diff --git a/apps/android/src/ui/link-helpers.ts b/apps/android/src/ui/link-helpers.ts new file mode 100644 index 0000000..847a3bf --- /dev/null +++ b/apps/android/src/ui/link-helpers.ts @@ -0,0 +1,600 @@ +import { Linking } from 'react-native' +import MarkdownIt from 'markdown-it' +import { installMath } from './markdown-math' +import { + markdownFilePath, + parseDshFileLink, + parseWorkspaceTarget, + classifyDshLinkPath, + classifyDshWebBrand, + type DshLinkIconKind, + type DshParsedFileLink, +} from './file-glyph' + +export { + markdownFilePath, + parseDshFileLink, + parseWorkspaceTarget, + classifyDshLinkPath, + classifyDshWebBrand, + type DshLinkIconKind, + type DshParsedFileLink, +} + +export type LinkKind = 'citation' | 'external' | 'workspace' | 'session' | 'standard' + +export type BlockReason = 'unsafe_protocol' | 'malformed_uri' + +export interface MarkdownLinkInfo { + kind: LinkKind + rawHref: string + safeHref: string | null + displayText: string + isBlocked: boolean + blockReason?: BlockReason + iconKind?: DshLinkIconKind + filePath?: string + fileLine?: number + citationIndex?: string + domain?: string + sessionId?: string +} + +export interface SanitizeResult { + safeHref: string | null + isBlocked: boolean + reason?: BlockReason + protocol?: string +} + +const ALLOWED_EXTERNAL_PROTOCOLS = new Set(['http:', 'https:', 'mailto:']) +const ALLOWED_ALL_PROTOCOLS = new Set(['http:', 'https:', 'mailto:', 'dsh-session:', 'workspace:']) + +const BLOCKED_PROTOCOLS = new Set([ + 'javascript:', + 'data:', + 'vbscript:', + 'file:', + 'blob:', + 'intent:', + 'content:', + 'about:', + 'filesystem:', + 'jar:', + 'chrome:', + 'chrome-extension:', +]) + +const SCHEME_REGEX = /^([a-z][a-z0-9+.-]*):/i +const CONTROL_CHARS_REGEX = /[\u0000-\u001F\u007F-\u009F\u200B-\u200D\uFEFF]/g +const URL_SCHEME_PREFIX_REGEX = /^[a-z][a-z0-9+.-]*:\/\//i +const WWW_PREFIX_REGEX = /^www\./i +const BARE_URL_TEXT_REGEX = /^(?:https?:\/\/|www\.)\S+$/i + +// 严格且安全的成对角标与锚点正则(无灾难性回溯) +const CITATION_BRACKET_REGEX = /^\[\^?(\d+)\]$/ +const CITATION_CARET_REGEX = /^\^(\d+)$/ +const CITATION_NAMED_REGEX = /^\[(?:来源|参考|cite|ref)\s*(?:[::]\s*)?(\d+)\]$/iu +const CITATION_ANCHOR_REGEX = /^#(?:cite|ref|fn|footnote)[-_](\d+)$/i + +function stripWrappingBrackets(href: string): string { + const trimmed = href.trim() + if (trimmed.startsWith('<') && trimmed.endsWith('>')) { + return trimmed.slice(1, -1).trim() + } + return trimmed +} + +export function getProtocol(href: string): string { + const match = SCHEME_REGEX.exec((href ?? '').trim()) + return match ? `${match[1]!.toLowerCase()}:` : '' +} + +/** + * 严格协议白名单校验与安全清洗: + * - 协议相对链接(以 // 开头)在处理相对路径前统一提升为 https:// 并进行完整 WHATWG 规范化 + * - 拦截危险伪协议(javascript:, data:, vbscript:, file:, blob: 等)及控制字符绕过 + * - 隐式将 www. 前缀提升为 https:// + * - 严格校验内部协议(dsh-session, workspace)载荷有效性 + */ +export function sanitizeAndValidateHref(rawHref: string): SanitizeResult { + if (typeof rawHref !== 'string') { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri' } + } + + const cleaned = stripWrappingBrackets(rawHref).replace(CONTROL_CHARS_REGEX, '').trim() + if (!cleaned) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri' } + } + + // 1. 协议相对 URL 必须在常规斜杠相对路径之前优先识别并补齐协议(必须包含有效公网域名点号,拒绝 UNC 本地机器名 //server/share) + let candidate = cleaned + if (candidate.startsWith('//')) { + if (candidate.startsWith('///') || candidate.startsWith('//\\') || !candidate.slice(2).split(/[/?#]/u)[0]?.includes('.')) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri' } + } + candidate = `https:${candidate}` + } else if (WWW_PREFIX_REGEX.test(candidate)) { + candidate = `https://${candidate}` + } + + const schemeMatch = SCHEME_REGEX.exec(candidate) + + // 无协议头:相对路径、页面锚点、Windows 盘符路径(D:/...) + if (!schemeMatch || (schemeMatch[1]!.length === 1 && (candidate.charAt(2) === '/' || candidate.charAt(2) === '\\'))) { + // 页面内纯锚点直接放行 + if (candidate.startsWith('#')) { + return { safeHref: candidate, isBlocked: false } + } + // 严格阻断路径穿越 (../) + if (/(?:^|[/\\])\.\.(?:[/\\]|$)/.test(candidate)) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri' } + } + // Windows 盘符绝对路径与常规相对路径:必须通过本地解析器验证 + const parsed = parseDshFileLink(candidate) + if (!parsed) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri' } + } + return { safeHref: candidate, isBlocked: false } + } + + const protocol = `${schemeMatch[1]!.toLowerCase()}:` + + if (BLOCKED_PROTOCOLS.has(protocol) || !ALLOWED_ALL_PROTOCOLS.has(protocol)) { + return { safeHref: null, isBlocked: true, reason: 'unsafe_protocol', protocol } + } + + // 2. 内部会话协议:必须能成功解码出有效 sessionId + if (protocol === 'dsh-session:') { + const sessionId = decodeSessionMentionUri(candidate) + if (!sessionId || sessionId.length === 0) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri', protocol } + } + return { safeHref: candidate, isBlocked: false, protocol } + } + + // 3. 内部工作区协议:必须是合法相对路径,拒绝 UNC、nested scheme、盘符及路径穿越 + if (protocol === 'workspace:') { + const raw = candidate.slice('workspace:'.length).trim() + const parsed = parseWorkspaceTarget(raw) + if (!parsed) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri', protocol } + } + return { safeHref: candidate, isBlocked: false, protocol } + } + + // 4. 网络协议:要求标准 // 双斜杠前缀并严格校验 + if (protocol === 'http:' || protocol === 'https:') { + if (!/^https?:\/\//i.test(candidate)) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri', protocol } + } + } + + try { + const parsed = new URL(candidate) + const parsedProtocol = parsed.protocol.toLowerCase() + + if (!ALLOWED_EXTERNAL_PROTOCOLS.has(parsedProtocol)) { + return { safeHref: null, isBlocked: true, reason: 'unsafe_protocol', protocol: parsedProtocol } + } + + if ((parsedProtocol === 'http:' || parsedProtocol === 'https:') && !parsed.hostname) { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri', protocol: parsedProtocol } + } + + return { safeHref: candidate, isBlocked: false, protocol: parsedProtocol } + } catch { + return { safeHref: null, isBlocked: true, reason: 'malformed_uri', protocol } + } +} + +/** + * 判定链接是否为引用来源角标(例如 [1], [^1], ^1, [来源 1], #fn-1): + * - 严格成对匹配,单字母或裸数字不误判为角标,除非通过明确的 #cite-1 锚点确认; + * - 前置长度保护(<= 32 字符),彻底消除回溯开销。 + */ +export function isCitationLink( + text: string, + href?: string, +): { isCitation: boolean; index?: string } { + const trimmedText = typeof text === 'string' ? text.trim() : '' + + if (trimmedText && trimmedText.length <= 32) { + const bMatch = CITATION_BRACKET_REGEX.exec(trimmedText) + if (bMatch && bMatch[1]) return { isCitation: true, index: bMatch[1] } + + const cMatch = CITATION_CARET_REGEX.exec(trimmedText) + if (cMatch && cMatch[1]) return { isCitation: true, index: cMatch[1] } + + const nMatch = CITATION_NAMED_REGEX.exec(trimmedText) + if (nMatch && nMatch[1]) return { isCitation: true, index: nMatch[1] } + } + + const trimmedHref = typeof href === 'string' ? href.trim() : '' + if (trimmedHref) { + const anchorMatch = CITATION_ANCHOR_REGEX.exec(trimmedHref) + if (anchorMatch && anchorMatch[1]) { + return { isCitation: true, index: anchorMatch[1] } + } + } + + return { isCitation: false } +} + +/** + * 长 URL 可读截断与格式化: + * 去除冗余 scheme 和 www,保留域名主体与末尾路径,中间折叠为省略号,防止移动端窄屏排版撑破。 + */ +export function formatDisplayUrl(rawUrl: string, maxLen = 42): string { + const trimmed = typeof rawUrl === 'string' ? rawUrl.trim() : '' + if (!trimmed) return '' + + const limit = Number.isFinite(maxLen) ? Math.max(8, Math.floor(maxLen)) : 42 + + let display = trimmed + .replace(URL_SCHEME_PREFIX_REGEX, '') + .replace(WWW_PREFIX_REGEX, '') + .replace(/\/+$/u, '') + + if (!display) display = trimmed + + if (display.length <= limit) return display + + const tailLength = Math.min(14, Math.floor(limit / 3)) + const headLength = Math.max(1, limit - tailLength - 1) + return `${display.slice(0, headLength)}…${display.slice(-tailLength)}` +} + +function extractDomain(href: string): string | undefined { + try { + const parsed = new URL(href) + if (!parsed.hostname) return undefined + return parsed.hostname.replace(WWW_PREFIX_REGEX, '') + } catch { + const match = /^[a-z][a-z0-9+.-]*:\/\/([^/?#]+)/iu.exec(href) + return match && match[1] ? match[1].replace(WWW_PREFIX_REGEX, '') : undefined + } +} + +function isBareUrlText(text: string, href: string): boolean { + const trimmedText = (text ?? '').trim() + if (!trimmedText) return false + const trimmedHref = (href ?? '').trim() + if (trimmedText === trimmedHref) return true + return BARE_URL_TEXT_REGEX.test(trimmedText) +} + +/** + * 链接多态分类器(1:1 对齐官方 DeepSeek Harness 桌面端逻辑): + * 优先级: + * 1. 外部 Web 链接(优先于无 scheme 的本地检查,避免 www. 域名被误判为本地文件) + * 2. 会话内部引用(dsh-session:) + * 3. 工作区内部协议(workspace:) + * 4. 引用来源角标(isCitationLink:[^1], [1], [来源 1]) + * 5. 本地工作区文件(parseDshFileLink:支持相对路径、盘符绝对路径、尖括号包路径、行号 #L...) + * 注:非锚点的无协议相对路径若被 parseDshFileLink 拒绝(如路径穿越、带查询参数、UNC 等),一律严格阻断(isBlocked: true),绝不放行! + * 6. 页面内部锚点(#...)与普通语义链接 + */ +export function classifyMarkdownLink(href: string, text: string): MarkdownLinkInfo { + const rawHref = typeof href === 'string' ? href : '' + const rawText = typeof text === 'string' ? text : '' + const trimmedHref = rawHref.trim() + const trimmedText = rawText.trim() + const plainLabel = trimmedText || trimmedHref + + if (!trimmedHref) { + const emptyCitation = isCitationLink(trimmedText) + if (emptyCitation.isCitation && emptyCitation.index) { + return { + kind: 'citation', + rawHref, + safeHref: null, + displayText: emptyCitation.index, + isBlocked: false, + citationIndex: emptyCitation.index, + } + } + return { + kind: 'standard', + rawHref, + safeHref: null, + displayText: plainLabel, + isBlocked: false, + } + } + + // 1. 外部 Web 链接判定(包括裸 www. 链接) + if (WWW_PREFIX_REGEX.test(trimmedHref) || /^https?:\/\//i.test(trimmedHref) || /^\/\//.test(trimmedHref)) { + const { safeHref, isBlocked, reason } = sanitizeAndValidateHref(trimmedHref) + if (isBlocked || !safeHref) { + return { + kind: 'standard', + rawHref, + safeHref: null, + displayText: plainLabel, + isBlocked: true, + blockReason: reason, + } + } + const domain = extractDomain(safeHref) + const iconKind = classifyDshWebBrand(safeHref) ?? 'globe' + const displayText = isBareUrlText(rawText, rawHref) + ? formatDisplayUrl(safeHref) + : plainLabel + + return { + kind: 'external', + rawHref, + safeHref, + displayText, + isBlocked: false, + domain, + iconKind, + } + } + + // 2. 内置会话引用协议 + if (trimmedHref.toLowerCase().startsWith('dsh-session:')) { + const { safeHref, isBlocked, reason } = sanitizeAndValidateHref(trimmedHref) + if (isBlocked || !safeHref) { + return { + kind: 'standard', + rawHref, + safeHref: null, + displayText: plainLabel, + isBlocked: true, + blockReason: reason, + } + } + const sessionId = decodeSessionMentionUri(safeHref) + return { + kind: 'session', + rawHref, + safeHref, + displayText: plainLabel || sessionId || '会话', + isBlocked: false, + sessionId, + iconKind: 'code', + } + } + + // 3. 内部 workspace 协议(剥离协议后进行严格相对路径本地文件校验) + if (trimmedHref.toLowerCase().startsWith('workspace:')) { + const rawTarget = trimmedHref.slice('workspace:'.length).trim() + const fileParsed = parseWorkspaceTarget(rawTarget) + if (!fileParsed) { + return { + kind: 'standard', + rawHref, + safeHref: null, + displayText: plainLabel, + isBlocked: true, + blockReason: 'malformed_uri', + } + } + const iconKind = classifyDshLinkPath(fileParsed.path) + return { + kind: 'workspace', + rawHref, + safeHref: trimmedHref, + displayText: plainLabel, + isBlocked: false, + iconKind, + filePath: fileParsed.path, + fileLine: fileParsed.line, + } + } + + // 4. 本地工作区文件(1:1 官方 DSH ow(e) 规则,优先于纯文本角标匹配) + const fileParsed = parseDshFileLink(trimmedHref) + if (fileParsed) { + const iconKind = classifyDshLinkPath(fileParsed.path) + return { + kind: 'workspace', + rawHref, + safeHref: trimmedHref, + displayText: plainLabel, + isBlocked: false, + iconKind, + filePath: fileParsed.path, + fileLine: fileParsed.line, + } + } + + // 5. 引用来源角标(仅允许目标为空或合法页面内锚点,非锚点的非法目标绝不能借角标文本逃逸阻断) + if (!trimmedHref || trimmedHref.startsWith('#')) { + const citation = isCitationLink(trimmedText, trimmedHref) + if (citation.isCitation && citation.index) { + return { + kind: 'citation', + rawHref, + safeHref: trimmedHref.startsWith('#') ? trimmedHref : null, + displayText: citation.index, + isBlocked: false, + citationIndex: citation.index, + } + } + } + + // 6. 纯页面内锚点 (#section) + if (trimmedHref.startsWith('#')) { + return { + kind: 'standard', + rawHref, + safeHref: trimmedHref, + displayText: plainLabel, + isBlocked: false, + } + } + + // 7. mailto 协议 + if (/^mailto:/i.test(trimmedHref)) { + const { safeHref, isBlocked, reason } = sanitizeAndValidateHref(trimmedHref) + return { + kind: 'standard', + rawHref, + safeHref: isBlocked ? null : safeHref, + displayText: plainLabel, + isBlocked, + blockReason: reason, + iconKind: 'globe', + } + } + + // 8. 其他所有被 parseDshFileLink 拒绝的相对路径(如 ../secret.txt, src/../a.js, query 等)或危险伪协议一律阻断! + return { + kind: 'standard', + rawHref, + safeHref: null, + displayText: plainLabel, + isBlocked: true, + blockReason: 'malformed_uri', + } +} + +/** + * 打开受信任的安全外部链接(仅放行 http:, https:, mailto:) + * 返回 true 表示已尝试向系统分派打开请求,非外部安全协议一律阻断。 + */ +export function openSafeExternalUrl(rawUrl: string): boolean { + const { safeHref, isBlocked } = sanitizeAndValidateHref(rawUrl) + if (isBlocked || !safeHref) return false + + const protocol = getProtocol(safeHref) + if (!ALLOWED_EXTERNAL_PROTOCOLS.has(protocol)) { + return false + } + + try { + const pending = Linking.openURL(safeHref) + if (pending && typeof (pending as Promise).catch === 'function') { + void (pending as Promise).catch(() => undefined) + } + return true + } catch { + return false + } +} + +const SESSION_MENTION_SCHEME = 'dsh-session:' +const BASE64_ALPHABET = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/' +const BASE64_VALUES: ReadonlyMap = new Map( + Array.from(BASE64_ALPHABET, (char, index) => [char, index] as const), +) + +export function decodeBase64ToBytes(input: string): Uint8Array | undefined { + if (typeof input !== 'string') return undefined + // Strict Base64 padding check: '=' must only appear at the end, at most 2, and length % 4 == 0 if padded + const padIndex = input.indexOf('=') + if (padIndex !== -1) { + const padding = input.slice(padIndex) + if (!/^={1,2}$/.test(padding) || input.length % 4 !== 0) { + return undefined + } + } + + const unpadded = input.replace(/=+$/u, '') + if (unpadded.length === 0 || unpadded.length % 4 === 1) return undefined + const bytes: number[] = [] + let buffer = 0 + let bits = 0 + for (let index = 0; index < unpadded.length; index += 1) { + const value = BASE64_VALUES.get(unpadded.charAt(index)) + if (value === undefined) return undefined + buffer = ((buffer << 6) | value) & 0xffffff + bits += 6 + if (bits >= 8) { + bits -= 8 + bytes.push((buffer >> bits) & 0xff) + } + } + + // Canonical Base64: unused trailing bits must be strictly zero + if (bits > 0) { + const unusedMask = (1 << bits) - 1 + if ((buffer & unusedMask) !== 0) return undefined + } + + return Uint8Array.from(bytes) +} + +export function decodeUtf8Bytes(bytes: Uint8Array): string | undefined { + if (typeof TextDecoder !== 'undefined') { + try { + return new TextDecoder('utf-8', { fatal: true }).decode(bytes) + } catch { + return undefined + } + } + + let result = '' + let index = 0 + while (index < bytes.length) { + const byte = bytes[index]! + let codePoint: number + let trailing: number + let minCodePoint: number + + if (byte < 0x80) { + codePoint = byte + trailing = 0 + minCodePoint = 0 + } else if (byte >= 0xc2 && byte <= 0xdf) { + codePoint = byte & 0x1f + trailing = 1 + minCodePoint = 0x80 + } else if (byte >= 0xe0 && byte <= 0xef) { + codePoint = byte & 0x0f + trailing = 2 + minCodePoint = 0x800 + } else if (byte >= 0xf0 && byte <= 0xf4) { + codePoint = byte & 0x07 + trailing = 3 + minCodePoint = 0x10000 + } else { + return undefined + } + + if (index + trailing >= bytes.length) return undefined + + for (let offset = 1; offset <= trailing; offset += 1) { + const continuation = bytes[index + offset]! + if ((continuation & 0xc0) !== 0x80) return undefined + codePoint = (codePoint << 6) | (continuation & 0x3f) + } + + // Overlong or surrogate or above maximum Unicode + if (codePoint < minCodePoint || (codePoint >= 0xd800 && codePoint <= 0xdfff) || codePoint > 0x10ffff) { + return undefined + } + + result += String.fromCodePoint(codePoint) + index += trailing + 1 + } + return result +} + +export function decodeSessionMentionUri(href: string): string | undefined { + if (typeof href !== 'string') return undefined + const value = href.trim() + if (value.length <= SESSION_MENTION_SCHEME.length) return undefined + if (!value.toLowerCase().startsWith(SESSION_MENTION_SCHEME)) return undefined + const encoded = value.slice(SESSION_MENTION_SCHEME.length).trim() + const normalized = encoded.replace(/-/gu, '+').replace(/_/gu, '/') + const bytes = decodeBase64ToBytes(normalized) + if (bytes === undefined) return undefined + const json = decodeUtf8Bytes(bytes) + if (json === undefined) return undefined + try { + const parsed: unknown = JSON.parse(json) + return typeof parsed === 'string' && parsed.length > 0 ? parsed : undefined + } catch { + return undefined + } +} + +export function createMarkdownParser(): MarkdownIt { + const parser = new MarkdownIt({ html: false, linkify: true, breaks: true }) + installMath(parser) + return parser +}