Agent-System-Frontend/src/previewsetup.ts

// 预览能力装配（裁定 D1–D4）：把重解析器经 configureFilePreviewLoaders 注入 @ui。
// 全部 dynamic import——对应格式首次预览时才下载 chunk，主 bundle 不含解析器。
// 链接预览 fetcher 走服务端 /api/url-preview（SSRF 防护在服务端执行端）。
import { configureFilePreviewLoaders, configureLinkPreviewFetcher, renderMarkdown, setSafeHTML, stripUnsafeLinks } from '@ui'
import { auth } from './auth'
import type { LinkPreviewData } from '@ui'
import { fitPagesToWidth } from './previewfit'

const fetchBuf = async (url: string | undefined): Promise<ArrayBuffer> => {
  const r = await fetch(url!)
  if (!r.ok) throw new Error(`HTTP ${r.status}`)
  return r.arrayBuffer()
}

// docx/OOXML 是 ZIP 容器，真文件必以 PK 魔数开头（PK\x03\x04 常规，PK\x05\x06 空包）。
// 智能体生成侧可能写出"内容是文本的伪 docx"，先嗅探再喂 mammoth：
// 非 ZIP 直接按文本/markdown 回退渲染，不再抛 JSZip 底层错误。
const isZipBytes = (buf: ArrayBuffer): boolean => {
  if (buf.byteLength < 4) return false
  const h = new Uint8Array(buf, 0, 4)
  return h[0] === 0x50 && h[1] === 0x4b &&
    ((h[2] === 0x03 && h[3] === 0x04) || (h[2] === 0x05 && h[3] === 0x06))
}

const renderTextFallback = (el: HTMLElement, buf: ArrayBuffer, name: string | undefined, kindLabel: string) => {
  const note = document.createElement('div')
  note.className = 'ui-fp-note'
  note.textContent = `「${name ?? '该文件'}」不是有效的 ${kindLabel} 文件（缺少 ZIP 结构），已按文本内容展示；需要正式格式请让智能体重新生成。`
  const body = document.createElement('div')
  body.className = 'doc-viewer'
  setSafeHTML(body, renderMarkdown(new TextDecoder().decode(buf), { breaks: true }))
  el.innerHTML = ''
  el.append(note, body)
}

const docxLoader = async (el: HTMLElement, source: { contentUrl?: string; name?: string }) => {
  const buf = await fetchBuf(source.contentUrl)
  if (!isZipBytes(buf)) {
    renderTextFallback(el, buf, source.name, 'DOCX')
    return
  }
  try {
    const { renderAsync } = await import('docx-preview')
    el.innerHTML = ''
    await renderAsync(buf, el, el, { className: 'docx', inWrapper: true, breakPages: true })
    stripUnsafeLinks(el)
    return fitPagesToWidth(el, el)
  } catch (err) {
    console.error('[preview:docx] page render failed, falling back', source.name, err)
  }
  const mammoth = await import('mammoth')
  const { value } = await mammoth.convertToHtml({ arrayBuffer: buf })
  el.innerHTML = ''
  const body = document.createElement('div')
  body.className = 'doc-viewer'
  setSafeHTML(body, value)
  el.append(body)
}

const xlsxLoader = async (el: HTMLElement, source: { contentUrl?: string; name?: string }) => {
  const buf = await fetchBuf(source.contentUrl)
  // .xls 是 OLE 复合文档，不是 ZIP；只对 OOXML（xlsx/xlsm，或未给文件名）做魔数嗅探。
  if (!/\.xls$/i.test(source.name ?? '') && !isZipBytes(buf)) {
    renderTextFallback(el, buf, source.name, 'XLSX')
    return
  }
  const XLSX = await import('xlsx')
  const wb = XLSX.read(buf, { type: 'array' })
  el.innerHTML = ''
  const tabs = document.createElement('div')
  tabs.className = 'ui-fp-sheet-tabs'
  const wrap = document.createElement('div')
  wrap.className = 'ui-fp-sheet-wrap'
  el.append(tabs, wrap)
  wb.SheetNames.forEach((name, i) => {
    const tab = document.createElement('button')
    tab.className = `ui-fp-sheet-tab ${i === 0 ? 'on' : ''}`
    tab.textContent = name
    tab.onclick = () => {
      tabs.querySelectorAll('.ui-fp-sheet-tab').forEach(t => t.classList.remove('on'))
      tab.classList.add('on')
      renderSheet(wrap, XLSX.utils.sheet_to_json<unknown[]>(wb.Sheets[name], { header: 1, defval: '' }))
    }
    tabs.append(tab)
    if (i === 0) renderSheet(wrap, XLSX.utils.sheet_to_json<unknown[]>(wb.Sheets[name], { header: 1, defval: '' }))
  })
}

function renderSheet(wrap: HTMLElement, rows: unknown[][]) {
  wrap.innerHTML = ''
  if (rows.length > 200) {
    const note = document.createElement('div')
    note.className = 'ui-fp-note'
    note.textContent = `仅显示前 200 行（共 ${rows.length} 行）——完整内容请下载`
    wrap.append(note)
  }
  const table = document.createElement('table')
  table.className = 'ui-fp-sheet'
  for (const row of rows.slice(0, 200)) {
    const tr = document.createElement('tr')
    for (const cell of row) {
      const td = document.createElement('td')
      td.textContent = cell == null ? '' : String(cell)
      td.title = td.textContent
      tr.append(td)
    }
    table.append(tr)
  }
  wrap.append(table)
}

const pptxLoader = async (el: HTMLElement, source: { contentUrl?: string; name?: string }) => {
  const buf = await fetchBuf(source.contentUrl)
  // .ppt 是老版二进制；OOXML（pptx/ppsx）才要求 ZIP 魔数。
  if (!/\.ppt$/i.test(source.name ?? '') && !isZipBytes(buf)) {
    renderTextFallback(el, buf, source.name, 'PPTX')
    return
  }
  // pptx2html@0.3.4 的 module 入口引用了大小写错配的 ./txml（Linux 打包必炸），
  // 按 D2 备选条款改用 pptx-preview（init(dom).preview(arrayBuffer) 全保真渲染）。
  const { init } = await import('pptx-preview')
  el.innerHTML = ''
  const stage = document.createElement('div')
  stage.className = 'ui-fp-slide-stage'
  el.append(stage)
  const previewer = init(stage, { width: stage.clientWidth || 900 })
  await previewer.preview(buf)
  stripUnsafeLinks(el)
}

export function wirePreviewCapabilities() {
  configureFilePreviewLoaders({ docx: docxLoader, xlsx: xlsxLoader, pptx: pptxLoader })
  configureLinkPreviewFetcher(async href => {
    const r = await fetch(auth.withKey(`/api/url-preview?url=${encodeURIComponent(href)}`))
    if (!r.ok) return { url: href, error: `HTTP ${r.status}` }
    return (await r.json()) as LinkPreviewData
  })
}