From 690779c5fbcae514583b71c4f6da4af53c9b3f34 Mon Sep 17 00:00:00 2001 From: Steven Date: Fri, 6 Nov 2026 02:08:36 +0800 Subject: [PATCH] =?UTF-8?q?fix(wechat):=20=E5=85=AC=E4=BC=97=E5=8F=B7?= =?UTF-8?q?=E6=96=B0=E7=89=88=E9=A1=B5=E9=9D=A2=E6=8A=93=E4=B8=8D=E5=88=B0?= =?UTF-8?q?=E6=AD=A3=E6=96=87=EF=BC=8C=E6=94=B9=E4=BB=8E=20cgiDataNew=20?= =?UTF-8?q?=E9=87=8C=E8=AF=BB?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 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。 --- 2.Design/API.md | 28 ++++- 2.Design/ROADMAP-M0-M1.md | 32 +++++- backend/app/api/knowledge.py | 94 +++++++++++++++- backend/app/services/article.py | 125 ++++++++++++++++++++- backend/tests/test_knowledge.py | 115 +++++++++++++++++++ backend/tests/test_wechat_article.py | 40 +++++++ web/e2e/knowledge-tags-delete.mjs | 160 +++++++++++++++++++++++++++ web/src/lib/types.ts | 6 + web/src/pages/Knowledge.tsx | 112 ++++++++++++++++++- web/src/pages/KnowledgeDoc.tsx | 23 +++- 10 files changed, 720 insertions(+), 15 deletions(-) create mode 100644 web/e2e/knowledge-tags-delete.mjs diff --git a/2.Design/API.md b/2.Design/API.md index f3c8fdd..c9e286f 100644 --- a/2.Design/API.md +++ b/2.Design/API.md @@ -513,7 +513,30 @@ Web 端未读数由 `AppShell` 每 30 秒轮询一次；在通知页标已读后 ### 5.2 `GET /knowledge/{item_id}` -详情，附最近 50 条 `activities`。不存在或跨租户统一 `7001`。 +详情，附最近 50 条 `activities`。不存在、跨租户、**已经删掉**的统一 `7001` +（删除是软删，但对读的人来说就是"没了"，否则收藏的链接还会打开一篇已删除的知识）。 + +### 5.2.1 `GET /knowledge/tags` — 侧栏那串标签 + +``` +GET /knowledge/tags?status=published&status=in_review +``` + +```json +{"items": [{"name": "发布", "count": 12}, {"name": "回滚", "count": 3}]} +``` + +- `count` 是**有多少条知识用了它**（`count(distinct item_id)`），不是引用次数。 +- 只统计未删除的知识；带了 `status` 就只数这些状态，和列表的 tab 对得上。 +- 按条数从多到少排，同数按名字。标签是分类之外的第二根轴：分类回答"属于哪一块"，标签回答"跟什么有关"。 + +### 5.2.2 `DELETE /knowledge/{item_id}?confirm=true` — 删除 + +- 必须显式带 `confirm=true`，否则 `1002`「删除不能撤销，带 confirm=true 再来一次」（防手滑）。 +- 只有**作者本人**或 `owner`/`admin` 能删，否则 `3004`。 +- 软删：`item.deleted_at` 落时间戳，列表、详情、检索、标签统计里都不再出现； + **检索切片真删**（`chunk` 表），否则"删掉的知识还能被搜出来"比删不掉更吓人。 +- 重复删不回错误，回 `message: "已经删过了"`，让重试的客户端不必区分。 ### 5.3 `POST /knowledge` — 手记或指定来源 @@ -758,7 +781,8 @@ Query： |---|---| | 判断 | `extract_url()`：去掉首尾标点后只剩一个链接；带了别的话（"看下 https://… 这个"）仍按任务处理 | | 抓取 | `httpx` 取页面（25 秒超时、8 MB 上限），`trafilatura` 抽正文；标题取 `og:title`／`

`／`` | -| 公众号 | `mp.weixin.qq.com` 单独走 DOM：标题 `#activity-name`、作者 `#js_name`、正文 `#js_content`，图片取 `data-src`（通用抽取器会把公众号的图丢光） | +| 公众号（旧页） | `mp.weixin.qq.com` 且页面有 `#js_content`：标题 `#activity-name`、作者 `#js_name`、正文 `#js_content`，图片取 `data-src`（通用抽取器会把公众号的图丢光） | +| 公众号（新版页） | 没有 `#js_content` 时（`appmsg_type = '9'` / `'10002'`）整篇在 `window.cgiDataNew` 里：从 `content_noencode` / `desc` 的 JS 字符串字面量读正文，按 `\xNN`、`\uNNNN` 反转义，话题标签先 `html.unescape` 再走 DOM；配图字段（`cdn_url`/`img_url`/`pic_url`）排在文字之后，前后各找 2500 字符合图；作者兜底取 `nick_name` | | 清洗 | 导航/页脚按行边界裁掉（`_trim_chrome`）；`trafilatura` 结果太短（<120 字）就退到容器打分兜底 | | 配图 | 正文里的外链图下载进我们自己的存储，最多 12 张、单张 8 MB，正文改写成 `/attachments/{id}/raw?签名`——公众号一换图，外链就是一排红叉 | | 落库 | 知识草稿（`status="draft"`，`doc_type="note"`），正文首行是 `> 原文：<url>` | diff --git a/2.Design/ROADMAP-M0-M1.md b/2.Design/ROADMAP-M0-M1.md index 7696b71..28386d3 100644 --- a/2.Design/ROADMAP-M0-M1.md +++ b/2.Design/ROADMAP-M0-M1.md @@ -857,7 +857,37 @@ `lxml / courlan / htmldate / jusText / lxml_html_clean / …` 再 `pip install --no-deps trafilatura`； `lxml` 6 起 `lxml.html.clean` 拆成 `lxml_html_clean`，不装就 `ImportError`。 -### 13.9 还没做 +#### 13.8.1 用户发来真链接，结果只抓到一个「。」 + +上线后用户发了 `https://mp.weixin.qq.com/s/ZEOsi4R4JcmZ_mhS6qpcww`，库里只有 67 个字符的正文。 +复盘：**公众号换了渲染方式**——新版页面（`window.appmsg_type = '9'` / `'10002'`）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（桌面 / iPhone Safari / MicroMessenger）拿到的字节数完全一样，所以也不是 UA 的问题。 + +修法（`backend/app/services/article.py`）： + +- 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的 HTML（`\x26lt;a class=\x26quot;wx_topic_link\x26quot;\x26gt;#调料…`）， + 先 `html.unescape` 再走 DOM。 +- 配图字段（`cdn_url` / `img_url` / `pic_url`）**排在文字字段之后**，所以配图要前后各找 2500 字符； + 剩下没用过的 `mmbiz.qpic.cn` 图按 12 张上限补在末尾（图文型「小绿书」的图不在文字附近）。 +- 解析顺序：旧 DOM（`#js_content`）→ 新版 JS 数据 → 通用抽取，一层层退。 +- 作者兜底从 `nick_name` 取，标题取 `og:title`。 + +结果：那篇 2373 字符 + 12 张图，蒜蓉酱那篇 2772 字符 + 13 张图，作者都取到了。 +**教训：抓取器写死在一种页面结构上，等于给自己埋一个定时炸弹；线上先看真实 HTML 再改。** + +### 13.9 知识库：标签清单与删除 + +用户提两件："文章都缺少删除功能"、"分类有列表显示，但是分类的下面应该还有标签的列表"。 + +- `GET /knowledge/tags`：按标签统计"有多少条知识用了它"，带 `status` 就跟着 tab 数； + 侧栏在分类树下面渲染成一排 `#标签 + 条数`，点一下筛选、再点取消。 +- `DELETE /knowledge/{item_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）和阅读页动作行各有一个入口。删完详情也回 `7001`， + 否则收藏的链接还能打开一篇"已经删掉的知识"。 + +### 13.10 还没做 - 只做了"进"（收消息建事项），没做"出"（任务有进展回一条微信通知）。 - 消息不认人：按通道创建者记账；要认人得先做微信身份绑定。 diff --git a/backend/app/api/knowledge.py b/backend/app/api/knowledge.py index 202d20e..a012ffb 100644 --- a/backend/app/api/knowledge.py +++ b/backend/app/api/knowledge.py @@ -22,7 +22,7 @@ from typing import Annotated, Any from fastapi import APIRouter, Query from pydantic import BaseModel, Field -from sqlalchemy import Select, and_, func, or_, select, update +from sqlalchemy import Select, and_, delete, func, or_, select, update from sqlalchemy.ext.asyncio import AsyncSession from ..audit import record as audit_record @@ -33,6 +33,7 @@ from ..models_knowledge import ( DOC_TYPES, KNOWLEDGE_STATUSES, MAX_CATEGORY_DEPTH, + Chunk, KnowledgeCategory, KnowledgeDoc, ) @@ -96,11 +97,23 @@ class StatusActionIn(BaseModel): # ── 内部工具 ──────────────────────────────────────────────────────────── async def _load_doc( - session: AsyncSession, ws_id: uuid.UUID, item_id: uuid.UUID + session: AsyncSession, ws_id: uuid.UUID, item_id: uuid.UUID, *, include_deleted: bool = False ) -> tuple[Item, KnowledgeDoc]: + """取一条知识。 + + 删除是软删（`deleted_at`），所以默认把已删除的当成"不存在"： + 否则删完还能从收藏的链接、浏览器历史里点开，用户会以为删除没生效。 + 只有删除接口自己需要 `include_deleted=True`，好让重复删回一句"已经删过了"。 + """ item = await session.get(Item, item_id) doc = await session.get(KnowledgeDoc, item_id) - if item is None or doc is None or item.workspace_id != ws_id or item.kind != "knowledge": + if ( + item is None + or doc is None + or item.workspace_id != ws_id + or item.kind != "knowledge" + or (item.deleted_at is not None and not include_deleted) + ): raise AppError(ErrorCode.KNOWLEDGE_NOT_FOUND, "这份知识不存在") return item, doc @@ -595,8 +608,39 @@ async def delete_category( return ok({"deleted": str(category_id), "uncategorized_docs": docs}) -# 注意路由顺序：`/knowledge/search` 必须排在 `/knowledge/{item_id}` 之前， -# 否则 "search" 会被当成 uuid 解析，返回 422。 +# 注意路由顺序：`/knowledge/search` 与 `/knowledge/tags` 必须排在 +# `/knowledge/{item_id}` 之前，否则会被当成 uuid 解析，返回 422。 +@router.get("/knowledge/tags", summary="知识用到的标签（带条数）") +async def list_knowledge_tags( + ctx: MemberDep, + session: SessionDep, + status: Annotated[list[str] | None, Query()] = None, +) -> dict: + """只统计"还在用"的标签，给右侧那串标签用，点一下按标签筛。 + + 条数是**知识条数**，不是引用次数——同一个标签在一条知识里挂两次也只算一条。 + """ + stmt: Select = ( + select(Tag.name, func.count(func.distinct(Item.id)).label("docs")) + .join(ItemTag, ItemTag.tag_id == Tag.id) + .join(Item, Item.id == ItemTag.item_id) + .join(KnowledgeDoc, KnowledgeDoc.item_id == Item.id) + .where( + Item.workspace_id == ctx.ws_id, + Item.kind == "knowledge", + Item.deleted_at.is_(None), + Tag.workspace_id == ctx.ws_id, + ) + .group_by(Tag.name) + .order_by(func.count(func.distinct(Item.id)).desc(), Tag.name) + ) + wanted = [s for s in (status or []) if s in KNOWLEDGE_STATUSES] + if wanted: + stmt = stmt.where(KnowledgeDoc.status_hint.in_(wanted)) + rows = (await session.execute(stmt)).all() + return ok({"items": [{"name": name, "count": count} for name, count in rows]}) + + @router.get("/knowledge/search", summary="检索已发布知识") async def search( ctx: MemberDep, @@ -756,6 +800,46 @@ async def patch_knowledge( return ok(await _knowledge_out(session, ws_id, item, doc)) +@router.delete("/knowledge/{item_id}", summary="删除知识（软删，连带删掉检索切片）") +async def delete_knowledge( + item_id: uuid.UUID, + ctx: MemberDep, + session: SessionDep, + confirm: Annotated[bool, Query(description="必须显式带 true，防手滑")] = False, +) -> dict: + """删除是软删：条目留在库里（活动记录、来源关系都还在），只是不再出现在列表与检索里。 + + 切片必须真删：检索只查 `chunk` 表，留着的话"已经删掉的知识"还能被搜出来， + 那比删不掉更吓人。 + """ + ws_id = ctx.ws_id + if not confirm: + raise AppError(ErrorCode.VALIDATION_FAILED, "删除不能撤销，带 confirm=true 再来一次") + item, doc = await _load_doc(session, ws_id, item_id, include_deleted=True) + if item.deleted_at is not None: + return ok({"id": str(item.id)}, message="已经删过了") + if item.creator_id != ctx.user.id and ctx.role not in ("owner", "admin"): + raise AppError(ErrorCode.INSUFFICIENT_ROLE, "只有作者本人或管理员能删这条知识") + + item.deleted_at = datetime.now(timezone.utc) + item.version = (item.version or 1) + 1 + await session.execute( + delete(Chunk).where(Chunk.workspace_id == ws_id, Chunk.item_id == item.id) + ) + doc.chunk_count = 0 + await session.flush() + await audit_record( + session, + workspace_id=ws_id, + actor_user_id=ctx.user.id, + action="knowledge.deleted", + target_type="item", + target_id=str(item.id), + payload={"title": item.title, "status": doc.status_hint}, + ) + return ok({"id": str(item.id)}, message="已删除") + + # ── 状态动作 ──────────────────────────────────────────────────────────── @router.post("/knowledge/{item_id}/status", summary="提交审议 / 发布 / 归档 / 恢复") async def change_status( diff --git a/backend/app/services/article.py b/backend/app/services/article.py index e1a93eb..b7fb283 100644 --- a/backend/app/services/article.py +++ b/backend/app/services/article.py @@ -15,6 +15,7 @@ from __future__ import annotations +import html as html_mod import ipaddress import re import socket @@ -55,6 +56,15 @@ WECHAT_UA = ( IMAGE_EXTS = {"jpeg": "jpg", "jpg": "jpg", "png": "png", "gif": "gif", "webp": "webp"} +# 公众号新版页面（2024 之后）不再把正文放进 `#js_content`，而是把整篇塞进 +# `window.cgiDataNew` 这个 JS 对象里，由前端渲染。所以还得会读 JS 字符串字面量： +# 换行写成 `\x0a`，话题链接是 `\x26lt;a …\x26gt;`（先转义成实体、再转义成 JS）。 +_WECHAT_FIELD_RE = re.compile(r"(?:content_noencode|desc)\s*:\s*'") +_WECHAT_IMAGE_RE = re.compile( + r"""(?:cdn_url|img_url|pic_url)\s*:\s*['"]((?:https?:)?//[^'"]*mmbiz\.qpic\.cn[^'"]*)['"]""" +) +_JS_ESCAPE_RE = re.compile(r"\\(x[0-9a-fA-F]{2}|u[0-9a-fA-F]{4}|.)", re.S) + class ArticleError(RuntimeError): """抓取/解析没成功。调用方据此决定降级方式。""" @@ -174,6 +184,10 @@ def _wechat_meta(html: str) -> tuple[str, str, list[tuple[str, str]]]: author = " ".join(p for p in parts if p) if author: break + if not author: + # 新版页面没有 #js_name，公众号名在 cgiDataNew 的 nick_name 字段里 + found = re.search(r"nick_name\s*:\s*'([^']{1,80})'", html) + author = _clean_text(found.group(1)) if found else "" return title, author, tree @@ -244,6 +258,115 @@ def _wechat_body(html: str) -> str: return "\n\n".join(s.strip() for s in "\n".join(out).split("\n\n") if s.strip()) +def _js_literal(text: str, quote_pos: int) -> str: + """读一个单引号 JS 字符串的原始内容（`quote_pos` 指向那个开引号）。""" + out: list[str] = [] + i = quote_pos + 1 + while i < len(text): + ch = text[i] + if ch == "\\": + out.append(text[i : i + 2]) + i += 2 + continue + if ch == "'": + break + out.append(ch) + i += 1 + return "".join(out) + + +def _unescape_js(raw: str) -> str: + def repl(match: re.Match[str]) -> str: + seq = match.group(1) + if seq[0] in "xu" and len(seq) > 1: + return chr(int(seq[1:], 16)) + return {"n": "\n", "t": "\t", "r": "\r", "'": "'", '"': '"', "\\": "\\"}.get(seq, seq) + + return _JS_ESCAPE_RE.sub(repl, raw) + + +def _wechat_data_body(html_text: str) -> str: + """新版公众号：从 `window.cgiDataNew` 的 JS 字段里把正文和配图捞出来。 + + 一篇文章是若干「图片 + 一段说明文字」拼起来的，JS 里是 + `cdn_url: '…'` 后面紧跟 `desc` / `content_noencode`，所以按出现顺序取文字， + 再往回找它自己那张图，配图就跟着段落走。 + """ + blocks: list[str] = [] + seen_text: set[str] = set() + used_images: set[str] = set() + for match in _WECHAT_FIELD_RE.finditer(html_text): + piece = _fragment_to_markdown(_unescape_js(_js_literal(html_text, match.end() - 1))) + if len(piece) < 30 or piece in seen_text: + continue + seen_text.add(piece) + image = _nearby_image(html_text, match.start(), used_images) + if image and "![](" not in piece: + used_images.add(image) + piece = f"{piece}\n\n![]({image})" + blocks.append(piece) + # 图文型文章（"小绿书"）的图不在文字附近，靠上面的就近配对会漏掉； + # 剩下的正文图统一补在末尾，宁可顺序差点也别丢图。 + leftover: list[str] = [] + for match in _WECHAT_IMAGE_RE.finditer(html_text): + image = match.group(1) + image = image if image.startswith("http") else f"https:{image}" + if image in used_images: + continue + used_images.add(image) + leftover.append(f"![]({image})") + if len(leftover) >= MAX_IMAGES: + break + if leftover: + blocks.append("\n\n".join(leftover)) + return "\n\n".join(blocks).strip() + + +def _fragment_to_markdown(fragment: str) -> str: + """把「实体转义 + 夹着标签」的正文片段变成 markdown 文本。 + + 公众号把话题标签写成 `<a class="wx_topic_link" …>#调料</a>`， + 先 unescape 再当 HTML 处理，标签会被去掉、只留文字；`<img>` 转成 markdown。 + """ + from lxml import html as lxml_html + + fragment = html_mod.unescape(fragment or "") + fragment = re.sub( + r"""<img[^>]*?(?:data-src|src)\s*=\s*["']([^"']+)["'][^>]*>""", + lambda m: f"\n\n![]({m.group(1)})\n\n", + fragment, + flags=re.I, + ) + fragment = re.sub(r"<br\s*/?>", "\n", fragment, flags=re.I) + try: + tree = lxml_html.fragment_fromstring(fragment, create_parent="div") + text = tree.text_content() + except Exception: # 片段结构烂到 lxml 都不认，直接剥标签 + text = re.sub(r"<[^>]+>", "", fragment) + text = text.replace("\u00a0", " ") + lines = [re.sub(r"[ \t\u3000]+", " ", line).strip() for line in text.splitlines()] + out: list[str] = [] + for line in lines: + if not line and (not out or not out[-1]): + continue + out.append(line) + return "\n".join(out).strip() + + +def _nearby_image(html_text: str, pos: int, used: set[str]) -> str | None: + """这条文字对应的配图：先往后找（新版页面图片字段排在文字之后），再往回找。""" + for start, end in ((pos, pos + 2500), (max(0, pos - 2500), pos)): + found = [ + m.group(1) + for m in _WECHAT_IMAGE_RE.finditer(html_text, start, min(end, len(html_text))) + if m.group(1) not in used + ] + if found: + image = found[0] + return image if image.startswith("http") else f"https:{image}" + return None + + def _block_markdown(html: str) -> str: """兜底的正文提取：找出"装着最多段文字"的那个容器，再按段落拼成 markdown。 @@ -371,7 +494,7 @@ def parse_article(html: str, url: str) -> Article: raise ArticleError("微信把这篇拦下来了（需要人工验证或链接已失效）") if is_wechat_article(url): title, author, _tree = _wechat_meta(html) - body = _wechat_body(html) + body = _wechat_body(html) or _wechat_data_body(html) if not body.strip(): # 公众号偶尔会换结构，退回通用抽取，宁可少图也别丢正文 article = _generic_article(html, url) diff --git a/backend/tests/test_knowledge.py b/backend/tests/test_knowledge.py index 568eb87..d9e9c62 100644 --- a/backend/tests/test_knowledge.py +++ b/backend/tests/test_knowledge.py @@ -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} diff --git a/backend/tests/test_wechat_article.py b/backend/tests/test_wechat_article.py index a15d463..ef0a7e8 100644 --- a/backend/tests/test_wechat_article.py +++ b/backend/tests/test_wechat_article.py @@ -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 <a 这种碎片 + assert "#调料" in got.body_md + assert "<" 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") diff --git a/web/e2e/knowledge-tags-delete.mjs b/web/e2e/knowledge-tags-delete.mjs new file mode 100644 index 0000000..456c925 --- /dev/null +++ b/web/e2e/knowledge-tags-delete.mjs @@ -0,0 +1,160 @@ +/** + * 知识库的两件小事：分类下面的标签清单能点着筛，文章能删。 + * + * 为什么单独写一条：标签统计和删除都是"后端好了但界面没入口"的典型， + * 看不见就等于没有；所以这里从界面点，最后用接口回查确认真的生效。 + * + * 用法： + * node web/e2e/knowledge-tags-delete.mjs # 默认 dev 5180 + * WEB=http://192.168.8.188:8200 PHONE=13605513683 PASSWORD=... node web/e2e/knowledge-tags-delete.mjs + */ +import { readFile } from "node:fs/promises"; +import { chromium } from "playwright-core"; + +const WEB = process.env.WEB ?? "http://localhost:5180"; +const API = process.env.API ?? `${WEB}/api/v1`; +const PHONE = process.env.PHONE ?? "13900000001"; +const CHROME = + process.env.CHROME ?? + `${process.env.LOCALAPPDATA}\\ms-playwright\\chromium-1223\\chrome-win64\\chrome.exe`; + +let pass = 0; +let fail = 0; +function check(name, ok, extra = "") { + if (ok) { + pass += 1; + console.log(`PASS ${name}`); + } else { + fail += 1; + console.log(`FAIL ${name} ${extra}`); + } +} + +async function readPassword() { + if (process.env.PASSWORD) return process.env.PASSWORD; + try { + const file = new URL("../../.local/e2e-account-password.txt", import.meta.url); + const text = (await readFile(file, "utf8")).trim(); + return text || null; + } catch { + return null; + } +} + +async function call(path, { method = "GET", body, token } = {}) { + const headers = { "Content-Type": "application/json" }; + if (token) headers.Authorization = `Bearer ${token}`; + const r = await fetch(`${API}${path}`, { + method, + headers, + body: body === undefined ? undefined : JSON.stringify(body), + }); + return r.json(); +} + +async function main() { + const password = await readPassword(); + if (!password) throw new Error("没有密码：配 PASSWORD 或写 .local/e2e-account-password.txt"); + const login = await call("/auth/login/password", { + method: "POST", + body: { phone: PHONE, password }, + }); + if (login.code !== 0) throw new Error(`登录失败：${login.message}`); + const token = login.data.tokens.access_token; + const spaces = (await call("/workspaces", { token })).data?.items ?? []; + if (!spaces.length) throw new Error("这个账号没有工作区"); + const ws = spaces[0].id; + + // 自己造数据：标签名带时间戳，避免和库里已有的同名标签混在一起数不准 + const stamp = Date.now().toString().slice(-6); + const tag = `核对${stamp}`; + const created = await call(`/workspaces/${ws}/knowledge`, { + method: "POST", + token, + body: { title: `临时核对：标签与删除 ${stamp}`, body_md: "这条是 e2e 建的，跑完会删掉。", tag_names: [tag] }, + }); + if (created.code !== 0) throw new Error(`建知识失败：${created.message}`); + const doc = created.data; + await call(`/workspaces/${ws}/knowledge/${doc.id}/status`, { + method: "POST", + token, + body: { action: "publish", version: doc.version }, + }); + + const browser = await chromium.launch({ executablePath: CHROME }); + const context = await browser.newContext({ viewport: { width: 1440, height: 960 } }); + await context.addInitScript( + ([access, refresh]) => { + localStorage.setItem("dadao.access", access); + localStorage.setItem("dadao.refresh", refresh); + }, + [token, login.data.tokens.refresh_token], + ); + const page = await context.newPage(); + const errors = []; + page.on("pageerror", (e) => errors.push(String(e))); + page.on("dialog", (d) => void d.accept()); + + await page.goto(`${WEB}/w/${ws}/knowledge`, { waitUntil: "networkidle" }); + + // ① 分类下面有「标签」块，且自己建的那个标签在里面 + const tagBox = page.locator("section", { hasText: "标签" }).last(); + const tagBtn = page.getByRole("button", { name: new RegExp(`#${tag}`) }).first(); + await tagBtn.waitFor({ timeout: 10000 }); + check("分类下面是标签清单", (await tagBox.count()) > 0); + check("刚建的标签出现在侧栏（带条数）", (await tagBtn.count()) > 0); + + // ② 点标签 → 列表只剩这一类 + await tagBtn.click(); + await page.waitForTimeout(800); + const titles = await page.locator("li p.truncate").allInnerTexts(); + check( + "按标签筛出的正好是自己那条", + titles.filter((t) => t.includes(stamp)).length === 1 && titles.length >= 1, + `titles=${titles.length}`, + ); + check("标签被选中", (await page.getByRole("button", { name: new RegExp(`#${tag}`) }).first().getAttribute("aria-pressed")) === "true"); + + // ③ 再点一次取消筛选 + await tagBtn.click(); + await page.waitForTimeout(600); + check( + "再点一次取消标签筛选", + (await page.getByRole("button", { name: new RegExp(`#${tag}`) }).first().getAttribute("aria-pressed")) === "false", + ); + + // ④ 卡片上的删除：确认框点"确定"，这条从列表消失 + const row = page.locator("li").filter({ hasText: stamp }).first(); + await row.hover(); + await page.getByRole("button", { name: `删除 临时核对：标签与删除 ${stamp}` }).first().click(); + await page.waitForTimeout(1200); + check("列表里已经看不到这条", (await page.locator("li").filter({ hasText: stamp }).count()) === 0); + + const afterDelete = await call(`/workspaces/${ws}/knowledge/${doc.id}`, { token }); + check("接口回查：这条确实没了", afterDelete.code !== 0, `code=${afterDelete.code}`); + + // ⑤ 阅读页也有删除入口（换一条已存在的打开看看） + const page2 = await call(`/workspaces/${ws}/knowledge?limit=1`, { token }); + const anyDoc = page2.data?.items?.[0]; + if (anyDoc) { + await page.goto(`${WEB}/w/${ws}/knowledge/${anyDoc.id}`, { waitUntil: "networkidle" }); + check("阅读页有「删除」按钮", (await page.getByRole("button", { name: "删除" }).count()) > 0); + } + + check("控制台零报错", errors.length === 0, errors.slice(0, 2).join(" | ")); + + await page.goto(`${WEB}/w/${ws}/knowledge`, { waitUntil: "networkidle" }); + await page.screenshot({ path: "e2e/artifacts/knowledge-tags.png" }); + await browser.close(); + + // 收尾：万一界面那步没删掉，这里补一刀，别给下次跑留垃圾 + await call(`/workspaces/${ws}/knowledge/${doc.id}?confirm=true`, { method: "DELETE", token }); + + console.log(`\n${fail === 0 ? "ALL PASS" : "FAILED"} — pass ${pass}, fail ${fail}`); + if (fail > 0) process.exitCode = 1; +} + +main().catch((err) => { + console.error("FAIL 脚本异常 —", err); + process.exitCode = 1; +}); diff --git a/web/src/lib/types.ts b/web/src/lib/types.ts index bf0b4d3..07e2741 100644 --- a/web/src/lib/types.ts +++ b/web/src/lib/types.ts @@ -191,6 +191,12 @@ export interface KnowledgeCategoryNode extends KnowledgeCategory { doc_count: number; } +/** 侧栏那串标签：`count` 是"有多少条知识用了它"，不是引用次数。 */ +export interface KnowledgeTagCount { + name: string; + count: number; +} + export interface Knowledge extends Item { knowledge: KnowledgeInfo; category: KnowledgeCategory | null; diff --git a/web/src/pages/Knowledge.tsx b/web/src/pages/Knowledge.tsx index ac9af81..c62ab52 100644 --- a/web/src/pages/Knowledge.tsx +++ b/web/src/pages/Knowledge.tsx @@ -28,6 +28,7 @@ import type { Attachment, Knowledge, KnowledgeCategoryNode, + KnowledgeTagCount, SearchHit, } from "../lib/types"; @@ -76,6 +77,9 @@ export function KnowledgePage() { /** null = 不按分类筛（全部）；"none" = 未分类；其它是分类 id（含子分类） */ const [catFilter, setCatFilter] = useState<string | null>(null); const [categories, setCategories] = useState<KnowledgeCategoryNode[]>([]); + /** 分类下面的标签清单：知识库就靠分类 + 标签两条轴找东西 */ + const [tags, setTags] = useState<KnowledgeTagCount[]>([]); + const [tagFilter, setTagFilter] = useState<string | null>(null); const [catBusy, setCatBusy] = useState(false); const reqId = useRef(0); @@ -90,10 +94,28 @@ export function KnowledgePage() { } }, [wsId]); + const loadTags = useCallback(async () => { + try { + const qs = new URLSearchParams(); + for (const s of TABS.find((t) => t.key === tab)!.statuses) qs.append("status", s); + const page = await api.get<{ items: KnowledgeTagCount[] }>( + `/workspaces/${wsId}/knowledge/tags?${qs.toString()}`, + ); + setTags(page.items); + } catch { + // 标签只是辅助导航，取不到不该挡住列表 + setTags([]); + } + }, [wsId, tab]); + useEffect(() => { void loadCategories(); }, [loadCategories]); + useEffect(() => { + void loadTags(); + }, [loadTags]); + async function createCategory(name: string, parentId: string | null, color: string | null = null) { setCatBusy(true); setError(null); @@ -186,6 +208,7 @@ export function KnowledgePage() { if (q.trim()) qs.set("q", q.trim()); if (catFilter === "none") qs.set("uncategorized", "true"); else if (catFilter) qs.set("category_id", catFilter); + if (tagFilter) qs.set("tag", tagFilter); if (nextCursor) qs.set("cursor", nextCursor); const page = await api.get<{ items: Knowledge[]; @@ -202,7 +225,7 @@ export function KnowledgePage() { if (mine === reqId.current) setLoading(false); } }, - [wsId, tab, q, catFilter], + [wsId, tab, q, catFilter, tagFilter], ); useEffect(() => { @@ -239,11 +262,30 @@ export function KnowledgePage() { setComposing(false); setTab("draft"); setCatFilter(null); + setTagFilter(null); openDoc(doc.id); void load(null); void loadCategories(); } + /** + * 删除是软删（后端 `deleted_at`），但界面上一旦删了就当它不在了： + * 所以列表重拉、标签计数重算，别留一条需要刷新才消失的幽灵。 + */ + async function removeDoc(doc: Knowledge) { + if (!window.confirm(`删除「${doc.title}」？删了不会回来，检索里也会一起消失。`)) return; + setError(null); + try { + await api.del(`/workspaces/${wsId}/knowledge/${doc.id}?confirm=true`); + setDocs((prev) => prev.filter((d) => d.id !== doc.id)); + void load(null); + void loadCategories(); + void loadTags(); + } catch (err) { + setError(err instanceof ApiError ? err.message : "删除失败"); + } + } + const pendingCount = docs.filter((d) => d.knowledge.status_hint !== "published").length; const activeCategory = categories.find((c) => c.id === catFilter) ?? null; @@ -291,6 +333,13 @@ export function KnowledgePage() { loading={loading} onSelect={(value) => { setCatFilter(value); + setTagFilter(null); + setHits(null); + }} + tags={tags} + tagSelected={tagFilter} + onSelectTag={(name) => { + setTagFilter((prev) => (prev === name ? null : name)); setHits(null); }} busy={catBusy} @@ -486,9 +535,20 @@ export function KnowledgePage() { </div> ) : null} </button> - <Button variant="ghost" size="sm" className="shrink-0" onClick={() => openDoc(d.id)}> - 打开 - </Button> + <div className="flex shrink-0 items-center gap-1"> + <Button variant="ghost" size="sm" onClick={() => openDoc(d.id)}> + 打开 + </Button> + {/* 删除只在悬停/聚焦时露出来：这一列动作不该比标题更抢眼 */} + <button + className="rounded-md p-1.5 text-ink-400 opacity-0 transition hover:bg-red-50 hover:text-red-600 focus:opacity-100 group-hover:opacity-100" + title="删除这条知识" + aria-label={`删除 ${d.title}`} + onClick={() => void removeDoc(d)} + > + <Trash size={14} /> + </button> + </div> </li> ))} </ul> @@ -548,12 +608,15 @@ function ColorSwatch({ ); } -/** 左侧分类树：一层一层缩进，点一下就把列表收窄到该分类（含子分类）。 */ +/** 左侧分类树：一层一层缩进，点一下就把列表收窄到该分类（含子分类）；分类下面是标签。 */ function CategorySidebar({ categories, selected, loading, onSelect, + tags, + tagSelected, + onSelectTag, busy, onCreate, onRename, @@ -564,6 +627,9 @@ function CategorySidebar({ selected: string | null; loading: boolean; onSelect: (value: string | null) => void; + tags: KnowledgeTagCount[]; + tagSelected: string | null; + onSelectTag: (name: string) => void; busy: boolean; onCreate: (name: string, parentId: string | null, color: string | null) => void; onRename: (cat: KnowledgeCategoryNode, name: string) => void; @@ -686,6 +752,42 @@ function CategorySidebar({ 还没有分类。先建一层「研发 / 部署」这样的目录，知识才有地方落脚。 </p> ) : null} + + {/* 标签是第二根轴：分类回答"这属于哪一块"，标签回答"它跟什么有关"。 + 按条数从多到少排，点一下切换筛选（再点一次取消）。 */} + {tags.length > 0 ? ( + <section className="mt-4 border-t border-line-200 pt-3"> + <div className="flex items-center justify-between px-1 pb-1"> + <span className="text-2xs font-medium uppercase tracking-wide text-ink-400">标签</span> + {tagSelected ? ( + <button className="link text-2xs" onClick={() => onSelectTag(tagSelected)}> + 清除 + </button> + ) : null} + </div> + <div className="flex flex-wrap gap-1 px-1"> + {tags.map((t) => { + const active = tagSelected === t.name; + return ( + <button + key={t.name} + aria-pressed={active} + onClick={() => onSelectTag(t.name)} + className={`inline-flex max-w-full items-center gap-1 rounded-full px-2 py-0.5 text-2xs transition-colors ${ + active + ? "bg-primary-600 text-white" + : "bg-primary-50 text-primary-700 hover:bg-primary-100" + }`} + title={`${t.count} 条知识用了这个标签`} + > + <span className="truncate">#{t.name}</span> + <span className={active ? "text-white/70" : "text-primary-400"}>{t.count}</span> + </button> + ); + })} + </div> + </section> + ) : null} </aside> ); } diff --git a/web/src/pages/KnowledgeDoc.tsx b/web/src/pages/KnowledgeDoc.tsx index 395af22..6f9526b 100644 --- a/web/src/pages/KnowledgeDoc.tsx +++ b/web/src/pages/KnowledgeDoc.tsx @@ -11,7 +11,7 @@ import { Link, useNavigate, useParams } from "react-router-dom"; import { AppShell } from "../components/AppShell"; import { AttachmentPanel } from "../components/AttachmentPanel"; import { Markdown } from "../components/Markdown"; -import { ChevronLeft, Link2, Pencil } from "../components/icons"; +import { ChevronLeft, Link2, Pencil, Trash } from "../components/icons"; import { Button, InlineAlert, SkeletonRows, StatusPill } from "../components/ui"; import { ApiError, api } from "../lib/client"; import { ACTION_LABEL, DOC_TYPE_LABEL, STATUS_LABEL, categoryColor } from "../lib/knowledge"; @@ -92,6 +92,22 @@ export function KnowledgeDocPage() { const status = doc?.knowledge.status_hint; const canEdit = status === "draft" || status === "in_review" || status === "published"; + + /** 删除：确认后回列表页。后端是软删，但界面按"没了"处理最不容易让人误会。 */ + async function remove() { + if (!doc) return; + if (!window.confirm(`删除「${doc.title}」？删了不会回来，检索里也会一起消失。`)) return; + setBusy(true); + setError(null); + try { + await api.del(`/workspaces/${wsId}/knowledge/${itemId}?confirm=true`); + navigate(`/w/${wsId}/knowledge`, { replace: true }); + } catch (err) { + setError(err instanceof ApiError ? err.message : "删除失败"); + } finally { + setBusy(false); + } + } const actions: { key: string; label: string; show: boolean }[] = [ { key: "submit_review", label: ACTION_LABEL.submit_review, show: status === "draft" }, { key: "withdraw", label: ACTION_LABEL.withdraw, show: status === "in_review" }, @@ -210,6 +226,11 @@ export function KnowledgeDocPage() { {a.label} </Button> ))} + {/* 删除放最后、用弱色 + 垃圾桶：它是不可逆的，不该和"发布"抢注意力 */} + <Button variant="ghost" disabled={busy} className="ml-auto text-ink-400 hover:text-red-600" onClick={() => void remove()}> + <Trash size={13} /> + 删除 + </Button> </div> <div className="mt-6">