aegon-lv97 1.0.0

package/gen-ai/components/AIExport/parseMarkdownHtml.ts

@@ -0,0 +1,653 @@
+/**
+ * 接口 sectionContent → HTML
+ *
+ * //0918 tip 兼容顺序：Markdown → HTML → Element Plus
+ * +link 继续走原解析：Markdown [text](url) / HTML <a> / el-link → a.report-tiptap-link
+ * tip 转成原 citation 标记（点击弹出 report-chapter-review-popover）
+ */
+import { Marked } from 'marked'
+//0918 tip / REF：与 editor 共用 resolve，避免导出侧半套迁移
+import { REF_MARKER_RE, REF_ANY_MARKER_RE, resolveRefMarker } from '../editor/reviewCitationTip'
+//0918
+
+function escapeHtml(value: string) {
+  return value
+    .replace(/&/g, '&amp;')
+    .replace(/</g, '&lt;')
+    .replace(/>/g, '&gt;')
+    .replace(/"/g, '&quot;')
+}
+
+function escapeAttr(value: string) {
+  return escapeHtml(String(value).replace(/\s+/g, ' ').trim())
+}
+
+function splitTipLink(tip: string) {
+  const raw = String(tip || '').trim()
+  const md = raw.match(/\[([^\]]+)\]\((https?:[^)\s]+)(?:\s+"[^"]*")?\)/i)
+  if (md?.[1] != null && md[2] != null) {
+    return {
+      text: raw.replace(md[0], '').replace(/\s+/g, ' ').trim(),
+      href: md[2].trim(),
+      linkLabel: md[1].trim(),
+    }
+  }
+  const html = raw.match(/<a\b[^>]*href=["'](https?:[^"']+)["'][^>]*>([\s\S]*?)<\/a>/i)
+  if (html?.[1] != null && html[2] != null) {
+    return {
+      text: raw.replace(html[0], '').replace(/\s+/g, ' ').trim(),
+      href: html[1].trim(),
+      linkLabel: html[2].replace(/<[^>]+>/g, '').trim(),
+    }
+  }
+  const paren = raw.match(/^(.*?)\s*\((https?:[^)\s]+)\)\s*$/i)
+  if (paren?.[1] != null && paren[2] != null) {
+    return {
+      text: paren[1].replace(/\s+/g, ' ').trim(),
+      href: paren[2].trim(),
+      linkLabel: paren[2].trim(),
+    }
+  }
+  return { text: raw, href: '', linkLabel: '' }
+}
+
+//0918 tip citation + REF 直鏈屬性
+function citationHtml(
+  id: string,
+  tip = '',
+  extra?: {
+    sourceName?: string
+    filePath?: string
+    filePreviewUrl?: string
+    chunkId?: string
+  }
+) {
+  const n = String(id).trim() || '1'
+  const { text, href, linkLabel } = splitTipLink(tip)
+  const tipAttr = text ? ` data-tip="${escapeAttr(text)}"` : ''
+  const hrefAttr = href ? ` data-href="${escapeAttr(href)}"` : ''
+  const labelAttr = linkLabel ? ` data-link-label="${escapeAttr(linkLabel)}"` : ''
+  const sourceNameAttr = extra?.sourceName
+    ? ` data-source-name="${escapeAttr(extra.sourceName)}"`
+    : ''
+  const filePathAttr = extra?.filePath ? ` data-file-path="${escapeAttr(extra.filePath)}"` : ''
+  const filePreviewAttr = extra?.filePreviewUrl
+    ? ` data-file-preview-url="${escapeAttr(extra.filePreviewUrl)}"`
+    : ''
+  const chunkIdAttr = extra?.chunkId ? ` data-chunk-id="${escapeAttr(extra.chunkId)}"` : ''
+  return `<span class="report-tiptap-citation" data-citation="${escapeHtml(n)}"${tipAttr}${hrefAttr}${labelAttr}${sourceNameAttr}${filePathAttr}${filePreviewAttr}${chunkIdAttr}>${escapeHtml(n)}</span>`
+}
+
+// REF_MARKER_RE / resolveRefMarker → ../editor/reviewCitationTip（0918 归类）
+
+function isRealHref(href: string) {
+  const value = String(href || '').trim()
+  if (!value || value === '#') return false
+  if (/^#(?:fn|footnote|cite|ref)/i.test(value)) return false
+  return /^(https?:|mailto:|tel:|\/|\.\/|\.\.\/|#\/)/i.test(value) || /^[a-z][a-z0-9+.-]*:/i.test(value)
+}
+
+function isCitationLabel(text: string) {
+  const label = text.replace(/<[^>]+>/g, '').trim()
+  return /^\[?\d+\]?$/.test(label) || /^[+*]?\d+$/.test(label)
+}
+
+function idFromLabel(label: string, fallback: string) {
+  const m = String(label).replace(/<[^>]+>/g, '').trim().match(/(\d+)/)
+  return m?.[1] ?? fallback
+}
+
+let citationSeq = 0
+function nextId(label: string) {
+  const fromLabel = idFromLabel(label, '')
+  if (fromLabel) return fromLabel
+  citationSeq += 1
+  return String(citationSeq)
+}
+
+function extractAttr(tag: string, name: string) {
+  const m = tag.match(new RegExp(`(?:^|\\s)${name}\\s*=\\s*(?:"([^"]*)"|'([^']*)'|([^\\s>]+))`, 'i'))
+  return (m?.[1] ?? m?.[2] ?? m?.[3] ?? '').trim()
+}
+
+const PLACEHOLDER = '%%GENAI_BLOCK_'
+
+function protectBlocks(source: string) {
+  const slots: string[] = []
+  const stash = (html: string) => {
+    const i = slots.length
+    slots.push(html)
+    return `${PLACEHOLDER}${i}%%`
+  }
+  let text = source.replace(
+    /<span\b[^>]*class="[^"]*report-tiptap-citation[^"]*"[^>]*>[\s\S]*?<\/span>/gi,
+    (m) => stash(m)
+  )
+  text = text.replace(/<(table|pre|blockquote|details)[\s\S]*?<\/\1>/gi, (m) => stash(m))
+  text = text.replace(/<(el-(?:tooltip|popover|link))\b[\s\S]*?<\/\1>/gi, (m) => stash(m))
+  return { text, slots }
+}
+
+function restoreBlocks(html: string, slots: string[]) {
+  return html.replace(new RegExp(`${PLACEHOLDER}(\\d+)%%`, 'g'), (_, i) => slots[Number(i)] ?? '')
+}
+
+/** Markdown 层 tip：脚注定义、[^n]、[+n]、[ref:uuid]；[+link] / [text](url) 不在此当 tip */
+//0918：先 REF（未命中删除），再旧 [+n]/[^n]
+function applyMarkdownTips(source: string) {
+  const footnotes = new Map<string, string>()
+  let text = source.replace(/^\[\^(\d+)\]:\s*(.+)$/gm, (_, id: string, tip: string) => {
+    footnotes.set(id, tip.trim())
+    return ''
+  })
+  text = text.replace(REF_MARKER_RE, (_full, chunkId: string) => {
+    const resolved = resolveRefMarker(chunkId)
+    // 未命中：删除占位，只显示已命中 tip；REF 入库后重解析
+    if (!resolved) return ''
+    return citationHtml(resolved.id, resolved.tip, resolved.extra)
+  })
+  //0918-1：匹配不到的 [ref:…] 一律过滤为空
+  text = text.replace(REF_ANY_MARKER_RE, '')
+  REF_MARKER_RE.lastIndex = 0
+  REF_ANY_MARKER_RE.lastIndex = 0
+  //0918-1
+  text = text.replace(/\[\+(\d+)\]/g, (_, id: string) => citationHtml(id))
+  text = text.replace(/\[\^(\d+)\](?!:)/g, (_, id: string) =>
+    citationHtml(id, footnotes.get(id) || '')
+  )
+  return { text, footnotes }
+}
+//0918
+
+function linkHtml(href: string, inner: string, title = '') {
+  const titleAttr = title ? ` title="${escapeAttr(title)}"` : ''
+  return `<a href="${escapeHtml(href)}" class="report-tiptap-link" target="_blank" rel="noopener noreferrer"${titleAttr}>${inner}</a>`
+}
+
+function isTipHref(href: string, title: string, _label?: string) {
+  if (isRealHref(href)) return false
+  if (title) return true
+  return /^#(?:fn|footnote|cite|ref)/i.test(href)
+}
+
+const markedParser = new Marked({
+  gfm: true,
+  breaks: true,
+  renderer: {
+    link({ href, title, tokens }) {
+      const text = this.parser.parseInline(tokens)
+      const tip = String(title || '').trim()
+      if (isTipHref(String(href || ''), tip)) {
+        return citationHtml(nextId(text), tip)
+      }
+      return linkHtml(String(href || '#'), text, tip)
+    },
+  },
+})
+
+/** HTML 层：已有 citation / title / 脚注 a / 数字 sup → tip；真实 a → +link */
+function applyHtmlTipsAndLinks(html: string) {
+  let out = html.replace(/<a\b([^>]*)>([\s\S]*?)<\/a>/gi, (m, attrs: string, inner: string) => {
+    if (/report-tiptap-citation/i.test(attrs)) return m
+    const href = extractAttr(attrs, 'href')
+    const title = extractAttr(attrs, 'title') || extractAttr(attrs, 'data-tip')
+    const label = inner.replace(/<[^>]+>/g, '').trim()
+    if (isTipHref(href, title) && !isRealHref(href)) {
+      return citationHtml(nextId(label || '1'), title)
+    }
+    if (/report-tiptap-link/i.test(attrs) && isRealHref(href)) return m
+    return linkHtml(href || '#', inner, title)
+  })
+
+  out = out.replace(/<(?:abbr|span)\b([^>]*)>([\s\S]*?)<\/(?:abbr|span)>/gi, (m, attrs: string, inner: string) => {
+    if (/report-tiptap-citation/i.test(attrs) || /report-tiptap-link/i.test(attrs)) return m
+    const citationId = extractAttr(attrs, 'data-citation')
+    const tip = extractAttr(attrs, 'data-tip') || extractAttr(attrs, 'title')
+    const label = inner.replace(/<[^>]+>/g, '').trim()
+    if (citationId) return citationHtml(citationId, tip)
+    if (tip && isCitationLabel(label)) return citationHtml(nextId(label), tip)
+    return m
+  })
+
+  out = out.replace(/<sup\b([^>]*)>([\s\S]*?)<\/sup>/gi, (m, attrs: string, inner: string) => {
+    const tip = extractAttr(attrs, 'title') || extractAttr(attrs, 'data-tip')
+    const label = inner.replace(/<[^>]+>/g, '').trim()
+    if (!tip || !isCitationLabel(label)) return m
+    return citationHtml(nextId(label), tip)
+  })
+
+  return out
+}
+
+/** Element Plus 层：el-tooltip / el-popover → 原 tip；el-link → +link */
+function applyElementPlusTipsAndLinks(html: string) {
+  let out = html.replace(
+    /<el-tooltip\b([^>]*)>([\s\S]*?)<\/el-tooltip>/gi,
+    (_m, attrs: string, inner: string) => {
+      const tip = extractAttr(attrs, 'content') || extractAttr(attrs, 'title')
+      const label = inner.replace(/<[^>]+>/g, '').trim()
+      if (!tip) return inner
+      return citationHtml(nextId(label || '1'), tip)
+    }
+  )
+  out = out.replace(
+    /<el-popover\b([^>]*)>([\s\S]*?)<\/el-popover>/gi,
+    (_m, attrs: string, inner: string) => {
+      const slotTip =
+        inner
+          .match(/<template[^>]*(?:#|v-slot:|slot=)[^>]*(?:content|default)[^>]*>([\s\S]*?)<\/template>/i)?.[1]
+          ?.replace(/<[^>]+>/g, ' ')
+          .replace(/\s+/g, ' ')
+          .trim() || ''
+      const tip = extractAttr(attrs, 'content') || extractAttr(attrs, 'title') || slotTip
+      const label = inner.replace(/<template[\s\S]*?<\/template>/gi, '').replace(/<[^>]+>/g, '').trim()
+      if (!tip) return inner.replace(/<template[\s\S]*?<\/template>/gi, '')
+      return citationHtml(nextId(label || '1'), tip)
+    }
+  )
+  out = out.replace(/<el-link\b([^>]*)>([\s\S]*?)<\/el-link>/gi, (_m, attrs: string, inner: string) => {
+    const href = extractAttr(attrs, 'href')
+    const title = extractAttr(attrs, 'title') || extractAttr(attrs, 'content')
+    const label = inner.replace(/<[^>]+>/g, '').trim()
+    if (!isRealHref(href) && title) {
+      return citationHtml(nextId(label || '1'), title)
+    }
+    return linkHtml(href || '#', inner, title)
+  })
+  return out
+}
+
+function applyFootnoteTips(html: string, footnotes: Map<string, string>) {
+  if (!footnotes.size) return html
+  return html.replace(
+    /<span class="report-tiptap-citation" data-citation="(\d+)"([^>]*)>([\s\S]*?)<\/span>/gi,
+    (m, id: string, rest: string, inner: string) => {
+      const raw = footnotes.get(id)
+      if (!raw) return m
+      const { text, href, linkLabel } = splitTipLink(raw)
+      if (/data-href=/.test(rest) && /data-tip=/.test(rest)) return m
+      const cleaned = String(rest).replace(/\sdata-(?:tip|href|link-label)="[^"]*"/gi, '')
+      const tipAttr = text ? ` data-tip="${escapeAttr(text)}"` : ''
+      const hrefAttr = href ? ` data-href="${escapeAttr(href)}"` : ''
+      const labelAttr = linkLabel ? ` data-link-label="${escapeAttr(linkLabel)}"` : ''
+      return `<span class="report-tiptap-citation" data-citation="${id}"${tipAttr}${hrefAttr}${labelAttr}${cleaned}>${inner}</span>`
+    }
+  )
+}
+
+const UNCLOSED_BLOCK_RE =
+  /<(table|thead|tbody|tr|td|th|pre|blockquote|ul|ol|li|p|h[1-6]|think|strong|b|em|i|u|s|code|mark|div|section|article|figure|figcaption|header|footer|details|summary|font|ins|del|caption|el-tooltip|el-popover|el-link|span|a)\b[^>]*>(?:(?!<\/\1>)[\s\S])*$/i
+
+export function stripThinkBlocks(source: string) {
+  return source
+    .replace(/<think\b[^>]*>[\s\S]*?<\/think>/gi, '')
+    .replace(/<think\b[\s\S]*$/i, '')
+}
+
+const CJK_CHAR =
+  /[\u3400-\u9FFF\uF900-\uFAFF\u3000-\u303F\u3040-\u30FF\uAC00-\uD7AF]/
+
+/** 補回可讀空格：空白規範化、中西交界（保留換行）；英文分詞交由後端保證 */
+export function normalizeReadableSpacing(source: string): string {
+  let text = String(source ?? '')
+  if (!text) return ''
+  text = text.replace(/\u00a0/g, ' ')
+  // 0914：空白規範化 + 中西交界（英文粘詞由後端輸出正常空格）
+  text = text.replace(/[^\S\n\r]+/g, ' ')
+  text = text.replace(
+    new RegExp(`(${CJK_CHAR.source})([A-Za-z0-9])`, 'g'),
+    '$1 $2'
+  )
+  text = text.replace(
+    new RegExp(`([A-Za-z0-9])(${CJK_CHAR.source})`, 'g'),
+    '$1 $2'
+  )
+  // 0914
+  return text
+}
+
+/** 是否像 Markdown（ATX 標題 / 列表 / 加粗），用於決定要不要拆 <p>/<br>/字面 \\n */
+function looksLikeMarkdown(text: string) {
+  const probe = String(text || '')
+    .replace(/<(?:br)\s*\/?>/gi, '\n')
+    .replace(/<\/(?:p|div)\s*>/gi, '\n')
+    .replace(/<(?:p|div)\b[^>]*>/gi, '')
+  return /(?:^|\n)\s{0,3}#{1,6}\s|(?:^|\n)\s{0,3}(?:[-*+]|\d{1,9}[.)])\s|\*{3}[^*\n]+\*{3}|\*{2}[^*\n]+\*{2}/m.test(
+    probe
+  )
+}
+
+/**   
+ * 接口正文常把 Markdown 塞进 <p>/<br> 或字面 \\n，
+ * marked 無法把行首 ### / * 認成標題與列表，畫面就會打出原始符號。
+ * 閉合 HTML 結構保留；僅在「像 Markdown」時拆開陷阱標籤。
+ */
+function normalizeMarkdownSource(source: string) {
+  let text = String(source ?? '')
+  if (!text) return ''
+
+  const realNl = (text.match(/\r?\n/g) || []).length
+  const escapedNl = (text.match(/\\n/g) || []).length
+  if (escapedNl > 0 && escapedNl >= realNl) {
+    const unescaped = text.replace(/\\r\\n/g, '\n').replace(/\\n/g, '\n').replace(/\\t/g, '\t')
+    if (looksLikeMarkdown(unescaped)) text = unescaped
+  }
+
+  if (!looksLikeMarkdown(text)) {
+    return normalizeReadableSpacing(text)
+  }
+
+  if (/<br\s*\/?>/i.test(text)) {
+    text = text.replace(/<br\s*\/?>/gi, '\n')
+  }
+
+  if (/<(?:p|div)\b/i.test(text)) {
+    const unwrapped = text
+      .replace(/<\/(?:p|div)\s*>\s*<(?:p|div)\b[^>]*>/gi, '\n\n')
+      .replace(/<(?:p|div)\b[^>]*>/gi, '')
+      .replace(/<\/(?:p|div)\s*>/gi, '\n')
+    if (looksLikeMarkdown(unwrapped)) text = unwrapped
+  }
+
+  return normalizeReadableSpacing(text)
+}
+
+/**
+ * 在 source 上掃描，自動跳過 ```fence``` 與行內 `code`，
+ * 避免代碼裡的 ** / * / # 被當成未閉合 Markdown。
+ */
+function scanMarkdownOutsideCode(
+  source: string,
+  visit: (i: number, ch: string) => number | void
+) {
+  let i = 0
+  let inFence = false
+  while (i < source.length) {
+    if (source.startsWith('```', i)) {
+      inFence = !inFence
+      i += 3
+      continue
+    }
+    if (inFence) {
+      i += 1
+      continue
+    }
+    if (source.charAt(i) === '`') {
+      let n = 0
+      while (source.charAt(i + n) === '`') n += 1
+      const open = source.slice(i, i + n)
+      const closeAt = source.indexOf(open, i + n)
+      if (closeAt < 0) {
+        // 半截行內 code：剩餘整段視為 code，不再掃描標記
+        return
+      }
+      i = closeAt + n
+      continue
+    }
+    const step = visit(i, source.charAt(i))
+    i += typeof step === 'number' && step > 0 ? step : 1
+  }
+}
+
+function lastUnpairedFence(source: string, fence: string) {
+  let odd = false
+  let last = -1
+  scanMarkdownOutsideCode(source, (i) => {
+    // 統計 ** 時整段跳過 ***，避免三星被拆成 ** + *
+    if (fence === '**' && source.startsWith('***', i)) return 3
+    if (fence === '__' && source.startsWith('___', i)) return 3
+    if (source.startsWith(fence, i)) {
+      odd = !odd
+      last = i
+      return fence.length
+    }
+  })
+  return odd ? last : -1
+}
+
+function lastUnpairedInlineTick(source: string) {
+  let odd = false
+  let last = -1
+  let i = 0
+  let inFence = false
+  while (i < source.length) {
+    if (source.startsWith('```', i)) {
+      inFence = !inFence
+      i += 3
+      continue
+    }
+    if (inFence) {
+      i += 1
+      continue
+    }
+    if (source.charAt(i) === '`') {
+      odd = !odd
+      last = i
+    }
+    i += 1
+  }
+  return odd ? last : -1
+}
+
+function lastUnpairedSingleStar(source: string) {
+  let odd = false
+  let last = -1
+  scanMarkdownOutsideCode(source, (i) => {
+    if (source.startsWith('***', i)) return 3
+    if (source.startsWith('**', i)) return 2
+    if (source.charAt(i) !== '*') return
+    const atLineStart = i === 0 || source.charAt(i - 1) === '\n'
+    const next = source.charAt(i + 1)
+    if (atLineStart && (next === ' ' || next === '\t')) return 1
+    odd = !odd
+    last = i
+  })
+  return odd ? last : -1
+}
+
+function lastUnpairedSingleUnderscore(source: string) {
+  let odd = false
+  let last = -1
+  scanMarkdownOutsideCode(source, (i) => {
+    if (source.startsWith('__', i)) return 2
+    if (source.charAt(i) !== '_') return
+    const prev = i === 0 ? ' ' : source.charAt(i - 1)
+    const next = source.charAt(i + 1)
+    if (/[A-Za-z0-9]/.test(prev) && /[A-Za-z0-9]/.test(next)) return 1
+    odd = !odd
+    last = i
+  })
+  return odd ? last : -1
+}
+
+/** 完整主題分隔線（HR），不應當半截標記 */
+function isCompleteThematicBreak(text: string) {
+  return /^\s*(?:-{3,}|\*{3,}|_{3,})\s*$/.test(text)
+}
+
+/** 打字机：切开已闭合前缀 vs 半截标记；半截内文继续打出，避免整段闪没 */
+function splitStreamStablePrefix(source: string) {
+  const cuts: number[] = []
+  const lastLt = source.lastIndexOf('<')
+  const lastGt = source.lastIndexOf('>')
+  if (lastLt > lastGt && /[a-zA-Z/!]/.test(source.charAt(lastLt + 1) || '')) {
+    cuts.push(lastLt)
+  }
+  const block = source.match(UNCLOSED_BLOCK_RE)
+  if (block && block.index != null) cuts.push(block.index)
+  const linkOpen = source.search(/\[[^\]]*$|\[[^\]]*\]\([^)]*$/)
+  if (linkOpen >= 0) cuts.push(linkOpen)
+  const plusOpen = source.search(/\[\+(?:\d+)?$|\[\^\d*$/)
+  if (plusOpen >= 0) cuts.push(plusOpen)
+  for (const fence of ['```', '***', '**', '__', '~~'] as const) {
+    const at = lastUnpairedFence(source, fence)
+    if (at >= 0) cuts.push(at)
+  }
+  const tick = lastUnpairedInlineTick(source)
+  if (tick >= 0) cuts.push(tick)
+  const star = lastUnpairedSingleStar(source)
+  if (star >= 0) cuts.push(star)
+  const under = lastUnpairedSingleUnderscore(source)
+  if (under >= 0) cuts.push(under)
+  const mdBlock = lastIncompleteMarkdownCut(source)
+  if (mdBlock >= 0) cuts.push(mdBlock)
+  if (!cuts.length) return { stable: source, rest: '' }
+  const at = Math.min(...cuts)
+  const stable = source.slice(0, at)
+  const rest = source.slice(at)
+  // 誤切：rest 本身已是完整 HR，應併入 stable 交給 marked
+  if (rest && isCompleteThematicBreak(rest.trim())) {
+    return { stable: source, rest: '' }
+  }
+  return { stable, rest }
+}
+
+/** 半截标题 / 列表 / 引用 / 表格 / setext 先不交给 marked，避免结构来回跳 */
+function lastIncompleteMarkdownCut(source: string) {
+  const cuts: number[] = []
+  const pushMarker = (re: RegExp) => {
+    const m = source.match(re)
+    if (!m || m.index == null) return
+    cuts.push(m[0].startsWith('\n') ? m.index + 1 : m.index)
+  }
+  // 僅標題符號、尚無正文
+  pushMarker(/(?:^|\n) {0,3}#{1,6}$/)
+  // ### 後只有空格、標題字還在打
+  pushMarker(/(?:^|\n) {0,3}#{1,6}[ \t]+$/)
+  pushMarker(/(?:^|\n) {0,3}(?:[-*+]|\d{1,9}[.)])$/)
+  pushMarker(/(?:^|\n) {0,3}>$/)
+  // 僅 1～2 個 -/_（等第 3 個才成 HR）；完整 --- 不切
+  pushMarker(/(?:^|\n) {0,3}(?:-{1,2}|_{1,2})$/)
+  // setext 下劃線未完成（1～2）；≥3 交給 marked 解析為標題/HR
+  const setext = source.match(/\n[ \t]*([-=])\1{0,1}$/)
+  if (setext && setext.index != null) cuts.push(setext.index + 1)
+  const table = incompleteTableCut(source)
+  if (table >= 0) cuts.push(table)
+  return cuts.length ? Math.min(...cuts) : -1
+}
+
+function incompleteTableCut(source: string) {
+  const lines = source.split('\n')
+  const lastLine = lines[lines.length - 1]
+  if (!lines.length || lastLine == null || !/^\s*\|/.test(lastLine)) return -1
+  let i = lines.length - 1
+  while (i > 0 && /^\s*\|/.test(lines[i - 1] ?? '')) i -= 1
+  const block = lines.slice(i)
+  const hasSep = block.some((line) => /^\s*\|?\s*:?-{2,}/.test(line))
+  const last = block[block.length - 1] || ''
+  const lastComplete = /\|.+\|/.test(last.trim())
+  if (hasSep && lastComplete && block.length >= 3) return -1
+  return i === 0 ? 0 : lines.slice(0, i).join('\n').length + 1
+}
+
+function htmlInnerPlain(rest: string) {
+  const gt = rest.indexOf('>')
+  if (gt < 0) return ''
+  let inner = rest.slice(gt + 1)
+  inner = inner.replace(/<\/[a-zA-Z][^>]*>/g, '').replace(/<[a-zA-Z][^>]*\/?>/g, '')
+  const lastLt = inner.lastIndexOf('<')
+  const lastGt = inner.lastIndexOf('>')
+  if (lastLt > lastGt) inner = inner.slice(0, lastLt)
+  return inner
+}
+
+/**
+ * 半截标记不把内文藏掉：`**xxx` / `<strong>xxx` 继续打字，只不画出 ** / 标签本身。
+ * 闭合后走正式解析变成粗体，与打字机进度同步、避免整段闪没。
+ */
+function visiblePlainFromIncomplete(rest: string) {
+  if (!rest) return ''
+  const leadWs = rest.match(/^[ \t]*/)?.[0] ?? ''
+  const t = rest.slice(leadWs.length)
+  if (!t) return ''
+  if (isCompleteThematicBreak(t)) return ''
+  if (t.startsWith('```')) {
+    const nl = t.indexOf('\n')
+    return nl < 0 ? '' : leadWs + t.slice(nl + 1)
+  }
+  if (t.startsWith('***')) return leadWs + t.slice(3)
+  if (t.startsWith('**')) return leadWs + t.slice(2)
+  if (t.startsWith('__')) return leadWs + t.slice(2)
+  if (t.startsWith('~~')) return leadWs + t.slice(2)
+  if (/^#{1,6}$/.test(t)) return ''
+  if (/^#{1,6}[ \t]+$/.test(t)) return ''
+  if (/^#{1,6}[ \t]+/.test(t)) return leadWs + t.replace(/^#{1,6}[ \t]+/, '')
+  if (/^([-*+]|\d{1,9}[.)])$/.test(t)) return ''
+  if (/^([-*+]|\d{1,9}[.)]) /.test(t)) return leadWs + t.replace(/^([-*+]|\d{1,9}[.)]) /, '')
+  if (t === '>') return ''
+  if (/^\s*\|/.test(t)) return leadWs + t.replace(/^\s*\|/, '').replace(/\|/g, ' ')
+  if (t.startsWith('`')) return leadWs + t.slice(1).replace(/`/g, '')
+  if (t.startsWith('[') || t.startsWith('![')) {
+    const inner = t.replace(/^!?\[/, '')
+    const close = inner.indexOf(']')
+    return leadWs + (close < 0 ? inner : inner.slice(0, close))
+  }
+  if (t.startsWith('<')) return leadWs + htmlInnerPlain(t)
+  if (t.startsWith('*')) return leadWs + t.slice(1)
+  if (t.startsWith('_')) return leadWs + t.slice(1)
+  return leadWs + t
+}
+
+function appendPlainTail(html: string, rest: string) {
+  if (!rest) return html
+  const escaped = escapeHtml(rest).replace(/\n/g, '<br>')
+  if (!html) return `<p>${escaped}</p>`
+  // 表格 / 标题半截单独成段，避免塞进上一块后再整体跳结构
+  if (/^\s*\|/.test(rest.trimStart()) || /^\s*#{1,6}(?:\s|$)/.test(rest.trimStart())) {
+    return `${html}<p>${escaped}</p>`
+  }
+  const closeRe = /<\/(?:p|li|h[1-6]|td|th)\b[^>]*>/gi
+  let last = -1
+  let m: RegExpExecArray | null
+  while ((m = closeRe.exec(html))) last = m.index
+  if (last >= 0) return html.slice(0, last) + escaped + html.slice(last)
+  return `${html}<p>${escaped}</p>`
+}
+
+/** 将 Markdown / HTML / Element Plus 解析为 TipTap 可 setContent 的片段 */
+export function parseMarkdownToHtml(source: string): string {
+  const text = normalizeMarkdownSource(String(source ?? ''))
+  if (!text.trim()) return ''
+  citationSeq = 0
+  try {
+    const withoutThink = stripThinkBlocks(text)
+    if (!withoutThink.trim()) return ''
+    const mdTips = applyMarkdownTips(withoutThink)
+    const protectedSrc = protectBlocks(mdTips.text)
+    const parsed = markedParser.parse(protectedSrc.text, { async: false })
+    let html = typeof parsed === 'string' ? parsed.trim() : ''
+    html = restoreBlocks(html, protectedSrc.slots)
+    html = applyHtmlTipsAndLinks(html)
+    html = applyElementPlusTipsAndLinks(html)
+    html = applyFootnoteTips(html, mdTips.footnotes)
+    return html
+  } catch {
+    return `<p>${escapeHtml(stripThinkBlocks(text))}</p>`
+  }
+}
+
+let streamStableCacheKey = ''
+let streamStableCacheHtml = ''
+
+/** 打字机逐帧：完整语法走解析；半截 ** / 标签不画，内文继续打出，打完再灌全文 */
+export function parseMarkdownToHtmlStream(source: string): string {
+  const text = stripThinkBlocks(normalizeMarkdownSource(String(source ?? '')))
+  if (!text) return ''
+  const { stable, rest } = splitStreamStablePrefix(text)
+  if (!rest) {
+    const html = parseMarkdownToHtml(text)
+    streamStableCacheKey = text
+    streamStableCacheHtml = html
+    return html
+  }
+  let html = ''
+  if (stable) {
+    if (stable === streamStableCacheKey) html = streamStableCacheHtml
+    else {
+      html = parseMarkdownToHtml(stable)
+      streamStableCacheKey = stable
+      streamStableCacheHtml = html
+    }
+  }
+  return appendPlainTail(html, visiblePlainFromIncomplete(rest))
+}