fix(knowledge): related_page 过滤精确化(子串包含 -> 归一候选键等值)

KBFF-v11c(派单 KBFF-v11c-relatedpage-exact-match-filter · A 侧 wecom-yanxue · 执行人 wecom-project)
本提交=代码面切片:未 push、未部署;生产部署另走发布单 + approval-gate(回滚 = git revert <本 sha>)。

- backend/app/api/knowledge_articles.py:related_page 过滤由 LIKE '%<v>%'(子串包含,短键/前缀超集命中)
  收敛为「归一候选键 IN ∪ 模板子路由 <键>/:<占位符>」;不可归一入参 -> 空集(不退化为全量)
- backend/app/utils/route_key.py(新增):匹配口径唯一实现处(归一化 / 候选键派生 / 模板模式)
- backend/tests/test_kbff_v11c_related_page_exact_match_20261012.py(新增):42 用例
  —— 生产 seed 7 值逐值回归(各中 1 行)、短键 /maps->0 行 /kpi->0 行、带参路由与尾斜杠/query/大小写兼容、
  改造前 contains 语义负对照、账套分区与 category 组合语义保持

(cherry picked from commit 2a5cb182c237ae724d4b506b68eb48810ba27b16)
This commit is contained in:
wecom-project
2026-10-18 16:31:52 +08:00
committed by Hermes CI Fix
parent 60b6ef3bfe
commit 5fa83e2532
3 changed files with 399 additions and 1 deletions
@@ -0,0 +1,255 @@
# -*- coding: utf-8 -*-
"""KBFF-v11c · `/knowledge-articles?related_page=` 过滤精确化回归(2026-10-12)
派单:`KBFF-v11c-relatedpage-exact-match-filter`(A 侧 wecom-yanxue)
口径实现:`app/utils/route_key.py`;改造点:`app/api/knowledge_articles.py`
纪律:本文件只新增测试,不改 app/**、conftest.py、任何既有 test_*.py。
覆盖:
§1 单位:normalize_route / candidate_route_keys / route_template_child_pattern(表驱动)
§2 负对照:证明改造前 contains 语义确会超集命中(同库同 seed 直构旧表达式)
§3 生产 seed 7 值逐值回归(值取自 backend/scripts/seed_knowledge_articles.py)
—— 每个值各自只中自己那 1 行;短键 `/maps` 0 行、`/kpi` 0 行
§4 带参路由 / 尾斜杠 / query-fragment / 大小写 / 不可归一 兼容面
§5 模板子路由(存储值写占位符,任意占位符命名均可命中)
§6 既有语义保持:账套分区 + category 组合过滤不受影响
"""
import pytest
from fastapi.testclient import TestClient
from sqlalchemy.orm import Session
from app.models import Entity
from app.models.knowledge_article import KnowledgeArticle
from app.utils.route_key import (
candidate_route_keys,
normalize_route,
route_template_child_pattern,
template_route,
)
from tests.conftest import auth_header, create_test_user
KA = "/api/cma/knowledge-articles"
#: 生产 seed 的 7 个 related_page 值(backend/scripts/seed_knowledge_articles.py:41,61,83,113,139,166,189)
SEED = [
(1, "/maps/canvas"),
(2, "/kpis"),
(3, "/budget"),
(4, "/deviations"),
(5, "/predict"),
(6, "/maps-review"),
(7, "/alerts"),
]
# ── 夹具 ────────────────────────────────────────────────────────────
def _ensure_entity(db: Session, eid: int, name: str = None, short: str = None) -> Entity:
e = db.query(Entity).filter(Entity.id == eid).first()
if e:
return e
e = Entity(id=eid, name=name or f"实体{eid}", short_name=short or f"E{eid}", status="active")
db.add(e)
db.commit()
db.refresh(e)
return e
def _login(client: TestClient, entity_id: int = 1) -> str:
r = client.post("/api/cma/auth/login", json={
"username": "testadmin", "password": "test-fixture-pass", "entity_id": entity_id,
})
assert r.status_code == 200, r.text
j = r.json()
return j.get("token") or j.get("access_token")
def _seed_rows(db: Session, rows) -> None:
"""rows: [(id, related_page)],全局共享(entity_id=None)便于账套无关复算。"""
for i, rp in rows:
db.add(KnowledgeArticle(title=f"seed-{i}", content="c", category="term",
entity_id=None, related_page=rp, sort_order=i))
db.commit()
# ── §1 单位 ─────────────────────────────────────────────────────────
@pytest.mark.parametrize("raw,expected", [
("/maps/canvas", "/maps/canvas"),
("/maps/canvas/", "/maps/canvas"),
("/maps//canvas/", "/maps/canvas"),
("maps/canvas", "/maps/canvas"),
("/MAPS/Canvas", "/maps/canvas"),
("/maps/canvas?id=42", "/maps/canvas"),
("/maps/canvas#top", "/maps/canvas"),
(" /budget ", "/budget"),
("/", "/"),
("", ""),
(" ", ""),
(None, ""),
("?only=query", ""),
])
def test_u1_normalize_route(raw, expected):
assert normalize_route(raw) == expected
@pytest.mark.parametrize("raw,expect_members,expect_absent", [
("/maps/canvas", {"/maps/canvas"}, {"/maps", "/maps/canvas/:id"}),
("/maps/canvas/42", {"/maps/canvas/42", "/maps/canvas", "/maps/canvas/:id"},
set()),
("/maps/canvas/:id", {"/maps/canvas/:id", "/maps/canvas"}, {"/maps"}),
("/maps", {"/maps"}, {"/maps/canvas"}),
("/kpi", {"/kpi"}, {"/kpis"}),
("", set(), set()),
(" ", set(), set()),
])
def test_u2_candidate_route_keys(raw, expect_members, expect_absent):
keys = candidate_route_keys(raw)
assert set(keys) >= expect_members, f"{raw} -> {keys}"
assert not (set(keys) & expect_absent), f"{raw} 不该派生 {expect_absent}:{keys}"
assert list(keys) == sorted(keys), "候选键须排序去重,便于同库复算与取证"
@pytest.mark.parametrize("path,expected", [
("/maps/canvas/42", "/maps/canvas/:id"),
("/maps/canvas", "/maps/canvas"),
("/kpis", "/kpis"),
])
def test_u3_template_route(path, expected):
assert template_route(path) == expected
def test_u4_template_child_pattern_shape():
assert route_template_child_pattern("/maps/canvas") == "/maps/canvas/:%"
# LIKE 元字符须转义(否则用户串可注入通配)
assert route_template_child_pattern("/a_b%c") == "/a\\_b\\%c/:%"
# ── §2 负对照(改造前语义) ──────────────────────────────────────────
def test_n1_contains_semantics_superset_control(client: TestClient, db: Session):
"""负对照:旧 `.contains()` 语义下 `/maps` 超集命中 2 行、`/kpi` 命中 `/kpis`。
证明本改造确有行为变化(不是“改了个等价写法”)。"""
create_test_user(db)
_seed_rows(db, SEED)
tok = _login(client)
assert client.get(f"{KA}?related_page=/maps/canvas", headers=auth_header(tok)).status_code == 200
def old_query(v):
return (db.query(KnowledgeArticle)
.filter(KnowledgeArticle.related_page.contains(v))
.order_by(KnowledgeArticle.id.asc()).all())
assert [r.related_page for r in old_query("/maps")] == ["/maps/canvas", "/maps-review"]
assert [r.related_page for r in old_query("/kpi")] == ["/kpis"]
# 生成 SQL 亦确认为子串包含
sql = str(db.query(KnowledgeArticle)
.filter(KnowledgeArticle.related_page.contains("/maps"))
.statement.compile(compile_kwargs={"literal_binds": True}))
assert "LIKE" in sql and "%" in sql
# ── §3 生产 seed 7 值逐值回归 ────────────────────────────────────────
def test_f1_seven_seed_values_exact(client: TestClient, db: Session):
create_test_user(db)
_seed_rows(db, SEED)
tok = _login(client)
for sid, rp in SEED:
r = client.get(f"{KA}?related_page={rp}", headers=auth_header(tok))
assert r.status_code == 200, r.text
got = [(i["id"], i["related_page"]) for i in r.json()["data"]]
assert got == [(sid, rp)], f"{rp} 期望精确 1 行({sid}),实得 {got}"
def test_f2_short_key_and_prefix_zero(client: TestClient, db: Session):
"""短键/前缀必须 0 行:/maps(否则连带 /maps/canvas + /maps-review)、/kpi(否则 /kpis)。"""
create_test_user(db)
_seed_rows(db, SEED)
tok = _login(client)
for prop in ["/maps", "/kpi", "/map", "/budg", "/alerts-extra", "/maps-revie"]:
r = client.get(f"{KA}?related_page={prop}", headers=auth_header(tok))
assert r.status_code == 200, r.text
assert r.json()["data"] == [], f"{prop} 应 0 行,实得 {r.json()['data']}"
# ── §4 兼容面 ───────────────────────────────────────────────────────
@pytest.mark.parametrize("prop,expect_id", [
("/maps/canvas/42", 1), # 带参实例 → 父页 + 模板派生
("/maps/canvas/:id", 1), # 模板形式直传
("/maps/canvas/", 1), # 尾斜杠
("/MAPS/CANVAS", 1), # 大小写
("/maps/canvas?id=42", 1), # query
("/budget%23top", 3), # fragment(编码传参,验证落库前的截断)
("%20%20/kpis%20%20", 2), # 首尾空白
("/alerts", 7),
])
def test_f3_compat_forms(client: TestClient, db: Session, prop, expect_id):
create_test_user(db)
_seed_rows(db, SEED)
tok = _login(client)
r = client.get(f"{KA}?related_page={prop}", headers=auth_header(tok))
assert r.status_code == 200, r.text
assert {i["id"] for i in r.json()["data"]} == {expect_id}, f"{prop} -> {r.json()['data']}"
def test_f4_unmatchable_prop_returns_empty_not_all(client: TestClient, db: Session):
"""不可归一入参(纯 query/空白)⇒ 空集,**不得**退化为“不过滤=全量”。"""
create_test_user(db)
_seed_rows(db, SEED)
tok = _login(client)
for prop in ["%20", "?only=query", "%23frag"]:
r = client.get(f"{KA}?related_page={prop}", headers=auth_header(tok))
assert r.status_code == 200, r.text
assert r.json()["data"] == [], f"{prop} 应空集,实得 {len(r.json()['data'])} 行"
# ── §5 模板子路由 ───────────────────────────────────────────────────
@pytest.mark.parametrize("stored,prop,hit", [
("/maps/canvas/:articleId", "/maps/canvas", True),
("/maps/canvas/:id/step", "/maps/canvas", True),
("/maps/canvas", "/maps/canvas", True),
("/maps-review", "/maps", False),
("/maps/canvas/detail", "/maps/canvas/detail", True),
])
def test_f5_template_child(db: Session, client: TestClient, stored, prop, hit):
create_test_user(db)
_seed_rows(db, [(1, stored)])
tok = _login(client)
r = client.get(f"{KA}?related_page={prop}", headers=auth_header(tok))
assert r.status_code == 200, r.text
got = r.json()["data"]
if hit:
assert [i["related_page"] for i in got] == [stored], f"{prop} 应命中 {stored},实得 {got}"
else:
assert got == [], f"{prop} 不应命中 {stored}"
# ── §6 既有语义保持 ────────────────────────────────────────────────
def test_f6_partition_and_category_still_hold(client: TestClient, db: Session):
"""账套分区(本企业 ∪ 全局)与 category 组合过滤在精确口径下不变。"""
_ensure_entity(db, 2, "陕西博海", "博海")
rows = [
KnowledgeArticle(title="K1", content="c", entity_id=1, related_page="/kpis", category="term"),
KnowledgeArticle(title="K2", content="c", entity_id=1, related_page="/kpis", category="practice"),
KnowledgeArticle(title="K3", content="c", entity_id=1, related_page="/budget", category="term"),
KnowledgeArticle(title="K4", content="c", entity_id=None, related_page=None, category="term"),
KnowledgeArticle(title="K5", content="c", entity_id=2, related_page="/kpis", category="term"),
]
db.add_all(rows)
db.commit()
for a in rows:
db.refresh(a)
create_test_user(db)
tok = _login(client)
rp = client.get(f"{KA}?related_page=/kpis", headers=auth_header(tok)).json()["data"]
assert {i["id"] for i in rp} == {rows[0].id, rows[1].id}
rc = client.get(f"{KA}?category=term", headers=auth_header(tok)).json()["data"]
assert {i["id"] for i in rc} == {rows[0].id, rows[2].id, rows[3].id}
rb = client.get(f"{KA}?related_page=/kpis&category=term", headers=auth_header(tok)).json()["data"]
assert {i["id"] for i in rb} == {rows[0].id}