MarKing

zh-CN

稳定性修复专版：修复 v1.6.1 起部分用户遇到的「打开笔记库即闪退」问题，并内置崩溃自动诊断——此后若再遇崩溃，应用会自动生成报告，复制粘贴即可反馈。建议所有 1.6.1 / 1.6.2 用户升级。

🐛 修复

• 打开笔记库即闪退 — 修复部分笔记库在启动或切换库时立即退出的问题。当文档中存在「包含链接且超过 200 字符的中文 / 日文 / emoji 行」时，建立双链索引的过程中会触发底层错误并终止整个应用；现已修复，此类库可正常打开与索引。
• 索引损坏导致反复闪退 — 个别库的索引缓存损坏或上次初始化被中断时，此前每次启动都会再次崩溃。现在应用会自动隔离可疑索引并重建（索引为派生缓存，重建不损失任何笔记内容），解除「坏库 → 反复崩」的死循环。

♻️ 优化

• 崩溃自动报告 — 若应用再发生底层崩溃，会自动生成诊断报告并用记事本弹出，复制内容即可反馈，无需自己查找日志文件。
• 异常日志覆盖补全 — 界面层的未捕获错误现在也会写入应用日志；启动与扫描过程新增关键节点记录，便于定位问题区间。

en-US

A stability-focused release: fixes the "instant crash when opening a vault" issue affecting some users since v1.6.1, and ships built-in crash self-diagnostics — should a low-level crash ever happen again, a ready-to-share report pops up automatically. Recommended update for all 1.6.1 / 1.6.2 users.

🐛 Fixes

• Vault Crashes on Open — Fixed an issue where certain vaults quit instantly during startup or vault switching. When a document contained a link on a CJK / emoji line longer than 200 characters, building the backlink index hit a low-level error that terminated the whole app. Such vaults now open and index normally.
• Corrupt Index Causing Repeat Crashes — If a vault's index cache was corrupted or a previous initialization was interrupted, every launch used to crash again. The app now quarantines the suspicious index and rebuilds it automatically — the index is a derived cache, so no notes are ever lost — breaking the "bad vault → crash loop".

♻️ Improvements

• Automatic Crash Reports — If a low-level crash ever happens again, a diagnostic report is generated and opened in your text editor automatically; copy its contents to report the issue, no log-hunting required.
• Broader Error Logging — Uncaught interface errors are now written to the application log as well, and startup / scan steps leave milestone entries that pinpoint where a problem occurred.

诊断预览版 —— 仅用于排查 1.6.1/1.6.2 闪退问题

此版本仅面向反馈过"安装后闪退"的用户，内置崩溃诊断能力，不是功能更新。

如果是来验证闪退修复的，请按以下步骤：

1. 安装后重复一次之前会闪退的操作（比如打开原来那个库）
2. 崩溃时会自动弹出一个记事本，把里面的文字全选复制，脱敏后粘贴到本 release 下方的 issue/评论区即可
3. 如果崩溃时没弹出记事本：Win+R 输入 %APPDATA%\com.marking.app\logs 打开日志目录，把 marking_rCURRENT.log 的最后 30 行提供到 issue 中；panic.log 若存在请一并提供。

注意：本版本为诊断构建，稳定性与正式版相同，但请不要作为日常版本长期使用。

zh-CN

AI 能力全面升级：支持 2026 年新模型并自动识别其能力，推理模型的「思考过程」现在可以实时展开观看；新增 Groq、MiniMax、OpenCode Go 三家供应商与自定义请求头；行内 AI 预览体验大幅打磨，另修复 Mermaid 导出与 Linux / macOS 启动问题。

✨ 新增功能

• 新模型即插即用 — 自动识别 GPT-5/6、Claude 4/5、Gemini 4、DeepSeek V4/R1、Qwen3.x、Kimi、GLM-5、Grok 4、MiniMax M2 等新一代模型的上下文长度、视觉输入与推理能力，添加后无需手工配置。
• 思考过程看得见 — DeepSeek R1、o 系等推理模型在续写、大纲、润色时会先展示「思考中」的推理过程，可折叠展开；思考内容不会混进正文。
• 更多供应商 — 内置 Groq、MiniMax、OpenCode Go 一键接入；每家供应商可单独配置自定义请求头，满足各类 API 网关的接入要求，密钥类信息加密存储。
• 行内 AI 预览升级 — 生成结果卡片在空间不足时自动翻到上方、滚动时贴边不消失；选中的原文以主题色高亮标记；中断生成时保留已写出的部分。
• 代码块换行控制 — 全局默认换行策略之外，单个代码块可通过工具栏临时切换换行 / 不换行，不改动文档原文。

♻️ 优化

• 推理模型更聪明 — 自动移除目标模型不支持的参数、为思考预留输出空间，避免「思考完没力气写正文」的空回复；验证模型连通性时自动用低推理强度，响应更快。
• 大文档防超限 — 内容超出模型上下文时自动压缩输出预算，实在放不下会给出明确提示，而不是返回看不懂的报错。
• 长思考不超时 — 推理模型思考时间较长，流式等待上限已相应放宽。

🐛 修复

• 推理模型偶发空回复 — 修复思考过程占满输出配额导致正文为空的问题。
• 大纲生成长时间无响应 — 推理模型生成大纲时界面长时间空白，现已与续写一致实时展示进度。
• Mermaid 导出问题 — 导出 PNG 文字偏移、超宽图被页面宽度裁断等问题修复，预览与导出渲染已统一。
• Linux 无法启动 — 修复部分 Linux 发行版缺少系统托盘组件导致 AppImage 启动即崩溃的问题，现在自动降级为无托盘运行。
• macOS 全屏关闭残留黑屏 — 全屏状态下关闭窗口不再留下黑屏桌面。
• 启动偶发卡住 — 修复托盘初始化偶发阻塞拖慢整体启动的问题。

en-US

A smarter AI core: 2026-era models are recognized automatically, and reasoning models now stream their thinking process in a collapsible view; Groq, MiniMax, and OpenCode Go join the built-in provider list alongside custom request headers; the inline-AI preview gets a thorough polish — plus fixes for Mermaid export and Linux / macOS startup issues.

✨ Features

• New Models, Plug and Play — GPT-5/6, Claude 4/5, Gemini 4, DeepSeek V4/R1, Qwen3.x, Kimi, GLM-5, Grok 4, MiniMax M2 and other next-gen models are recognized automatically — context length, vision input, and reasoning capability included. No manual configuration needed.
• Watch the Thinking — Reasoning models like DeepSeek R1 and the o-series now show a collapsible "thinking" stream while composing, outlining, or refining; reasoning never leaks into your document.
• More Providers — Groq, MiniMax, and OpenCode Go are built in; each provider supports custom request headers for API gateways, with credential-grade encrypted storage.
• Inline-AI Preview Upgraded — The result card flips above the cursor when space below runs out, pins to the editor edge instead of vanishing on scroll, marks the source range with a theme-aware highlight, and keeps partial output when you cancel mid-stream.
• Per-Block Line Wrap — Beyond the global default, each code block gets a toolbar toggle for wrap / no-wrap — without touching the document source.

♻️ Improvements

• Smarter Reasoning Requests — Unsupported sampling parameters are stripped automatically and output budget is reserved for thinking, eliminating "thought itself out of words" empty replies; connectivity checks use low reasoning effort for faster feedback.
• Large-Document Guard — When input exceeds the model's context, the output budget shrinks automatically; if it still won't fit you get a readable message instead of a cryptic error.
• No More Thinking Timeouts — The streaming idle limit is relaxed to accommodate long reasoning phases.

🐛 Fixes

• Empty Replies from Reasoning Models — Fixed thinking tokens exhausting the output budget and returning an empty response.
• Outline Generation Appearing Frozen — Reasoning models left the outline dialog blank for tens of seconds; progress now streams in real time like Compose.
• Mermaid Export — Fixed text drift in exported PNGs and extra-wide diagrams clipped by page width; preview and export now share one rendering pipeline.
• Linux Won't Launch — AppImage no longer crashes on distributions missing system-tray libraries; it degrades gracefully to a tray-less mode.
• macOS Black Space After Closing Fullscreen — Closing a fullscreen window no longer leaves a black desktop space behind.
• Occasional Startup Stall — Tray initialization can no longer blo

zh-CN

新增 WebDAV 同步通道：把笔记库同步到坚果云、NextCloud 等支持 WebDAV 的网盘，配合同步计划预览与定时自动同步，跨设备协作更稳；同期上线双链自动重命名同步与 Excalidraw 外部文件引用。

✨ 新增功能

• WebDAV 同步 — 除了 Git，现在还能把笔记库同步到坚果云、NextCloud 等支持 WebDAV 协议的网盘。填写服务器地址与账号即可开启，保存后自动同步；可以先查看本次的「同步计划」（将上传 / 下载 / 删除哪些文件）再确认执行，也可以按固定间隔（默认 5 分钟）自动同步。多设备之间的新增、修改、删除都会双向传播。
• 冲突保护 — 两端同时修改同一文件时，不会覆盖任何一侧：远端版本会另存为副本（文件名 (conflicted copy 日期).ext），进入「待处理副本」让你选择保留哪个，内容不丢失。
• 弱网可靠 — 同步遇到网络中断、超时或服务端限流时会自动退避重试（最长 8 秒、最多 4 次）并遵循服务器的重试等待建议；认证失败等确定性错误不会空转重试。单条出错的条目不影响其他文件，重试不会产生重复文件。
• 同步活动日志与多设备概况 — 每次同步的上传 / 下载 / 冲突明细均可回看；「多设备同步状态」一览各端平台与最近同步时间。
• 公网明文 HTTP 拦截 — 默认拦截公网明文 http:// 同步地址防止凭据泄露；局域网、自托管与 Tailscale 环境自动豁免，确需公网明文时可显式放行。
• 自动重命名同步 — 重命名或移动文件 / 文件夹时，自动同步全库 [[双链]]（含别名、块引用、标题引用、嵌入与图片链接，以及目录前缀路径形态）；可先预览修改清单再执行，支持撤销，并可设置「自动 / 手动 / 预览」三档偏好，代码块内的示例不误改。
• Excalidraw 外部文件引用 — 画布可保存为独立 .excalidraw 文件，用 Obsidian 兼容的 ![[file.excalidraw]] 语法嵌入文档（与内嵌 base64 模式共存），适合大画布与跨文档复用同一画布。

♻️ 优化

• 同步通道选择 — 同步设置中可选择 Git 或 WebDAV 通道，两条通道各自独立配置，互不干扰。
• 文件列表摘要提速 — 列表第二行摘要改为派生缓存：扫描与保存时即时提取纯文本摘要（剥离标记与代码块），打开大库时列表加载更快。
• 多设备状态合并 — 分类、收藏、标签、模板随同步做实体级合并，两台设备同时修改不再丢一侧。

🐛 修复

• 反链与知识图谱覆盖不全 — 经同步到达、外部新建或从未打开过的文档此前不进入反向链接与关系图谱；现已纳入全库索引，外部重命名 / 移动也会同步更新链接指向。
• Mermaid 图表显示过小 — 小型图表此前按原始像素渲染显得太小，超宽图则被缩到不可读且无放大手段。现在小图自动放大至容器宽（上限 2x），超宽图按原始尺寸显示并支持横向滚动；双击图表可直接进入全屏，全屏打开自动适配视口大小。
• 中文摘要标记残留 — 修复中文相邻加粗（如「文字加粗文字」）时 ** 残留在列表摘要中的问题。
• 同步可靠性加固 — 远端目录缺失自愈、远端目录误删防护、并发同步串行化、macOS / Windows 文件名 Unicode 归一、冲突副本防撞名等多项加固。

en-US

WebDAV sync arrives: sync your vault to WebDAV services like Nutstore or NextCloud, with sync-plan preview and interval auto-sync — alongside auto link-renaming sync and Excalidraw external-file embeds.

✨ Features

• WebDAV Sync — Besides Git, you can now sync your vault to WebDAV-compatible services such as Nutstore and NextCloud. Enter the server address and account to enable it, and saving auto-syncs. Review the sync plan (which files will be uploaded / downloaded / deleted) before running it, or set a fixed interval (5 minutes by default) for automatic syncing. Adds, edits, and deletions propagate both ways across devices.
• Conflict Protection — When the same file is edited on both ends, neither side is overwritten: the remote version is saved as a copy (filename (conflicted copy YYYY-MM-DD).ext) and appears under "Pending Copies" for you to pick which to keep — nothing is lost.
• Reliable on Flaky Networks — Sync automatically retries with backoff on network drops, timeouts, or server-side rate-limiting (up to 8s, 4 attempts), honoring the server's suggested retry delay. Deterministic errors such as auth failures are not retried. A single failed item doesn't block the rest, and retries never create duplicates.
• Sync Activity Log & Multi-Device Overview — Review the upload / download / conflict details of every sync run; "Multi-Device Sync Status" shows each device's platform and last sync time at a glance.
• Plain-HTTP Interception on Public Networks — Plaintext http:// sync endpoints on public networks are blocked by default to prevent credential leaks; LAN / self-hosted / Tailscale environments are automatically exempted, with an explicit opt-out switch.
• Auto Link-Renaming Sync — Rename or move a file / folder and MarKing updates every [[wikilink]] across the vault automatically (aliases, block / heading refs, embeds, image links, and directory-prefix paths). Preview the change list before applying, undo any time, and pick auto / manual / preview preference — code samples are left untouched.
• Excalidraw External-File Embed — Save a canvas as a standalone .excalidraw file and embed it with the Obsidian-compatible ![[file.excalidraw]] syntax (coexists with the inline base64 mode) — ideal for large canvases and reusing the same canvas across documents.

♻️ Improvements

• Sync Channel Selection — Choose Git or WebDAV as your sync channel in sett