fix(wechat): 公众号新版页面抓不到正文，改从 cgiDataNew 里读

用户发来 https://mp.weixin.qq.com/s/ZEOsi4R4JcmZ_mhS6qpcww，存进库的正文只有 67 个字符（"原文链接 + 一个句号"）。复盘发现公众号换了渲染方式：新版页面把整篇塞进 window.cgiDataNew 这个 JS 对象由前端渲染，页面里根本没有 id="js_content"；三种 UA 拿到的字节数完全一样，也不是 UA 的问题。 - 从 JS 字符串字面量里读正文（content_noencode / desc），按 \xNN、\uNNNN 反转义，话题标签是双重转义过的 HTML，先 unescape 再走 DOM。 - 配图字段排在文字之后，所以前后各找 2500 字符合图；剩下没用过的 mmbiz 图按 12 张上限补在末尾（图文型文章靠就近配对会漏）。 - 解析顺序：旧 DOM → 新版 JS 数据 → 通用抽取，一层层退。实测两篇：2373 字符 + 12 张图、2772 字符 + 13 张图，作者也都取到了。同时补上用户提的两件： - 知识能删了：DELETE /knowledge/{id}?confirm=true，软删 + 真删检索切片，只有作者或 owner/admin 能删；列表卡片（悬停出现）和阅读页动作行各一个入口。删完详情也回 7001，否则收藏的链接还能打开一篇已删除的知识。 - 分类下面有标签清单了：GET /knowledge/tags 按标签数"有多少条知识用了它"，侧栏渲染成 #标签 + 条数，点一下筛选、再点取消。
2026-11-06 02:08:36 +08:00
parent 626f83fb9c
commit 690779c5fb
10 changed files with 720 additions and 15 deletions
@@ -423,3 +423,118 @@ async def test_list_pages_through_every_document(client: AsyncClient) -> None:
    assert cursor is None
    assert sorted(seen) == titles
    assert len(seen) == len(set(seen)), "翻页翻出了重复条目"
+
+
+async def test_delete_removes_from_list_and_search(client: AsyncClient) -> None:
+    """删掉的知识不能还留在列表和检索里——切片要一起删。"""
+    owner = await login(client)
+    ws = await create_workspace(client, owner, "删除知识")
+    note = unwrap(
+        await client.post(
+            f"{V1}/workspaces/{ws['id']}/knowledge",
+            json={
+                "title": "回滚前先停灰度",
+                "body_md": "灰度流量上涨后错误率飙升，第一步先停灰度。",
+                "tag_names": ["发布", "回滚"],
+            },
+            headers=owner["headers"],
+        )
+    )
+    unwrap(
+        await client.post(
+            f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}/status",
+            json={"action": "publish", "version": note["version"]},
+            headers=owner["headers"],
+        )
+    )
+
+    # 不带 confirm 不给删，防手滑
+    resp = await client.delete(
+        f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}", headers=owner["headers"]
+    )
+    assert resp.json()["code"] == 1002
+
+    unwrap(
+        await client.delete(
+            f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}",
+            params={"confirm": "true"},
+            headers=owner["headers"],
+        )
+    )
+
+    page = unwrap(
+        await client.get(f"{V1}/workspaces/{ws['id']}/knowledge", headers=owner["headers"])
+    )
+    assert page["items"] == []
+    hits = unwrap(
+        await client.get(
+            f"{V1}/workspaces/{ws['id']}/knowledge/search",
+            params={"q": "灰度"},
+            headers=owner["headers"],
+        )
+    )
+    assert hits["items"] == []
+    tags = unwrap(
+        await client.get(f"{V1}/workspaces/{ws['id']}/knowledge/tags", headers=owner["headers"])
+    )
+    assert tags["items"] == [], "删掉的知识不该再给标签贡献条数"
+
+    # 删完再从收藏的链接点进来：要当它不存在，不能还渲染出一篇"删掉了的知识"
+    detail = await client.get(
+        f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}", headers=owner["headers"]
+    )
+    assert detail.json()["code"] == 7001
+    again = await client.delete(
+        f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}",
+        params={"confirm": "true"},
+        headers=owner["headers"],
+    )
+    assert again.json()["message"] == "已经删过了", "重复删不该报错，回一句就够"
+
+
+async def test_only_author_or_admin_can_delete(client: AsyncClient) -> None:
+    owner = await login(client)
+    ws = await create_workspace(client, owner, "谁能删")
+    member = await _member(client, owner, ws)
+    note = unwrap(
+        await client.post(
+            f"{V1}/workspaces/{ws['id']}/knowledge",
+            json={"title": "别人的笔记", "body_md": "随手记的"},
+            headers=owner["headers"],
+        )
+    )
+
+    resp = await client.delete(
+        f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}",
+        params={"confirm": "true"},
+        headers=member["headers"],
+    )
+    assert resp.json()["code"] == 3004
+
+    # 作者自己能删
+    unwrap(
+        await client.delete(
+            f"{V1}/workspaces/{ws['id']}/knowledge/{note['id']}",
+            params={"confirm": "true"},
+            headers=owner["headers"],
+        )
+    )
+
+
+async def test_tags_endpoint_counts_documents(client: AsyncClient) -> None:
+    owner = await login(client)
+    ws = await create_workspace(client, owner, "标签统计")
+    for title, tags in (("甲", ["发布", "回滚"]), ("乙", ["发布"])):
+        unwrap(
+            await client.post(
+                f"{V1}/workspaces/{ws['id']}/knowledge",
+                json={"title": title, "tag_names": tags},
+                headers=owner["headers"],
+            )
+        )
+
+    got = unwrap(
+        await client.get(f"{V1}/workspaces/{ws['id']}/knowledge/tags", headers=owner["headers"])
+    )
+    counts = {row["name"]: row["count"] for row in got["items"]}
+    assert counts == {"发布": 2, "回滚": 1}
@@ -43,6 +43,28 @@ BLOCKED_HTML = """
 <html><body><div class="weui-msg"><p>环境异常</p><p>完成验证后即可继续访问</p></div></body></html>
 """

+# 新版公众号页面：没有 #js_content，正文全在 window.cgiDataNew 里；
+# 换行是 \x0a，话题标签是双重转义过的 HTML 片段，图片字段排在文字前面。
+# 正文抽成常量：desc 与 content_noencode 同文（真实页面也是这样，后者更权威），
+# 写成一行会超行宽，所以拆开拼。
+_WECHAT_DATA_BODY = (
+    r"食材： 蒜瓣500g、小米辣60g\x0a\x0a步骤：\x0a1\u20e3蒜瓣打碎，"
+    r"放在冷水里泡10分钟（泡一下蒜不会发苦）\x0a2\u20e3油烧热，转小火把蒜泥炒香"
+    r"\x0a\x0a注意\x26lt;a class=\x26quot;wx_topic_link\x26quot;\x26gt;"
+    r"#调料\x26lt;/a\x26gt;"
+)
+WECHAT_DATA_HTML = (
+    "<html><head>\n"
+    '<meta property="og:title" content="自制万能蒜蓉酱" />\n'
+    "</head><body>\n<script>\n  window.cgiDataNew = {\n"
+    "    nick_name: '破冰记的随笔',\n    item_list: [{\n"
+    "      cdn_url: 'http://mmbiz.qpic.cn/mmbiz_jpg/xyz/640?wx_fmt=jpeg',\n"
+    "      title: '自制万能蒜蓉酱',\n"
+    "      desc: '" + _WECHAT_DATA_BODY + "',\n"
+    "      content_noencode: '" + _WECHAT_DATA_BODY + "'\n"
+    "    }]\n  };\n</script>\n</body></html>\n"
+)
+

 def test_extract_url_only_when_message_is_a_link() -> None:
    assert article_service.extract_url("https://mp.weixin.qq.com/s/abc") == (
@@ -71,6 +93,24 @@ def test_wechat_article_parses_title_body_and_images() -> None:
    assert "扫码关注" not in got.body_md


+def test_new_wechat_page_reads_js_data() -> None:
+    """新版页面没有 #js_content：正文在 cgiDataNew 的 JS 字符串里，得能读出来。"""
+    got = article_service.parse_article(WECHAT_DATA_HTML, "https://mp.weixin.qq.com/s/abc")
+
+    assert got.title == "自制万能蒜蓉酱"
+    assert got.author == "破冰记的随笔"
+    assert "蒜瓣打碎" in got.body_md
+    # \x0a 要还原成换行，否则整篇挤成一行
+    assert "\n" in got.body_md
+    # 双重转义的话题标签还原成文字，不能留下 &lt;a 这种碎片
+    assert "#调料" in got.body_md
+    assert "&lt;" not in got.body_md
+    # 配图要跟着正文进来（正文里换成本地地址之前，先能取到这张图）
+    assert "mmbiz.qpic.cn/mmbiz_jpg/xyz/640" in got.body_md
+    # desc 与 content_noencode 是同一段文字，不能重复两遍
+    assert got.body_md.count("蒜瓣打碎") == 1
+
+
 def test_blocked_wechat_page_is_reported() -> None:
    with pytest.raises(article_service.ArticleError):
        article_service.parse_article(BLOCKED_HTML, "https://mp.weixin.qq.com/s/abc")