),
],
rejected=rejected or [],
)
def _patch(monkeypatch, copy):
monkeypatch.setattr(gemini_text, "is_configured", lambda: True)
async def _gen(place_name, category, facts, **kw):
return copy
monkeypatch.setattr(gemini_text, "generate_copy", _gen)
async def _place_with_facts(client, h, n_verified=5):
pid = (await client.post("/v1/place", headers=h, json={"name": "카피펜션", "category": 1})).json()["place"]["place_id"]
await client.post(f"/v1/place/{pid}/verify", headers=h, json={"source": 2, "road_address": f"주소{uuid.uuid4().hex[:6]}"})
keys = ["check_in_time", "check_out_time", "cancel_policy", "cooking_allowed", "pet_allowed"][:n_verified]
for k in keys:
await client.post(f"/v1/place/{pid}/fact", headers=h, json={"key": k, "value": "15:00"})
return pid
async def _faq_rows(db_engine, pid):
async with db_engine.begin() as c:
return (await c.execute(
text("SELECT question, answer, source_fact_ids, status, generated_by FROM place_faqs "
"WHERE place_id = :p ORDER BY sort_order"),
{"p": uuid.UUID(pid)},
)).all()
async def test_copy_generates_intro_and_faq_as_published(auth_headers, client, db_engine, monkeypatch):
"""검증: 소개문·FAQ 를 생성한다."""
_patch(monkeypatch, _copy())
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
body = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()
assert body["result"]["success"] is True
assert body["grounded_facts"] == 5
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
job = (await client.get(f"/v1/job/{body['job_id']}", headers=h)).json()["job"]
assert job["status"] == JobStatus.DONE.value, job.get("last_error")
assert job["result"]["intro"] is True
assert job["result"]["faqs"] == 1
rows = await _faq_rows(db_engine, pid)
assert rows[0][3] == FactStatus.VERIFIED.value, "★ 생성된 FAQ 는 바로 노출값이어야 한다"
assert rows[0][2] == ["check_in_time"], "근거 fact 가 기록돼야 한다"
# 소개문도 fact 로 들어가고, 승인 없이 바로 노출값이다
facts = (await client.get(f"/v1/place/{pid}/fact/list", headers=h)).json()
intro = [f for f in facts["facts"] if f["key"] == "intro"]
assert intro and intro[0]["status"] == FactStatus.VERIFIED.value
assert intro[0]["source_type"] == SourceType.LLM.value
async def _must_not_call_llm(*args, **kwargs):
raise AssertionError("★ 근거 없이 LLM 을 불렀다")
async def test_resume_reuses_completed_job(auth_headers, client, db_engine, monkeypatch):
_patch(monkeypatch, _copy())
h = await auth_headers("resume-owner")
pid = await _place_with_facts(client, h)
path = f"/v1/place/{pid}/copy"
first = (await client.post(path, headers=h, json={})).json()
pending = (await client.post(path, headers=h, json={"resume": True})).json()
assert pending["job_id"] == first["job_id"]
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
resumed = (await client.post(path, headers=h, json={"resume": True})).json()
assert resumed["job_id"] == first["job_id"]
assert resumed["status"] == JobStatus.DONE.value
assert resumed["created"] is False
job = (await client.get(f"/v1/job/{first['job_id']}", headers=h)).json()["job"]
assert job["place_id"] == pid
assert [(s["id"], s["status"]) for s in job["progress"]["steps"]] == [
("prepare", "done"), ("generate", "done"), ("save", "done"), ("faq_fill", "done"),
]
fresh = (await client.post(path, headers=h, json={})).json()
assert fresh["job_id"] != first["job_id"]
async def test_copy_progress_is_visible_during_generation(auth_headers, client, db_engine, monkeypatch):
h = await auth_headers("progress-owner")
pid = await _place_with_facts(client, h)
_patch(monkeypatch, _copy())
started = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()
async def generate(*args, **kwargs):
row = await JobQueue().get(started["job_id"])
assert row["status"] == JobStatus.RUNNING.value
assert [step["status"] for step in row["progress"]["steps"]] == ["done", "running", "pending", "pending"]
return _copy()
monkeypatch.setattr(gemini_text, "generate_copy", generate)
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
row = await JobQueue().get(started["job_id"])
assert row["status"] == JobStatus.DONE.value, row["last_error"]
async def test_copy_job_cannot_be_read_by_another_owner(auth_headers, client, monkeypatch):
_patch(monkeypatch, _copy())
owner = await auth_headers("owner")
other = await auth_headers("other")
pid = await _place_with_facts(client, owner)
started = (await client.post(f"/v1/place/{pid}/copy", headers=owner, json={})).json()
response = (await client.get(f"/v1/job/{started['job_id']}", headers=other)).json()
assert response["result"]["success"] is False
assert not response.get("job")
async def test_copy_refuses_without_facts_when_no_catalog(auth_headers, client, monkeypatch):
"""검증: 확인된 fact 가 하나도 없는 **호텔**(펜션 카탈로그 제외 대상)에서 생성을 시도한다."""
monkeypatch.setattr(gemini_text, "is_configured", lambda: True)
h = await auth_headers("u1")
pid = (await client.post("/v1/place", headers=h, json={"name": "빈호텔", "category": 1})).json()["place"]["place_id"]
await client.post(f"/v1/place/{pid}/verify", headers=h,
json={"source": 2, "road_address": "빈호텔주소", "category_name": "호텔"})
r = await client.post(f"/v1/place/{pid}/copy", headers=h, json={})
assert r.json()["result"]["code"] == ErrorType.FAQ_UNGROUNDED.value
async def test_unverified_facts_are_not_used_as_grounding(auth_headers, client, db_engine, monkeypatch):
"""검증: 크롤링으로 들어온 미검증 fact 만 있는 펜션."""
monkeypatch.setattr(gemini_text, "is_configured", lambda: True)
monkeypatch.setattr(gemini_text, "generate_copy", _must_not_call_llm)
h = await auth_headers("u1")
pid = (await client.post("/v1/place", headers=h, json={"name": "미검증펜션", "category": 1})).json()["place"]["place_id"]
await client.post(f"/v1/place/{pid}/verify", headers=h, json={"source": 2, "road_address": "미검증주소"})
await client.post(f"/v1/place/{pid}/fact", headers=h, json={
"key": "check_in_time", "value": "15:00",
"source_type": SourceType.CRAWL.value, "source_url": "https://ota.test/1"})
# CRAWL 은 이제 즉시 노출된다.
async with db_engine.begin() as connection:
await connection.execute(text("UPDATE place_facts SET status = :s WHERE place_id = :p"),
{"s": FactStatus.UNVERIFIED.value, "p": uuid.UUID(pid)})
body = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()
assert body["result"]["success"] is True and body["grounded_facts"] == 0
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
job = (await client.get(f"/v1/job/{body['job_id']}", headers=h)).json()["job"]
assert job["status"] == JobStatus.DONE.value, job.get("last_error")
assert job["result"]["faqs"] == 0 and job["result"]["faq_fill"] == 20
assert job["progress"]["steps"][1]["status"] == "skipped"
assert job["progress"]["steps"][1]["reason"] == "no_facts"
live = [r for r in await _faq_rows(db_engine, pid) if r[3] in _PUBLISHABLE]
assert len(live) == 20 and all(r[4] == SourceType.TEMPLATE.value for r in live)
assert any("체크인" in r[0] for r in live)
async def test_zero_facts_still_fill_twenty_without_api_key(auth_headers, client, db_engine, monkeypatch):
"""검증: fact 0건 · GEMINI 키 없음인 펜션(분류 없음 — 스테이머뭄처럼)."""
monkeypatch.setattr(gemini_text, "is_configured", lambda: False)
monkeypatch.setattr(gemini_text, "generate_copy", _must_not_call_llm)
h = await auth_headers("u1")
pid = (await client.post("/v1/place", headers=h, json={"name": "무근거펜션", "category": 1})).json()["place"]["place_id"]
await client.post(f"/v1/place/{pid}/verify", headers=h, json={"source": 2, "road_address": "무근거주소"})
body = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()
assert body["result"]["success"] is True, body["result"]
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
job = (await client.get(f"/v1/job/{body['job_id']}", headers=h)).json()["job"]
assert job["status"] == JobStatus.DONE.value, job.get("last_error")
assert job["result"]["faq_fill"] == 20
live = [r for r in await _faq_rows(db_engine, pid) if r[3] in _PUBLISHABLE]
assert len(live) == 20 and len({r[0] for r in live}) == 20
async def test_faq_without_grounding_is_dropped(auth_headers, client, db_engine, monkeypatch):
"""검증: 근거 fact 가 비어 있는 FAQ 가 생성물에 섞여 온다."""
_patch(monkeypatch, _copy(faqs=[
gemini_text.GeneratedFaq("체크인은?", "15시입니다.", ["check_in_time"]),
gemini_text.GeneratedFaq("수영장 있나요?", "네 있습니다.", []),
]))
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
job_id = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()["job_id"]
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
job = (await client.get(f"/v1/job/{job_id}", headers=h)).json()["job"]
assert job["result"]["faqs"] == 1
assert any("수영장" in str(r) for r in job["result"]["rejected"])
rows = await _faq_rows(db_engine, pid)
assert len([r for r in rows if r[4] == SourceType.LLM.value]) == 1
# 채우기가 넣은 문의 안내에도 '수영장 있나요?' 의 거짓 답("네 있습니다")은 없다.
assert not any("있습니다" in r[1] for r in rows if r[4] == SourceType.TEMPLATE.value)
async def test_rejected_sentences_are_reported(auth_headers, client, monkeypatch):
"""검증: 클라이언트가 반려한 문장이 있다."""
_patch(monkeypatch, _copy(intro=None, rejected=[("수영장을 갖추고 있습니다", "fact 에 없는 시설 '수영장'")]))
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
job_id = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()["job_id"]
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
job = (await client.get(f"/v1/job/{job_id}", headers=h)).json()["job"]
assert job["result"]["intro"] is False
assert any("수영장" in str(r) for r in job["result"]["rejected"])
async def test_regeneration_keeps_faq_the_owner_corrected(auth_headers, client, db_engine, monkeypatch):
"""검증: 사장님이 고친 FAQ 가 있는 상태에서 재생성한다."""
_patch(monkeypatch, _copy())
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
await client.post(f"/v1/place/{pid}/copy", headers=h, json={})
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
# 사장님이 문구를 고쳐 승인 — 잠긴 상태(CORRECTED)가 되고 책임 주체가 사람으로 넘어온다.
async with db_engine.begin() as c:
await c.execute(
text("UPDATE place_faqs SET status = :s, generated_by = :g, question = :q "
"WHERE place_id = :p AND generated_by = :llm"),
{"s": FactStatus.CORRECTED.value, "g": SourceType.OWNER.value,
"q": "체크인 시간이 어떻게 되나요?", "p": uuid.UUID(pid), "llm": SourceType.LLM.value},
)
_patch(monkeypatch, _copy(faqs=[gemini_text.GeneratedFaq("새 질문?", "새 답변", ["check_in_time"])]))
await client.post(f"/v1/place/{pid}/copy", headers=h, json={})
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
rows = await _faq_rows(db_engine, pid)
kept = [r for r in rows if r[3] == FactStatus.CORRECTED.value]
assert kept and kept[0][0] == "체크인 시간이 어떻게 되나요?", "★ 사장님이 고친 FAQ 가 재생성에 밀려났다"
# 새로 생성된 것은 바로 노출값으로 함께 선다.
assert any(r[0] == "새 질문?" and r[3] == FactStatus.VERIFIED.value for r in rows)
_PUBLISHABLE = (FactStatus.VERIFIED.value, FactStatus.CORRECTED.value)
async def test_copy_fills_faqs_to_target_without_duplicates(auth_headers, client, db_engine, monkeypatch):
"""검증: 생성된 FAQ 가 1건뿐이고, 사장님이 직접 쓴 FAQ 가 1건 있다."""
_patch(monkeypatch, _copy())
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
owner = await client.post(f"/v1/place/{pid}/faq", headers=h, json={"question": "보증금이 있나요?", "answer": "없습니다."})
assert owner.json()["result"]["success"] is True
job_id = (await client.post(f"/v1/place/{pid}/copy", headers=h, json={})).json()["job_id"]
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
job = (await client.get(f"/v1/job/{job_id}", headers=h)).json()["job"]
assert job["status"] == JobStatus.DONE.value, job.get("last_error")
assert job["result"]["faqs"] == 1
assert job["result"]["faq_fill"] == 18
live = [r for r in await _faq_rows(db_engine, pid) if r[3] in _PUBLISHABLE]
assert len(live) == 20
assert len({r[0] for r in live}) == 20, "같은 질문이 두 번 들어갔다"
fills = [r for r in live if r[4] == SourceType.TEMPLATE.value]
assert len(fills) == 18
fill_text = " ".join(r[0] for r in fills)
for topic in ("체크인", "체크아웃", "취소", "요리", "반려동물", "보증금"):
assert topic not in fill_text, f"★ 이미 답이 있는 주제 '{topic}' 에 문의 안내가 붙었다"
assert all("문의" in r[1] or "확인" in r[1] for r in fills), "문의 안내가 아닌 답이 섞였다"
async def test_regeneration_replaces_fill_faqs(auth_headers, client, db_engine, monkeypatch):
"""검증: 생성을 두 번 돌린다."""
_patch(monkeypatch, _copy())
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
for _ in range(2):
await client.post(f"/v1/place/{pid}/copy", headers=h, json={})
await Worker("w", JobQueue(), build_handler(), job_deadline_sec=30).drain()
live = [r for r in await _faq_rows(db_engine, pid) if r[3] in _PUBLISHABLE]
assert len(live) == 20
assert len({r[0] for r in live}) == 20
async def test_copy_requires_api_key(auth_headers, client, monkeypatch):
"""검증: GEMINI_API_KEY 없이 생성을 시도한다."""
monkeypatch.setattr(gemini_text, "is_configured", lambda: False)
h = await auth_headers("u1")
pid = await _place_with_facts(client, h)
r = await client.post(f"/v1/place/{pid}/copy", headers=h, json={})
assert r.json()["result"]["code"] == ErrorType.GENERATOR_NOT_CONFIGURED.value