dsh-cait 1.0.3 → 1.0.4

package/lib/formats.js

@@ -1,917 +1,917 @@
-/**
- * Dependency-free parsers/builders for the first-tier CAT exchange formats:
- * CSV, SRT, WebVTT, PO and a basic XLIFF 1.2 subset.
- *
- * These formats are intentionally shallow: inline markup is flattened to text,
- * complex subtitle styling and XLIFF extensions are not round-tripped. The
- * original source text is kept so the normal CAT/TM/glossary flow applies.
- */
-import { CatError } from './errors.js'
-import { decodeXml, escapeXml } from './xml.js'
-
-const SRT_TIME = /^(\d{1,3}):(\d{2}):(\d{2})[,.](\d{1,3})$/
-const VTT_TIME = /^(?:(\d{1,3}):)?(\d{1,2}):(\d{2})\.(\d{1,3})$/
-const INLINE_XML = /<\/?(?:bpt|ept|ph|g|x|bx|ex|mrk|sub|hi|span)\b[^>]*>/gi
-const INLINE_TOKEN = /\{(\d+)\}/g
-
-function formatError(code, message) {
-  return new CatError(code, message)
-}
-
-function readXmlAttr(source, name) {
-  const match = String(source ?? '').match(new RegExp(`\\b${name}\\s*=\\s*["']([^"']*)["']`, 'i'))
-  return match ? decodeXml(match[1]) : ''
-}
-
-/**
- * Convert XLIFF inline markup to stable {n} tokens while keeping the original
- * XML for export. This prevents <ph>, <bpt>/<ept> and <g> placeholders from
- * being silently dropped when a translation is written back.
- */
-function extractInline(raw) {
-  const tokens = []
-  const text = String(raw ?? '').replace(INLINE_XML, (xml) => {
-    const token = `{${tokens.length + 1}}`
-    tokens.push({ token, xml })
-    return token
-  })
-  return { text: decodeXml(text).trim(), tokens }
-}
-
-/** Render a tokenised string back to XML, escaping only the text portions. */
-function renderInline(value, tokens) {
-  const text = String(value ?? '')
-  const map = new Map((Array.isArray(tokens) ? tokens : []).map((item) => [item.token, item.xml]))
-  let output = ''
-  let last = 0
-  for (const match of text.matchAll(INLINE_TOKEN)) {
-    output += escapeXml(text.slice(last, match.index))
-    output += map.get(match[0]) ?? escapeXml(match[0])
-    last = match.index + match[0].length
-  }
-  output += escapeXml(text.slice(last))
-  return output
-}
-
-/** Tokens present in the source but missing from the target. */
-export function missingInlineTokens(tokens, target) {
-  if (!Array.isArray(tokens) || !tokens.length) return []
-  const value = String(target ?? '')
-  return tokens.map((item) => item.token).filter((token) => !value.includes(token))
-}
-
-/** Append placeholders the translator omitted, so export never silently loses them. */
-function completeInline(value, tokens) {
-  let output = String(value ?? '')
-  for (const item of Array.isArray(tokens) ? tokens : []) {
-    if (!output.includes(item.token)) output += `${output && !/\s$/.test(output) ? ' ' : ''}${item.token}`
-  }
-  return output
-}
-
-export function parseCsvRows(text) {
-  const input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
-  const rows = []
-  let row = []
-  let field = ''
-  let quoted = false
-
-  for (let index = 0; index < input.length; index += 1) {
-    const char = input[index]
-    if (quoted) {
-      if (char === '"') {
-        if (input[index + 1] === '"') {
-          field += '"'
-          index += 1
-        } else {
-          quoted = false
-        }
-      } else {
-        field += char
-      }
-      continue
-    }
-
-    if (char === '"' && field === '') {
-      quoted = true
-    } else if (char === ',') {
-      row.push(field)
-      field = ''
-    } else if (char === '\n') {
-      row.push(field)
-      rows.push(row)
-      row = []
-      field = ''
-    } else {
-      field += char
-    }
-  }
-
-  if (quoted) throw formatError('CSV_PARSE', 'CSV 引号未闭合。')
-  if (field.length || row.length) {
-    row.push(field)
-    rows.push(row)
-  }
-  return rows
-}
-
-export function parseCsv(text, { sourceColumn, targetColumn } = {}) {
-  const rows = parseCsvRows(text)
-  if (!rows.length) throw formatError('CSV_PARSE', 'CSV 中没有可用的行。')
-
-  const normalize = (value) => String(value ?? '').trim().toLowerCase()
-  const sourceAliases = new Set(['source', 'src', 'original', '原文', '源文', '源文本'])
-  const targetAliases = new Set(['target', 'tgt', 'translation', '译文', '目標', '目标', '目标文本'])
-  const first = rows[0].map(normalize)
-  const hasHeader = first.some((value) => sourceAliases.has(value) || targetAliases.has(value))
-  const resolveColumn = (rawValue, aliases, fallback) => {
-    const value = String(rawValue ?? '').trim()
-    if (!value) return fallback
-    if (/^\d+$/.test(value)) {
-      const index = Number(value) - 1
-      if (index < 0) throw formatError('CSV_PARSE', 'CSV 列号从 1 开始。')
-      return index
-    }
-    const index = first.findIndex((item) => item === value.toLowerCase())
-    if (index < 0) throw formatError('CSV_PARSE', `CSV 中找不到列“${value}”。`)
-    return index
-  }
-
-  let sourceIndex = first.findIndex((value) => sourceAliases.has(value))
-  let targetIndex = first.findIndex((value) => targetAliases.has(value))
-  let start = 0
-
-  if (sourceColumn !== undefined || targetColumn !== undefined) {
-    sourceIndex = resolveColumn(sourceColumn, sourceAliases, sourceIndex >= 0 ? sourceIndex : 0)
-    targetIndex = resolveColumn(targetColumn, targetAliases, targetIndex)
-    start = hasHeader ? 1 : 0
-  } else if (sourceIndex >= 0 && targetIndex >= 0) {
-    start = 1
-  } else {
-    sourceIndex = 0
-    targetIndex = rows[0].length > 1 ? 1 : -1
-  }
-
-  const maxColumns = rows.reduce((max, row) => Math.max(max, row.length), 0)
-  if (sourceIndex < 0 || sourceIndex >= maxColumns) {
-    throw formatError('CSV_PARSE', 'CSV 源文列不存在。')
-  }
-  if (targetIndex >= maxColumns) {
-    throw formatError('CSV_PARSE', 'CSV 译文列不存在。')
-  }
-
-  const segments = []
-  for (let index = start; index < rows.length; index += 1) {
-    const row = rows[index]
-    const source = String(row[sourceIndex] ?? '').trim()
-    if (!source) continue
-    const target = targetIndex >= 0 ? String(row[targetIndex] ?? '').trim() : ''
-    segments.push({
-      source,
-      target,
-      paragraph: segments.length,
-      meta: { format: 'csv' },
-    })
-  }
-
-  if (!segments.length) throw formatError('CSV_NO_UNITS', 'CSV 中没有可用的源文行。')
-  return { format: 'csv', segments, warnings: [] }
-}
-
-function parseSrtTime(value) {
-  const match = String(value).trim().match(SRT_TIME)
-  if (!match) return null
-  const [, hours, minutes, seconds, millis] = match
-  return Number(hours) * 3600000 + Number(minutes) * 60000 + Number(seconds) * 1000 + Number(millis.padEnd(3, '0'))
-}
-
-export function formatSrtTime(milliseconds) {
-  const total = Math.max(0, Math.round(Number(milliseconds) || 0))
-  const hours = Math.floor(total / 3600000)
-  const minutes = Math.floor((total % 3600000) / 60000)
-  const seconds = Math.floor((total % 60000) / 1000)
-  const millis = total % 1000
-  return `${String(hours).padStart(2, '0')}:${String(minutes).padStart(2, '0')}:${String(seconds).padStart(2, '0')},${String(millis).padStart(3, '0')}`
-}
-
-function stripSubtitleMarkup(value) {
-  const assOverrides = String(value).match(/\{\\[^}]+\}/g) ?? []
-  return {
-    // Semantic tags such as <i>, <b>, <font> and WebVTT <c>/<v>/<ruby> are
-    // meaningful translation content and stay in the string verbatim.
-    text: decodeXml(String(value).replace(/<br\s*\/?>/gi, '\n').replace(/\{\\[^}]+\}/g, '')),
-    count: assOverrides.length,
-  }
-}
-
-function parseSubtitleBlocks(text, kind) {
-  const input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n').trim()
-  if (!input) throw formatError(`${kind.toUpperCase()}_PARSE`, `${kind.toUpperCase()} 文件内容为空。`)
-
-  const blocks = input.split(/\n{2,}/)
-  const segments = []
-  const warnings = []
-  let markupCount = 0
-
-  for (const block of blocks) {
-    const lines = block.split('\n')
-    if (!lines.length) continue
-    let cursor = 0
-    if (lines[0].trim().toUpperCase().startsWith('WEBVTT')) {
-      lines.shift()
-      while (lines.length && !lines[0].trim()) lines.shift()
-      if (!lines.length) continue
-    }
-    if (lines[0].trim().startsWith('NOTE') || lines[0].trim().startsWith('STYLE') || lines[0].trim().startsWith('REGION')) continue
-
-    let identifier = ''
-    if (!lines[cursor].includes('-->')) {
-      identifier = lines[cursor].trim()
-      cursor += 1
-    }
-    if (!lines[cursor]?.includes('-->')) continue
-    const timing = lines[cursor].trim()
-    cursor += 1
-
-    const timingMatch = timing.match(/^(\S+)\s*-->\s*(\S+)(.*)$/)
-    if (!timingMatch) {
-      throw formatError(`${kind.toUpperCase()}_PARSE`, `字幕时间轴无效: ${timing}`)
-    }
-    const start = kind === 'srt' ? parseSrtTime(timingMatch[1]) : parseVttTime(timingMatch[1])
-    const end = kind === 'srt' ? parseSrtTime(timingMatch[2]) : parseVttTime(timingMatch[2])
-    if (start === null || end === null) {
-      throw formatError(`${kind.toUpperCase()}_PARSE`, `字幕时间轴无效: ${timing}`)
-    }
-    if (end <= start) {
-      throw formatError(
-        `${kind.toUpperCase()}_PARSE`,
-        `字幕结束时间必须晚于开始时间: ${timing}`,
-      )
-    }
-
-    const rawText = lines.slice(cursor).join('\n').trim()
-    if (!rawText) continue
-    const stripped = stripSubtitleMarkup(rawText)
-    markupCount += stripped.count
-    const text = stripped.text.trim()
-    if (!text) continue
-    segments.push({
-      source: text,
-      target: '',
-      paragraph: segments.length,
-      meta: {
-        format: kind,
-        start,
-        end,
-        settings: timingMatch[3].trim(),
-        identifier,
-      },
-    })
-  }
-
-  if (!segments.length) throw formatError(`${kind.toUpperCase()}_NO_UNITS`, `${kind.toUpperCase()} 中没有可用的字幕。`)
-  if (markupCount) {
-    warnings.push({
-      code: `${kind.toUpperCase()}_INLINE_MARKUP_LOST`,
-      count: markupCount,
-      message: `${kind.toUpperCase()} 含 ${markupCount} 个 ASS 样式覆盖标记（{\\...}），导入时已移除；语义标签已保留。`,
-    })
-  }
-  return { format: kind, segments, warnings }
-}
-
-export function parseSrt(text) {
-  return parseSubtitleBlocks(text, 'srt')
-}
-
-function parseVttTime(value) {
-  const match = String(value).trim().match(VTT_TIME)
-  if (!match) return null
-  const [, hours, minutes, seconds, millis] = match
-  return Number(hours || 0) * 3600000 + Number(minutes) * 60000 + Number(seconds) * 1000 + Number(millis.padEnd(3, '0'))
-}
-
-export function formatVttTime(milliseconds) {
-  const total = Math.max(0, Math.round(Number(milliseconds) || 0))
-  const hours = Math.floor(total / 3600000)
-  const minutes = Math.floor((total % 3600000) / 60000)
-  const seconds = Math.floor((total % 60000) / 1000)
-  const millis = total % 1000
-  return `${String(hours).padStart(2, '0')}:${String(minutes).padStart(2, '0')}:${String(seconds).padStart(2, '0')}.${String(millis).padStart(3, '0')}`
-}
-
-export function parseVtt(text) {
-  return parseSubtitleBlocks(text, 'vtt')
-}
-
-function subtitleText(segment) {
-  return String(segment.target || segment.source || '').trim()
-}
-
-function cueTimes(segment, fallbackStart) {
-  const start = Number(segment.meta?.start)
-  const end = Number(segment.meta?.end)
-  if (Number.isFinite(start) && start >= 0 && Number.isFinite(end) && end > start) {
-    return { start, end }
-  }
-  // Text-only documents (and any cue without usable timing metadata) start
-  // where the previous cue ended. This avoids the old index * 2000 fallback
-  // overlapping an explicitly timed cue that sits earlier in the document.
-  const safeStart = Number.isFinite(start) && start >= 0 ? start : fallbackStart
-  return { start: safeStart, end: safeStart + 2000 }
-}
-
-function buildCues(segments) {
-  let cursor = 0
-  return segments.map((segment) => {
-    const times = cueTimes(segment, cursor)
-    cursor = Math.max(cursor, times.end)
-    return { segment, times }
-  })
-}
-
-export function buildSrt(doc) {
-  return buildCues(doc.segments)
-    .map(({ segment, times }, index) => [
-      String(index + 1),
-      `${formatSrtTime(times.start)} --> ${formatSrtTime(times.end)}`,
-      subtitleText(segment),
-      '',
-    ].join('\n'))
-    .join('\n')
-}
-
-export function buildVtt(doc) {
-  const cues = buildCues(doc.segments).map(({ segment, times }) => {
-    const identifier = segment.meta?.identifier ? `${segment.meta.identifier}\n` : ''
-    return `${identifier}${formatVttTime(times.start)} --> ${formatVttTime(times.end)}\n${subtitleText(segment)}\n`
-  })
-  return `WEBVTT\n\n${cues.join('\n')}`
-}
-
-function parsePoQuoted(raw) {
-  const match = String(raw).trim().match(/^"((?:\\.|[^"\\])*)"$/)
-  if (!match) return null
-  return match[1].replace(/\\(n|t|r|"|\\)/g, (all, char) => {
-    if (char === 'n') return '\n'
-    if (char === 't') return '\t'
-    if (char === 'r') return '\r'
-    return char
-  })
-}
-
-function poField(line) {
-  const match = line.match(/^(msgctxt|msgid_plural|msgid|msgstr(?:\[\d+\])?)\s+(.*)$/)
-  if (!match) return null
-  const value = parsePoQuoted(match[2])
-  if (value === null) return null
-  return { name: match[1], value }
-}
-
-export function parsePo(text) {
-  const input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
-  if (!input.trim()) throw formatError('PO_PARSE', 'PO 文件内容为空。')
-
-  const segments = []
-  const obsolete = []
-  let entry = null
-  let current = null
-  let pendingComments = []
-  let obsoleteBlock = null
-
-  const flushObsolete = () => {
-    if (!obsoleteBlock?.length) return
-    const block = obsoleteBlock.join('\n').trimEnd()
-    if (block) obsolete.push(block)
-    obsoleteBlock = null
-  }
-
-  const flush = () => {
-    if (!entry) return
-    const source = String(entry.msgid ?? '').trim()
-    if (source) {
-      const msgidPlural = String(entry.msgid_plural ?? '').trim()
-      const pluralKeys = Object.keys(entry.msgstr ?? {})
-        .filter((key) => /^\d+$/.test(key))
-        .sort((left, right) => Number(left) - Number(right))
-      const plurals = pluralKeys
-        .map((key) => String(entry.msgstr[key] ?? '').trim())
-        .filter(Boolean)
-      const target = String(entry.msgstr?.[''] ?? plurals[0] ?? '').trim()
-      const comments = entry.comments.filter(Boolean)
-      segments.push({
-        source,
-        target,
-        paragraph: segments.length,
-        meta: {
-          format: 'po',
-          context: String(entry.msgctxt ?? '').trim(),
-          msgidPlural,
-          plural: Boolean(msgidPlural),
-          ...(comments.length ? { comments } : {}),
-          ...(plurals.length ? { plurals } : {}),
-        },
-      })
-    }
-    entry = null
-    current = null
-  }
-
-  for (const rawLine of input.split('\n')) {
-    const line = rawLine.trim()
-    if (!line) {
-      flush()
-      flushObsolete()
-      continue
-    }
-    if (line.startsWith('#~')) {
-      if (!obsoleteBlock) {
-        flush()
-        obsoleteBlock = []
-      }
-      obsoleteBlock.push(rawLine.replace(/^#~\s?/, ''))
-      continue
-    }
-    if (obsoleteBlock) {
-      obsoleteBlock.push(rawLine)
-      continue
-    }
-    if (line.startsWith('#')) {
-      const comment = line.replace(/^#[.,:|\s]*/, '').trim()
-      if (comment) {
-        if (entry) entry.comments.push(comment)
-        else pendingComments.push(comment)
-      }
-      continue
-    }
-
-    const field = poField(line)
-    if (field) {
-      entry ??= { msgstr: {}, comments: pendingComments.splice(0) }
-      current = field.name
-      if (current.startsWith('msgstr')) {
-        entry.msgstr[current === 'msgstr' ? '' : current.match(/\[(\d+)\]/)?.[1] ?? '0'] = field.value
-      } else {
-        entry[current] = field.value
-      }
-      continue
-    }
-    if (line.startsWith('"') && current) {
-      const value = parsePoQuoted(line)
-      if (value === null) continue
-      if (current.startsWith('msgstr')) {
-        const key = current === 'msgstr' ? '' : current.match(/\[(\d+)\]/)?.[1] ?? '0'
-        entry.msgstr[key] = `${entry.msgstr[key] ?? ''}${value}`
-      } else {
-        entry[current] = `${entry[current] ?? ''}${value}`
-      }
-    }
-  }
-  flush()
-  flushObsolete()
-
-  if (!segments.length) throw formatError('PO_NO_UNITS', 'PO 中没有可用的 msgid/msgstr 条目。')
-  return { format: 'po', segments, obsolete, warnings: [] }
-}
-
-function poEscape(value) {
-  return String(value ?? '')
-    .replace(/\\/g, '\\\\')
-    .replace(/"/g, '\\"')
-    .replace(/\n/g, '\\n')
-    .replace(/\t/g, '\\t')
-    .replace(/\r/g, '\\r')
-}
-
-function poLine(field, value) {
-  return `${field} "${poEscape(value)}"`
-}
-
-export function buildPo(doc) {
-  const header = [
-    'msgid ""',
-    'msgstr ""',
-    '"Content-Type: text/plain; charset=UTF-8\\n"',
-    `"Language: ${doc.targetLang}\\n"`,
-    '',
-  ].join('\n')
-
-  const entries = doc.segments.map((segment) => {
-    const context = String(segment.meta?.context ?? '').trim()
-    const plural = String(segment.meta?.msgidPlural ?? '').trim()
-    const target = String(segment.target || segment.source || '').trim()
-    const comments = Array.isArray(segment.meta?.comments) ? segment.meta.comments : []
-    const lines = comments.filter(Boolean).map((comment) => `#. ${comment}`)
-    if (context) lines.push(poLine('msgctxt', context))
-    lines.push(poLine('msgid', segment.source))
-    if (plural) {
-      const forms = Array.isArray(segment.meta?.plurals) ? segment.meta.plurals : []
-      lines.push(poLine('msgid_plural', plural))
-      lines.push(poLine('msgstr[0]', target))
-      for (let index = 1; index < Math.max(forms.length, 2); index += 1) {
-        lines.push(poLine(`msgstr[${index}]`, forms[index] || target))
-      }
-    } else {
-      lines.push(poLine('msgstr', target))
-    }
-    return lines.join('\n')
-  })
-
-  const obsolete = Array.isArray(doc.formatMeta?.poObsolete)
-    ? doc.formatMeta.poObsolete
-      .filter((block) => typeof block === 'string' && block.trim())
-      .map((block) => block.split('\n').map((line) => `#~ ${line}`).join('\n'))
-    : []
-
-  const blocks = [...entries, ...obsolete].filter(Boolean)
-  return `${header}\n${blocks.join('\n\n')}\n`
-}
-
-function parseXliffUnit(body, attrs, format, tokens) {
-  const sourceMatch = body.match(/<(?:[\w.-]+:)?source\b[^>]*>([\s\S]*?)<\/(?:[\w.-]+:)?source>/i)
-    ?? body.match(/<(?:[\w.-]+:)?source\b[^>]*\/>/i)
-  if (!sourceMatch) return null
-  const sourceRaw = sourceMatch[1] ?? ''
-  const source = extractInline(sourceRaw)
-  if (!source.text) return null
-  const targetMatch = body.match(/<(?:[\w.-]+:)?target\b[^>]*>([\s\S]*?)<\/(?:[\w.-]+:)?target>/i)
-    ?? body.match(/<(?:[\w.-]+:)?target\b[^>]*\/>/i)
-  const target = targetMatch ? extractInline(targetMatch[1] ?? '').text : ''
-  tokens.push(...source.tokens)
-  return {
-    source: source.text,
-    target,
-    paragraph: 0,
-    meta: {
-      format,
-      xliffId: readXmlAttr(attrs, 'id'),
-      ...(source.tokens.length ? { inline: source.tokens } : {}),
-    },
-  }
-}
-
-function parseXliff2(input, rootAttrs) {
-  const sourceLang = readXmlAttr(rootAttrs, 'srcLang') || readXmlAttr(rootAttrs, 'source-language')
-  const targetLang = readXmlAttr(rootAttrs, 'trgLang') || readXmlAttr(rootAttrs, 'target-language')
-  const units = []
-  const inlineTokens = []
-  const unitRe = /<(?:[\w.-]+:)?unit\b([^>]*)>([\s\S]*?)<\/(?:[\w.-]+:)?unit>/gi
-  let match
-  while ((match = unitRe.exec(input))) {
-    const unit = parseXliffUnit(match[2], match[1], 'xliff2', inlineTokens)
-    if (!unit) continue
-    unit.paragraph = units.length
-    units.push(unit)
-  }
-  if (!units.length) throw formatError('XLIFF_NO_UNITS', 'XLIFF 中没有可用的 unit/segment。')
-  return { format: 'xliff2', sourceLang, targetLang, segments: units, warnings: [] }
-}
-
-export function parseXliff(text) {
-  const input = String(text ?? '').replace(/^\uFEFF/, '')
-  if (!input.trim()) throw formatError('XLIFF_PARSE', 'XLIFF 文件内容为空。')
-
-  const rootTag = input.match(/<((?:[\w.-]+:)?xliff)\b([^>]*)>/i)
-  const rootAttrs = rootTag?.[2] ?? ''
-  const version = readXmlAttr(rootAttrs, 'version')
-  const looksLikeXliff2 = /^2\./i.test(version)
-    || (!/<(?:[\w.-]+:)?trans-unit\b/i.test(input) && /<(?:[\w.-]+:)?unit\b/i.test(input))
-  if (looksLikeXliff2) return parseXliff2(input, rootAttrs)
-
-  const fileTag = input.match(/<((?:[\w.-]+:)?file)\b([^>]*)>/i)
-  const sourceLang = readXmlAttr(fileTag?.[2], 'source-language')
-  const targetLang = readXmlAttr(fileTag?.[2], 'target-language')
-
-  const units = []
-  const unitRe = /<(?:[\w.-]+:)?trans-unit\b([^>]*)>([\s\S]*?)<\/(?:[\w.-]+:)?trans-unit>/gi
-  let match
-  while ((match = unitRe.exec(input))) {
-    const unit = parseXliffUnit(match[2], match[1], 'xliff', [])
-    if (!unit) continue
-    unit.paragraph = units.length
-    units.push(unit)
-  }
-
-  if (!units.length) throw formatError('XLIFF_NO_UNITS', 'XLIFF 中没有可用的 trans-unit。')
-  return { format: 'xliff', sourceLang, targetLang, segments: units, warnings: [] }
-}
-
-export function buildXliff(doc) {
-  const units = doc.segments.map((segment, index) => {
-    const id = String(segment.meta?.xliffId || index + 1)
-    const tokens = Array.isArray(segment.meta?.inline) ? segment.meta.inline : []
-    const target = completeInline(segment.target || segment.source, tokens)
-    return [
-      `      <trans-unit id="${escapeXml(id)}">`,
-      `        <source xml:lang="${escapeXml(doc.sourceLang)}">${renderInline(segment.source, tokens)}</source>`,
-      `        <target xml:lang="${escapeXml(doc.targetLang)}">${renderInline(target, tokens)}</target>`,
-      '      </trans-unit>',
-    ].join('\n')
-  })
-  return [
-    '<?xml version="1.0" encoding="UTF-8"?>',
-    '<xliff version="1.2">',
-    `  <file source-language="${escapeXml(doc.sourceLang)}" target-language="${escapeXml(doc.targetLang)}" datatype="plaintext" original="${escapeXml(doc.name)}">`,
-    '    <body>',
-    ...units,
-    '    </body>',
-    '  </file>',
-    '</xliff>',
-    '',
-  ].join('\n')
-}
-
-export function buildXliff2(doc) {
-  const units = doc.segments.map((segment, index) => {
-    const id = String(segment.meta?.xliffId || index + 1)
-    const tokens = Array.isArray(segment.meta?.inline) ? segment.meta.inline : []
-    const target = completeInline(segment.target || segment.source, tokens)
-    return [
-      `    <unit id="${escapeXml(id)}">`,
-      '      <segment>',
-      `        <source>${renderInline(segment.source, tokens)}</source>`,
-      `        <target>${renderInline(target, tokens)}</target>`,
-      '      </segment>',
-      '    </unit>',
-    ].join('\n')
-  })
-  return [
-    '<?xml version="1.0" encoding="UTF-8"?>',
-    '<xliff xmlns="urn:oasis:names:tc:xliff:document:2.0" version="2.0"',
-    `  srcLang="${escapeXml(doc.sourceLang)}" trgLang="${escapeXml(doc.targetLang)}">`,
-    '  <file id="f1">',
-    ...units,
-    '  </file>',
-    '</xliff>',
-    '',
-  ].join('\n')
-}
-
-function htmlToText(input) {
-  let text = String(input ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
-  text = text
-    .replace(/<!--[\s\S]*?-->/g, '')
-    .replace(/<(script|style|noscript|head)\b[\s\S]*?<\/\1>/gi, '')
-    .replace(/<br\s*\/?>/gi, '\n')
-    .replace(/<\/(?:p|div|li|h[1-6]|section|article|blockquote|pre|tr)>/gi, '\n\n')
-    .replace(/<\/(?:td|th)>/gi, '\t')
-    .replace(/<[^>]+>/g, '')
-  return decodeXml(text)
-    .replace(/[ \t]+\n/g, '\n')
-    .replace(/\n{3,}/g, '\n\n')
-    .trim()
-}
-
-export function parseHtml(text) {
-  const input = String(text ?? '')
-  if (!input.trim()) throw formatError('HTML_PARSE', 'HTML 文件内容为空。')
-  const plain = htmlToText(input)
-  if (!plain) throw formatError('HTML_NO_UNITS', 'HTML 中没有可用的正文。')
-  const segments = plain
-    .split(/\n{2,}/)
-    .map((part) => part.trim())
-    .filter(Boolean)
-    .map((source, index) => ({
-      source,
-      target: '',
-      paragraph: index,
-      meta: { format: 'html' },
-    }))
-  if (!segments.length) throw formatError('HTML_NO_UNITS', 'HTML 中没有可用的正文。')
-  return { format: 'html', segments, warnings: [] }
-}
-
-export function buildHtml(doc) {
-  const groups = new Map()
-  for (const segment of doc.segments) {
-    const key = segment.paragraph ?? 0
-    if (!groups.has(key)) groups.set(key, [])
-    groups.get(key).push(String(segment.target || segment.source || '').trim())
-  }
-  const paragraphs = [...groups.values()]
-    .map((parts) => parts.filter(Boolean).join(' '))
-    .filter(Boolean)
-    .map((text) => `  <p>${escapeXml(text)}</p>`)
-    .join('\n')
-  return [
-    '<!doctype html>',
-    '<html lang="' + escapeXml(doc.targetLang || 'en') + '">',
-    '<head><meta charset="utf-8"><title>' + escapeXml(doc.name || 'CAIT export') + '</title></head>',
-    '<body>',
-    paragraphs,
-    '</body>',
-    '</html>',
-    '',
-  ].join('\n')
-}
-
-function escapeJsonPointer(value) {
-  return String(value).replace(/~/g, '~0').replace(/\//g, '~1')
-}
-
-function unescapeJsonPointer(value) {
-  return String(value).replace(/~1/g, '/').replace(/~0/g, '~')
-}
-
-function walkJsonStrings(value, pointer, visit) {
-  if (typeof value === 'string') {
-    visit(value, pointer)
-    return
-  }
-  if (Array.isArray(value)) {
-    value.forEach((item, index) => walkJsonStrings(item, `${pointer}/${index}`, visit))
-    return
-  }
-  if (value && typeof value === 'object') {
-    for (const [key, item] of Object.entries(value)) {
-      walkJsonStrings(item, `${pointer}/${escapeJsonPointer(key)}`, visit)
-    }
-  }
-}
-
-function setJsonPointer(root, pointer, value) {
-  const parts = String(pointer).split('/').slice(1).map(unescapeJsonPointer)
-  if (!parts.length) return value
-  let current = root
-  for (let index = 0; index < parts.length - 1; index += 1) {
-    if (current == null) throw new Error('JSON path does not exist')
-    current = current[parts[index]]
-  }
-  if (current == null) throw new Error('JSON path does not exist')
-  current[parts[parts.length - 1]] = value
-  return root
-}
-
-export function parseJson(text) {
-  const input = String(text ?? '')
-  if (!input.trim()) throw formatError('JSON_PARSE', 'JSON 文件内容为空。')
-  let root
-  try {
-    root = JSON.parse(input)
-  } catch (error) {
-    throw formatError('JSON_PARSE', `JSON 解析失败: ${error.message}`)
-  }
-  const segments = []
-  walkJsonStrings(root, '', (source, pointer) => {
-    if (!source.trim()) return
-    segments.push({
-      source,
-      target: '',
-      paragraph: segments.length,
-      meta: { format: 'json', jsonPath: pointer },
-    })
-  })
-  if (!segments.length) throw formatError('JSON_NO_UNITS', 'JSON 中没有可翻译的字符串。')
-  return {
-    format: 'json',
-    segments,
-    formatMeta: { jsonTemplate: JSON.stringify(root) },
-    warnings: [],
-  }
-}
-
-export function buildJson(doc) {
-  let root
-  try {
-    root = doc.formatMeta?.jsonTemplate ? JSON.parse(doc.formatMeta.jsonTemplate) : {}
-  } catch {
-    root = {}
-  }
-  for (const segment of doc.segments) {
-    const pointer = segment.meta?.jsonPath
-    if (!pointer) continue
-    const text = String(segment.target || segment.source || '')
-    setJsonPointer(root, pointer, text)
-  }
-  return `${JSON.stringify(root, null, 2)}\n`
-}
-
-const RTF_DESTINATION_GROUP = /\{\\\*?\\?(?:fonttbl|colortbl|stylesheet|info|pict|header|footer|object|themedata|colorschememapping|datastore|latentstyles|rsidtbl|generator|xmlnstbl)\b[^{}]*(?:\{[^{}]*\}[^{}]*)*\}/gi
-
-function rtfUnicode(code) {
-  const value = Number(code)
-  return String.fromCharCode(value < 0 ? value + 65536 : value)
-}
-
-export function parseRtf(text) {
-  let input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
-  if (!input.trim()) throw formatError('RTF_PARSE', 'RTF 文件内容为空。')
-  if (!input.trimStart().startsWith('{\\rtf')) {
-    throw formatError('RTF_PARSE', 'RTF 文件缺少 {\\rtf 头。')
-  }
-  let previous
-  do {
-    previous = input
-    input = input.replace(RTF_DESTINATION_GROUP, '')
-  } while (previous !== input)
-
-  let output = ''
-  let index = 0
-  let unicodeSkip = 1
-  while (index < input.length) {
-    const char = input[index]
-    if (char === '{' || char === '}') {
-      index += 1
-      continue
-    }
-    if (char !== '\\') {
-      output += char
-      index += 1
-      continue
-    }
-    const next = input[index + 1]
-    if (next === '\\' || next === '{' || next === '}') {
-      output += next
-      index += 2
-      continue
-    }
-    if (next === "'") {
-      const hex = input.slice(index + 2, index + 4)
-      if (!/^[0-9a-f]{2}$/i.test(hex)) throw formatError('RTF_PARSE', 'RTF 十六进制转义无效。')
-      output += Buffer.from([Number.parseInt(hex, 16)]).toString('latin1')
-      index += 4
-      continue
-    }
-    if (next === '*') {
-      index += 2
-      continue
-    }
-    let cursor = index + 1
-    let sign = 1
-    if (input[cursor] === '-') {
-      sign = -1
-      cursor += 1
-    }
-    const wordStart = cursor
-    while (/[a-z]/i.test(input[cursor] ?? '')) cursor += 1
-    const word = input.slice(wordStart, cursor).toLowerCase()
-    const numberStart = cursor
-    while (/\d/.test(input[cursor] ?? '')) cursor += 1
-    const number = input.slice(numberStart, cursor)
-    if (input[cursor] === ' ') cursor += 1
-    index = cursor
-    if (!word) continue
-    const value = number ? sign * Number(number) : null
-    if (word === 'par') output += '\n\n'
-    else if (word === 'line') output += '\n'
-    else if (word === 'tab') output += '\t'
-    else if (word === 'uc' && value !== null) unicodeSkip = Math.max(0, value)
-    else if (word === 'u' && value !== null) {
-      output += rtfUnicode(value)
-      let skip = unicodeSkip
-      while (skip > 0 && index < input.length) {
-        index += 1
-        skip -= 1
-      }
-    } else if (word === 'bin' && value !== null && value > 0) {
-      index += value
-    }
-  }
-
-  const segments = output
-    .split(/\n{2,}/)
-    .map((part) => part.replace(/^\n+|\n+$/g, '').trim())
-    .filter(Boolean)
-    .map((source, paragraph) => ({
-      source,
-      target: '',
-      paragraph,
-      meta: { format: 'rtf' },
-    }))
-  if (!segments.length) throw formatError('RTF_NO_UNITS', 'RTF 中没有可用的正文。')
-  return { format: 'rtf', segments, warnings: [] }
-}
-
-function escapeRtf(value) {
-  let output = ''
-  for (const char of String(value)) {
-    const code = char.codePointAt(0)
-    if (char === '\\' || char === '{' || char === '}') output += `\\${char}`
-    else if (char === '\n') output += '\\line '
-    else if (char === '\t') output += '\\tab '
-    else if (code >= 0x20 && code < 0x7f) output += char
-    else if (code <= 0xffff) output += `\\u${code}?`
-    else {
-      const high = Math.floor((code - 0x10000) / 0x400) + 0xd800
-      const low = ((code - 0x10000) % 0x400) + 0xdc00
-      output += `\\u${high}?\\u${low}?`
-    }
-  }
-  return output
-}
-
-export function buildRtf(doc) {
-  const groups = new Map()
-  for (const segment of doc.segments) {
-    const key = segment.paragraph ?? 0
-    if (!groups.has(key)) groups.set(key, [])
-    groups.get(key).push(String(segment.target || segment.source || '').trim())
-  }
-  const body = [...groups.values()]
-    .map((parts) => parts.filter(Boolean).join(' '))
-    .filter(Boolean)
-    .map((text) => `\\pard\\f0\\fs22 ${escapeRtf(text)}\\par`)
-    .join('\n')
-  return `{\\rtf1\\ansi\\deff0{\\fonttbl{\\f0 Calibri;}}\\viewkind4\\uc1\n${body}\n}\n`
-}
-
+/**
+ * Dependency-free parsers/builders for the first-tier CAT exchange formats:
+ * CSV, SRT, WebVTT, PO and a basic XLIFF 1.2 subset.
+ *
+ * These formats are intentionally shallow: inline markup is flattened to text,
+ * complex subtitle styling and XLIFF extensions are not round-tripped. The
+ * original source text is kept so the normal CAT/TM/glossary flow applies.
+ */
+import { CatError } from './errors.js'
+import { decodeXml, escapeXml } from './xml.js'
+
+const SRT_TIME = /^(\d{1,3}):(\d{2}):(\d{2})[,.](\d{1,3})$/
+const VTT_TIME = /^(?:(\d{1,3}):)?(\d{1,2}):(\d{2})\.(\d{1,3})$/
+const INLINE_XML = /<\/?(?:bpt|ept|ph|g|x|bx|ex|mrk|sub|hi|span)\b[^>]*>/gi
+const INLINE_TOKEN = /\{(\d+)\}/g
+
+function formatError(code, message) {
+  return new CatError(code, message)
+}
+
+function readXmlAttr(source, name) {
+  const match = String(source ?? '').match(new RegExp(`\\b${name}\\s*=\\s*["']([^"']*)["']`, 'i'))
+  return match ? decodeXml(match[1]) : ''
+}
+
+/**
+ * Convert XLIFF inline markup to stable {n} tokens while keeping the original
+ * XML for export. This prevents <ph>, <bpt>/<ept> and <g> placeholders from
+ * being silently dropped when a translation is written back.
+ */
+function extractInline(raw) {
+  const tokens = []
+  const text = String(raw ?? '').replace(INLINE_XML, (xml) => {
+    const token = `{${tokens.length + 1}}`
+    tokens.push({ token, xml })
+    return token
+  })
+  return { text: decodeXml(text).trim(), tokens }
+}
+
+/** Render a tokenised string back to XML, escaping only the text portions. */
+function renderInline(value, tokens) {
+  const text = String(value ?? '')
+  const map = new Map((Array.isArray(tokens) ? tokens : []).map((item) => [item.token, item.xml]))
+  let output = ''
+  let last = 0
+  for (const match of text.matchAll(INLINE_TOKEN)) {
+    output += escapeXml(text.slice(last, match.index))
+    output += map.get(match[0]) ?? escapeXml(match[0])
+    last = match.index + match[0].length
+  }
+  output += escapeXml(text.slice(last))
+  return output
+}
+
+/** Tokens present in the source but missing from the target. */
+export function missingInlineTokens(tokens, target) {
+  if (!Array.isArray(tokens) || !tokens.length) return []
+  const value = String(target ?? '')
+  return tokens.map((item) => item.token).filter((token) => !value.includes(token))
+}
+
+/** Append placeholders the translator omitted, so export never silently loses them. */
+function completeInline(value, tokens) {
+  let output = String(value ?? '')
+  for (const item of Array.isArray(tokens) ? tokens : []) {
+    if (!output.includes(item.token)) output += `${output && !/\s$/.test(output) ? ' ' : ''}${item.token}`
+  }
+  return output
+}
+
+export function parseCsvRows(text) {
+  const input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
+  const rows = []
+  let row = []
+  let field = ''
+  let quoted = false
+
+  for (let index = 0; index < input.length; index += 1) {
+    const char = input[index]
+    if (quoted) {
+      if (char === '"') {
+        if (input[index + 1] === '"') {
+          field += '"'
+          index += 1
+        } else {
+          quoted = false
+        }
+      } else {
+        field += char
+      }
+      continue
+    }
+
+    if (char === '"' && field === '') {
+      quoted = true
+    } else if (char === ',') {
+      row.push(field)
+      field = ''
+    } else if (char === '\n') {
+      row.push(field)
+      rows.push(row)
+      row = []
+      field = ''
+    } else {
+      field += char
+    }
+  }
+
+  if (quoted) throw formatError('CSV_PARSE', 'CSV 引号未闭合。')
+  if (field.length || row.length) {
+    row.push(field)
+    rows.push(row)
+  }
+  return rows
+}
+
+export function parseCsv(text, { sourceColumn, targetColumn } = {}) {
+  const rows = parseCsvRows(text)
+  if (!rows.length) throw formatError('CSV_PARSE', 'CSV 中没有可用的行。')
+
+  const normalize = (value) => String(value ?? '').trim().toLowerCase()
+  const sourceAliases = new Set(['source', 'src', 'original', '原文', '源文', '源文本'])
+  const targetAliases = new Set(['target', 'tgt', 'translation', '译文', '目標', '目标', '目标文本'])
+  const first = rows[0].map(normalize)
+  const hasHeader = first.some((value) => sourceAliases.has(value) || targetAliases.has(value))
+  const resolveColumn = (rawValue, aliases, fallback) => {
+    const value = String(rawValue ?? '').trim()
+    if (!value) return fallback
+    if (/^\d+$/.test(value)) {
+      const index = Number(value) - 1
+      if (index < 0) throw formatError('CSV_PARSE', 'CSV 列号从 1 开始。')
+      return index
+    }
+    const index = first.findIndex((item) => item === value.toLowerCase())
+    if (index < 0) throw formatError('CSV_PARSE', `CSV 中找不到列“${value}”。`)
+    return index
+  }
+
+  let sourceIndex = first.findIndex((value) => sourceAliases.has(value))
+  let targetIndex = first.findIndex((value) => targetAliases.has(value))
+  let start = 0
+
+  if (sourceColumn !== undefined || targetColumn !== undefined) {
+    sourceIndex = resolveColumn(sourceColumn, sourceAliases, sourceIndex >= 0 ? sourceIndex : 0)
+    targetIndex = resolveColumn(targetColumn, targetAliases, targetIndex)
+    start = hasHeader ? 1 : 0
+  } else if (sourceIndex >= 0 && targetIndex >= 0) {
+    start = 1
+  } else {
+    sourceIndex = 0
+    targetIndex = rows[0].length > 1 ? 1 : -1
+  }
+
+  const maxColumns = rows.reduce((max, row) => Math.max(max, row.length), 0)
+  if (sourceIndex < 0 || sourceIndex >= maxColumns) {
+    throw formatError('CSV_PARSE', 'CSV 源文列不存在。')
+  }
+  if (targetIndex >= maxColumns) {
+    throw formatError('CSV_PARSE', 'CSV 译文列不存在。')
+  }
+
+  const segments = []
+  for (let index = start; index < rows.length; index += 1) {
+    const row = rows[index]
+    const source = String(row[sourceIndex] ?? '').trim()
+    if (!source) continue
+    const target = targetIndex >= 0 ? String(row[targetIndex] ?? '').trim() : ''
+    segments.push({
+      source,
+      target,
+      paragraph: segments.length,
+      meta: { format: 'csv' },
+    })
+  }
+
+  if (!segments.length) throw formatError('CSV_NO_UNITS', 'CSV 中没有可用的源文行。')
+  return { format: 'csv', segments, warnings: [] }
+}
+
+function parseSrtTime(value) {
+  const match = String(value).trim().match(SRT_TIME)
+  if (!match) return null
+  const [, hours, minutes, seconds, millis] = match
+  return Number(hours) * 3600000 + Number(minutes) * 60000 + Number(seconds) * 1000 + Number(millis.padEnd(3, '0'))
+}
+
+export function formatSrtTime(milliseconds) {
+  const total = Math.max(0, Math.round(Number(milliseconds) || 0))
+  const hours = Math.floor(total / 3600000)
+  const minutes = Math.floor((total % 3600000) / 60000)
+  const seconds = Math.floor((total % 60000) / 1000)
+  const millis = total % 1000
+  return `${String(hours).padStart(2, '0')}:${String(minutes).padStart(2, '0')}:${String(seconds).padStart(2, '0')},${String(millis).padStart(3, '0')}`
+}
+
+function stripSubtitleMarkup(value) {
+  const assOverrides = String(value).match(/\{\\[^}]+\}/g) ?? []
+  return {
+    // Semantic tags such as <i>, <b>, <font> and WebVTT <c>/<v>/<ruby> are
+    // meaningful translation content and stay in the string verbatim.
+    text: decodeXml(String(value).replace(/<br\s*\/?>/gi, '\n').replace(/\{\\[^}]+\}/g, '')),
+    count: assOverrides.length,
+  }
+}
+
+function parseSubtitleBlocks(text, kind) {
+  const input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n').trim()
+  if (!input) throw formatError(`${kind.toUpperCase()}_PARSE`, `${kind.toUpperCase()} 文件内容为空。`)
+
+  const blocks = input.split(/\n{2,}/)
+  const segments = []
+  const warnings = []
+  let markupCount = 0
+
+  for (const block of blocks) {
+    const lines = block.split('\n')
+    if (!lines.length) continue
+    let cursor = 0
+    if (lines[0].trim().toUpperCase().startsWith('WEBVTT')) {
+      lines.shift()
+      while (lines.length && !lines[0].trim()) lines.shift()
+      if (!lines.length) continue
+    }
+    if (lines[0].trim().startsWith('NOTE') || lines[0].trim().startsWith('STYLE') || lines[0].trim().startsWith('REGION')) continue
+
+    let identifier = ''
+    if (!lines[cursor].includes('-->')) {
+      identifier = lines[cursor].trim()
+      cursor += 1
+    }
+    if (!lines[cursor]?.includes('-->')) continue
+    const timing = lines[cursor].trim()
+    cursor += 1
+
+    const timingMatch = timing.match(/^(\S+)\s*-->\s*(\S+)(.*)$/)
+    if (!timingMatch) {
+      throw formatError(`${kind.toUpperCase()}_PARSE`, `字幕时间轴无效: ${timing}`)
+    }
+    const start = kind === 'srt' ? parseSrtTime(timingMatch[1]) : parseVttTime(timingMatch[1])
+    const end = kind === 'srt' ? parseSrtTime(timingMatch[2]) : parseVttTime(timingMatch[2])
+    if (start === null || end === null) {
+      throw formatError(`${kind.toUpperCase()}_PARSE`, `字幕时间轴无效: ${timing}`)
+    }
+    if (end <= start) {
+      throw formatError(
+        `${kind.toUpperCase()}_PARSE`,
+        `字幕结束时间必须晚于开始时间: ${timing}`,
+      )
+    }
+
+    const rawText = lines.slice(cursor).join('\n').trim()
+    if (!rawText) continue
+    const stripped = stripSubtitleMarkup(rawText)
+    markupCount += stripped.count
+    const text = stripped.text.trim()
+    if (!text) continue
+    segments.push({
+      source: text,
+      target: '',
+      paragraph: segments.length,
+      meta: {
+        format: kind,
+        start,
+        end,
+        settings: timingMatch[3].trim(),
+        identifier,
+      },
+    })
+  }
+
+  if (!segments.length) throw formatError(`${kind.toUpperCase()}_NO_UNITS`, `${kind.toUpperCase()} 中没有可用的字幕。`)
+  if (markupCount) {
+    warnings.push({
+      code: `${kind.toUpperCase()}_INLINE_MARKUP_LOST`,
+      count: markupCount,
+      message: `${kind.toUpperCase()} 含 ${markupCount} 个 ASS 样式覆盖标记（{\\...}），导入时已移除；语义标签已保留。`,
+    })
+  }
+  return { format: kind, segments, warnings }
+}
+
+export function parseSrt(text) {
+  return parseSubtitleBlocks(text, 'srt')
+}
+
+function parseVttTime(value) {
+  const match = String(value).trim().match(VTT_TIME)
+  if (!match) return null
+  const [, hours, minutes, seconds, millis] = match
+  return Number(hours || 0) * 3600000 + Number(minutes) * 60000 + Number(seconds) * 1000 + Number(millis.padEnd(3, '0'))
+}
+
+export function formatVttTime(milliseconds) {
+  const total = Math.max(0, Math.round(Number(milliseconds) || 0))
+  const hours = Math.floor(total / 3600000)
+  const minutes = Math.floor((total % 3600000) / 60000)
+  const seconds = Math.floor((total % 60000) / 1000)
+  const millis = total % 1000
+  return `${String(hours).padStart(2, '0')}:${String(minutes).padStart(2, '0')}:${String(seconds).padStart(2, '0')}.${String(millis).padStart(3, '0')}`
+}
+
+export function parseVtt(text) {
+  return parseSubtitleBlocks(text, 'vtt')
+}
+
+function subtitleText(segment) {
+  return String(segment.target || segment.source || '').trim()
+}
+
+function cueTimes(segment, fallbackStart) {
+  const start = Number(segment.meta?.start)
+  const end = Number(segment.meta?.end)
+  if (Number.isFinite(start) && start >= 0 && Number.isFinite(end) && end > start) {
+    return { start, end }
+  }
+  // Text-only documents (and any cue without usable timing metadata) start
+  // where the previous cue ended. This avoids the old index * 2000 fallback
+  // overlapping an explicitly timed cue that sits earlier in the document.
+  const safeStart = Number.isFinite(start) && start >= 0 ? start : fallbackStart
+  return { start: safeStart, end: safeStart + 2000 }
+}
+
+function buildCues(segments) {
+  let cursor = 0
+  return segments.map((segment) => {
+    const times = cueTimes(segment, cursor)
+    cursor = Math.max(cursor, times.end)
+    return { segment, times }
+  })
+}
+
+export function buildSrt(doc) {
+  return buildCues(doc.segments)
+    .map(({ segment, times }, index) => [
+      String(index + 1),
+      `${formatSrtTime(times.start)} --> ${formatSrtTime(times.end)}`,
+      subtitleText(segment),
+      '',
+    ].join('\n'))
+    .join('\n')
+}
+
+export function buildVtt(doc) {
+  const cues = buildCues(doc.segments).map(({ segment, times }) => {
+    const identifier = segment.meta?.identifier ? `${segment.meta.identifier}\n` : ''
+    return `${identifier}${formatVttTime(times.start)} --> ${formatVttTime(times.end)}\n${subtitleText(segment)}\n`
+  })
+  return `WEBVTT\n\n${cues.join('\n')}`
+}
+
+function parsePoQuoted(raw) {
+  const match = String(raw).trim().match(/^"((?:\\.|[^"\\])*)"$/)
+  if (!match) return null
+  return match[1].replace(/\\(n|t|r|"|\\)/g, (all, char) => {
+    if (char === 'n') return '\n'
+    if (char === 't') return '\t'
+    if (char === 'r') return '\r'
+    return char
+  })
+}
+
+function poField(line) {
+  const match = line.match(/^(msgctxt|msgid_plural|msgid|msgstr(?:\[\d+\])?)\s+(.*)$/)
+  if (!match) return null
+  const value = parsePoQuoted(match[2])
+  if (value === null) return null
+  return { name: match[1], value }
+}
+
+export function parsePo(text) {
+  const input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
+  if (!input.trim()) throw formatError('PO_PARSE', 'PO 文件内容为空。')
+
+  const segments = []
+  const obsolete = []
+  let entry = null
+  let current = null
+  let pendingComments = []
+  let obsoleteBlock = null
+
+  const flushObsolete = () => {
+    if (!obsoleteBlock?.length) return
+    const block = obsoleteBlock.join('\n').trimEnd()
+    if (block) obsolete.push(block)
+    obsoleteBlock = null
+  }
+
+  const flush = () => {
+    if (!entry) return
+    const source = String(entry.msgid ?? '').trim()
+    if (source) {
+      const msgidPlural = String(entry.msgid_plural ?? '').trim()
+      const pluralKeys = Object.keys(entry.msgstr ?? {})
+        .filter((key) => /^\d+$/.test(key))
+        .sort((left, right) => Number(left) - Number(right))
+      const plurals = pluralKeys
+        .map((key) => String(entry.msgstr[key] ?? '').trim())
+        .filter(Boolean)
+      const target = String(entry.msgstr?.[''] ?? plurals[0] ?? '').trim()
+      const comments = entry.comments.filter(Boolean)
+      segments.push({
+        source,
+        target,
+        paragraph: segments.length,
+        meta: {
+          format: 'po',
+          context: String(entry.msgctxt ?? '').trim(),
+          msgidPlural,
+          plural: Boolean(msgidPlural),
+          ...(comments.length ? { comments } : {}),
+          ...(plurals.length ? { plurals } : {}),
+        },
+      })
+    }
+    entry = null
+    current = null
+  }
+
+  for (const rawLine of input.split('\n')) {
+    const line = rawLine.trim()
+    if (!line) {
+      flush()
+      flushObsolete()
+      continue
+    }
+    if (line.startsWith('#~')) {
+      if (!obsoleteBlock) {
+        flush()
+        obsoleteBlock = []
+      }
+      obsoleteBlock.push(rawLine.replace(/^#~\s?/, ''))
+      continue
+    }
+    if (obsoleteBlock) {
+      obsoleteBlock.push(rawLine)
+      continue
+    }
+    if (line.startsWith('#')) {
+      const comment = line.replace(/^#[.,:|\s]*/, '').trim()
+      if (comment) {
+        if (entry) entry.comments.push(comment)
+        else pendingComments.push(comment)
+      }
+      continue
+    }
+
+    const field = poField(line)
+    if (field) {
+      entry ??= { msgstr: {}, comments: pendingComments.splice(0) }
+      current = field.name
+      if (current.startsWith('msgstr')) {
+        entry.msgstr[current === 'msgstr' ? '' : current.match(/\[(\d+)\]/)?.[1] ?? '0'] = field.value
+      } else {
+        entry[current] = field.value
+      }
+      continue
+    }
+    if (line.startsWith('"') && current) {
+      const value = parsePoQuoted(line)
+      if (value === null) continue
+      if (current.startsWith('msgstr')) {
+        const key = current === 'msgstr' ? '' : current.match(/\[(\d+)\]/)?.[1] ?? '0'
+        entry.msgstr[key] = `${entry.msgstr[key] ?? ''}${value}`
+      } else {
+        entry[current] = `${entry[current] ?? ''}${value}`
+      }
+    }
+  }
+  flush()
+  flushObsolete()
+
+  if (!segments.length) throw formatError('PO_NO_UNITS', 'PO 中没有可用的 msgid/msgstr 条目。')
+  return { format: 'po', segments, obsolete, warnings: [] }
+}
+
+function poEscape(value) {
+  return String(value ?? '')
+    .replace(/\\/g, '\\\\')
+    .replace(/"/g, '\\"')
+    .replace(/\n/g, '\\n')
+    .replace(/\t/g, '\\t')
+    .replace(/\r/g, '\\r')
+}
+
+function poLine(field, value) {
+  return `${field} "${poEscape(value)}"`
+}
+
+export function buildPo(doc) {
+  const header = [
+    'msgid ""',
+    'msgstr ""',
+    '"Content-Type: text/plain; charset=UTF-8\\n"',
+    `"Language: ${doc.targetLang}\\n"`,
+    '',
+  ].join('\n')
+
+  const entries = doc.segments.map((segment) => {
+    const context = String(segment.meta?.context ?? '').trim()
+    const plural = String(segment.meta?.msgidPlural ?? '').trim()
+    const target = String(segment.target || segment.source || '').trim()
+    const comments = Array.isArray(segment.meta?.comments) ? segment.meta.comments : []
+    const lines = comments.filter(Boolean).map((comment) => `#. ${comment}`)
+    if (context) lines.push(poLine('msgctxt', context))
+    lines.push(poLine('msgid', segment.source))
+    if (plural) {
+      const forms = Array.isArray(segment.meta?.plurals) ? segment.meta.plurals : []
+      lines.push(poLine('msgid_plural', plural))
+      lines.push(poLine('msgstr[0]', target))
+      for (let index = 1; index < Math.max(forms.length, 2); index += 1) {
+        lines.push(poLine(`msgstr[${index}]`, forms[index] || target))
+      }
+    } else {
+      lines.push(poLine('msgstr', target))
+    }
+    return lines.join('\n')
+  })
+
+  const obsolete = Array.isArray(doc.formatMeta?.poObsolete)
+    ? doc.formatMeta.poObsolete
+      .filter((block) => typeof block === 'string' && block.trim())
+      .map((block) => block.split('\n').map((line) => `#~ ${line}`).join('\n'))
+    : []
+
+  const blocks = [...entries, ...obsolete].filter(Boolean)
+  return `${header}\n${blocks.join('\n\n')}\n`
+}
+
+function parseXliffUnit(body, attrs, format, tokens) {
+  const sourceMatch = body.match(/<(?:[\w.-]+:)?source\b[^>]*>([\s\S]*?)<\/(?:[\w.-]+:)?source>/i)
+    ?? body.match(/<(?:[\w.-]+:)?source\b[^>]*\/>/i)
+  if (!sourceMatch) return null
+  const sourceRaw = sourceMatch[1] ?? ''
+  const source = extractInline(sourceRaw)
+  if (!source.text) return null
+  const targetMatch = body.match(/<(?:[\w.-]+:)?target\b[^>]*>([\s\S]*?)<\/(?:[\w.-]+:)?target>/i)
+    ?? body.match(/<(?:[\w.-]+:)?target\b[^>]*\/>/i)
+  const target = targetMatch ? extractInline(targetMatch[1] ?? '').text : ''
+  tokens.push(...source.tokens)
+  return {
+    source: source.text,
+    target,
+    paragraph: 0,
+    meta: {
+      format,
+      xliffId: readXmlAttr(attrs, 'id'),
+      ...(source.tokens.length ? { inline: source.tokens } : {}),
+    },
+  }
+}
+
+function parseXliff2(input, rootAttrs) {
+  const sourceLang = readXmlAttr(rootAttrs, 'srcLang') || readXmlAttr(rootAttrs, 'source-language')
+  const targetLang = readXmlAttr(rootAttrs, 'trgLang') || readXmlAttr(rootAttrs, 'target-language')
+  const units = []
+  const inlineTokens = []
+  const unitRe = /<(?:[\w.-]+:)?unit\b([^>]*)>([\s\S]*?)<\/(?:[\w.-]+:)?unit>/gi
+  let match
+  while ((match = unitRe.exec(input))) {
+    const unit = parseXliffUnit(match[2], match[1], 'xliff2', inlineTokens)
+    if (!unit) continue
+    unit.paragraph = units.length
+    units.push(unit)
+  }
+  if (!units.length) throw formatError('XLIFF_NO_UNITS', 'XLIFF 中没有可用的 unit/segment。')
+  return { format: 'xliff2', sourceLang, targetLang, segments: units, warnings: [] }
+}
+
+export function parseXliff(text) {
+  const input = String(text ?? '').replace(/^\uFEFF/, '')
+  if (!input.trim()) throw formatError('XLIFF_PARSE', 'XLIFF 文件内容为空。')
+
+  const rootTag = input.match(/<((?:[\w.-]+:)?xliff)\b([^>]*)>/i)
+  const rootAttrs = rootTag?.[2] ?? ''
+  const version = readXmlAttr(rootAttrs, 'version')
+  const looksLikeXliff2 = /^2\./i.test(version)
+    || (!/<(?:[\w.-]+:)?trans-unit\b/i.test(input) && /<(?:[\w.-]+:)?unit\b/i.test(input))
+  if (looksLikeXliff2) return parseXliff2(input, rootAttrs)
+
+  const fileTag = input.match(/<((?:[\w.-]+:)?file)\b([^>]*)>/i)
+  const sourceLang = readXmlAttr(fileTag?.[2], 'source-language')
+  const targetLang = readXmlAttr(fileTag?.[2], 'target-language')
+
+  const units = []
+  const unitRe = /<(?:[\w.-]+:)?trans-unit\b([^>]*)>([\s\S]*?)<\/(?:[\w.-]+:)?trans-unit>/gi
+  let match
+  while ((match = unitRe.exec(input))) {
+    const unit = parseXliffUnit(match[2], match[1], 'xliff', [])
+    if (!unit) continue
+    unit.paragraph = units.length
+    units.push(unit)
+  }
+
+  if (!units.length) throw formatError('XLIFF_NO_UNITS', 'XLIFF 中没有可用的 trans-unit。')
+  return { format: 'xliff', sourceLang, targetLang, segments: units, warnings: [] }
+}
+
+export function buildXliff(doc) {
+  const units = doc.segments.map((segment, index) => {
+    const id = String(segment.meta?.xliffId || index + 1)
+    const tokens = Array.isArray(segment.meta?.inline) ? segment.meta.inline : []
+    const target = completeInline(segment.target || segment.source, tokens)
+    return [
+      `      <trans-unit id="${escapeXml(id)}">`,
+      `        <source xml:lang="${escapeXml(doc.sourceLang)}">${renderInline(segment.source, tokens)}</source>`,
+      `        <target xml:lang="${escapeXml(doc.targetLang)}">${renderInline(target, tokens)}</target>`,
+      '      </trans-unit>',
+    ].join('\n')
+  })
+  return [
+    '<?xml version="1.0" encoding="UTF-8"?>',
+    '<xliff version="1.2">',
+    `  <file source-language="${escapeXml(doc.sourceLang)}" target-language="${escapeXml(doc.targetLang)}" datatype="plaintext" original="${escapeXml(doc.name)}">`,
+    '    <body>',
+    ...units,
+    '    </body>',
+    '  </file>',
+    '</xliff>',
+    '',
+  ].join('\n')
+}
+
+export function buildXliff2(doc) {
+  const units = doc.segments.map((segment, index) => {
+    const id = String(segment.meta?.xliffId || index + 1)
+    const tokens = Array.isArray(segment.meta?.inline) ? segment.meta.inline : []
+    const target = completeInline(segment.target || segment.source, tokens)
+    return [
+      `    <unit id="${escapeXml(id)}">`,
+      '      <segment>',
+      `        <source>${renderInline(segment.source, tokens)}</source>`,
+      `        <target>${renderInline(target, tokens)}</target>`,
+      '      </segment>',
+      '    </unit>',
+    ].join('\n')
+  })
+  return [
+    '<?xml version="1.0" encoding="UTF-8"?>',
+    '<xliff xmlns="urn:oasis:names:tc:xliff:document:2.0" version="2.0"',
+    `  srcLang="${escapeXml(doc.sourceLang)}" trgLang="${escapeXml(doc.targetLang)}">`,
+    '  <file id="f1">',
+    ...units,
+    '  </file>',
+    '</xliff>',
+    '',
+  ].join('\n')
+}
+
+function htmlToText(input) {
+  let text = String(input ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
+  text = text
+    .replace(/<!--[\s\S]*?-->/g, '')
+    .replace(/<(script|style|noscript|head)\b[\s\S]*?<\/\1>/gi, '')
+    .replace(/<br\s*\/?>/gi, '\n')
+    .replace(/<\/(?:p|div|li|h[1-6]|section|article|blockquote|pre|tr)>/gi, '\n\n')
+    .replace(/<\/(?:td|th)>/gi, '\t')
+    .replace(/<[^>]+>/g, '')
+  return decodeXml(text)
+    .replace(/[ \t]+\n/g, '\n')
+    .replace(/\n{3,}/g, '\n\n')
+    .trim()
+}
+
+export function parseHtml(text) {
+  const input = String(text ?? '')
+  if (!input.trim()) throw formatError('HTML_PARSE', 'HTML 文件内容为空。')
+  const plain = htmlToText(input)
+  if (!plain) throw formatError('HTML_NO_UNITS', 'HTML 中没有可用的正文。')
+  const segments = plain
+    .split(/\n{2,}/)
+    .map((part) => part.trim())
+    .filter(Boolean)
+    .map((source, index) => ({
+      source,
+      target: '',
+      paragraph: index,
+      meta: { format: 'html' },
+    }))
+  if (!segments.length) throw formatError('HTML_NO_UNITS', 'HTML 中没有可用的正文。')
+  return { format: 'html', segments, warnings: [] }
+}
+
+export function buildHtml(doc) {
+  const groups = new Map()
+  for (const segment of doc.segments) {
+    const key = segment.paragraph ?? 0
+    if (!groups.has(key)) groups.set(key, [])
+    groups.get(key).push(String(segment.target || segment.source || '').trim())
+  }
+  const paragraphs = [...groups.values()]
+    .map((parts) => parts.filter(Boolean).join(' '))
+    .filter(Boolean)
+    .map((text) => `  <p>${escapeXml(text)}</p>`)
+    .join('\n')
+  return [
+    '<!doctype html>',
+    '<html lang="' + escapeXml(doc.targetLang || 'en') + '">',
+    '<head><meta charset="utf-8"><title>' + escapeXml(doc.name || 'CAIT export') + '</title></head>',
+    '<body>',
+    paragraphs,
+    '</body>',
+    '</html>',
+    '',
+  ].join('\n')
+}
+
+function escapeJsonPointer(value) {
+  return String(value).replace(/~/g, '~0').replace(/\//g, '~1')
+}
+
+function unescapeJsonPointer(value) {
+  return String(value).replace(/~1/g, '/').replace(/~0/g, '~')
+}
+
+function walkJsonStrings(value, pointer, visit) {
+  if (typeof value === 'string') {
+    visit(value, pointer)
+    return
+  }
+  if (Array.isArray(value)) {
+    value.forEach((item, index) => walkJsonStrings(item, `${pointer}/${index}`, visit))
+    return
+  }
+  if (value && typeof value === 'object') {
+    for (const [key, item] of Object.entries(value)) {
+      walkJsonStrings(item, `${pointer}/${escapeJsonPointer(key)}`, visit)
+    }
+  }
+}
+
+function setJsonPointer(root, pointer, value) {
+  const parts = String(pointer).split('/').slice(1).map(unescapeJsonPointer)
+  if (!parts.length) return value
+  let current = root
+  for (let index = 0; index < parts.length - 1; index += 1) {
+    if (current == null) throw new Error('JSON path does not exist')
+    current = current[parts[index]]
+  }
+  if (current == null) throw new Error('JSON path does not exist')
+  current[parts[parts.length - 1]] = value
+  return root
+}
+
+export function parseJson(text) {
+  const input = String(text ?? '')
+  if (!input.trim()) throw formatError('JSON_PARSE', 'JSON 文件内容为空。')
+  let root
+  try {
+    root = JSON.parse(input)
+  } catch (error) {
+    throw formatError('JSON_PARSE', `JSON 解析失败: ${error.message}`)
+  }
+  const segments = []
+  walkJsonStrings(root, '', (source, pointer) => {
+    if (!source.trim()) return
+    segments.push({
+      source,
+      target: '',
+      paragraph: segments.length,
+      meta: { format: 'json', jsonPath: pointer },
+    })
+  })
+  if (!segments.length) throw formatError('JSON_NO_UNITS', 'JSON 中没有可翻译的字符串。')
+  return {
+    format: 'json',
+    segments,
+    formatMeta: { jsonTemplate: JSON.stringify(root) },
+    warnings: [],
+  }
+}
+
+export function buildJson(doc) {
+  let root
+  try {
+    root = doc.formatMeta?.jsonTemplate ? JSON.parse(doc.formatMeta.jsonTemplate) : {}
+  } catch {
+    root = {}
+  }
+  for (const segment of doc.segments) {
+    const pointer = segment.meta?.jsonPath
+    if (!pointer) continue
+    const text = String(segment.target || segment.source || '')
+    setJsonPointer(root, pointer, text)
+  }
+  return `${JSON.stringify(root, null, 2)}\n`
+}
+
+const RTF_DESTINATION_GROUP = /\{\\\*?\\?(?:fonttbl|colortbl|stylesheet|info|pict|header|footer|object|themedata|colorschememapping|datastore|latentstyles|rsidtbl|generator|xmlnstbl)\b[^{}]*(?:\{[^{}]*\}[^{}]*)*\}/gi
+
+function rtfUnicode(code) {
+  const value = Number(code)
+  return String.fromCharCode(value < 0 ? value + 65536 : value)
+}
+
+export function parseRtf(text) {
+  let input = String(text ?? '').replace(/^\uFEFF/, '').replace(/\r\n?/g, '\n')
+  if (!input.trim()) throw formatError('RTF_PARSE', 'RTF 文件内容为空。')
+  if (!input.trimStart().startsWith('{\\rtf')) {
+    throw formatError('RTF_PARSE', 'RTF 文件缺少 {\\rtf 头。')
+  }
+  let previous
+  do {
+    previous = input
+    input = input.replace(RTF_DESTINATION_GROUP, '')
+  } while (previous !== input)
+
+  let output = ''
+  let index = 0
+  let unicodeSkip = 1
+  while (index < input.length) {
+    const char = input[index]
+    if (char === '{' || char === '}') {
+      index += 1
+      continue
+    }
+    if (char !== '\\') {
+      output += char
+      index += 1
+      continue
+    }
+    const next = input[index + 1]
+    if (next === '\\' || next === '{' || next === '}') {
+      output += next
+      index += 2
+      continue
+    }
+    if (next === "'") {
+      const hex = input.slice(index + 2, index + 4)
+      if (!/^[0-9a-f]{2}$/i.test(hex)) throw formatError('RTF_PARSE', 'RTF 十六进制转义无效。')
+      output += Buffer.from([Number.parseInt(hex, 16)]).toString('latin1')
+      index += 4
+      continue
+    }
+    if (next === '*') {
+      index += 2
+      continue
+    }
+    let cursor = index + 1
+    let sign = 1
+    if (input[cursor] === '-') {
+      sign = -1
+      cursor += 1
+    }
+    const wordStart = cursor
+    while (/[a-z]/i.test(input[cursor] ?? '')) cursor += 1
+    const word = input.slice(wordStart, cursor).toLowerCase()
+    const numberStart = cursor
+    while (/\d/.test(input[cursor] ?? '')) cursor += 1
+    const number = input.slice(numberStart, cursor)
+    if (input[cursor] === ' ') cursor += 1
+    index = cursor
+    if (!word) continue
+    const value = number ? sign * Number(number) : null
+    if (word === 'par') output += '\n\n'
+    else if (word === 'line') output += '\n'
+    else if (word === 'tab') output += '\t'
+    else if (word === 'uc' && value !== null) unicodeSkip = Math.max(0, value)
+    else if (word === 'u' && value !== null) {
+      output += rtfUnicode(value)
+      let skip = unicodeSkip
+      while (skip > 0 && index < input.length) {
+        index += 1
+        skip -= 1
+      }
+    } else if (word === 'bin' && value !== null && value > 0) {
+      index += value
+    }
+  }
+
+  const segments = output
+    .split(/\n{2,}/)
+    .map((part) => part.replace(/^\n+|\n+$/g, '').trim())
+    .filter(Boolean)
+    .map((source, paragraph) => ({
+      source,
+      target: '',
+      paragraph,
+      meta: { format: 'rtf' },
+    }))
+  if (!segments.length) throw formatError('RTF_NO_UNITS', 'RTF 中没有可用的正文。')
+  return { format: 'rtf', segments, warnings: [] }
+}
+
+function escapeRtf(value) {
+  let output = ''
+  for (const char of String(value)) {
+    const code = char.codePointAt(0)
+    if (char === '\\' || char === '{' || char === '}') output += `\\${char}`
+    else if (char === '\n') output += '\\line '
+    else if (char === '\t') output += '\\tab '
+    else if (code >= 0x20 && code < 0x7f) output += char
+    else if (code <= 0xffff) output += `\\u${code}?`
+    else {
+      const high = Math.floor((code - 0x10000) / 0x400) + 0xd800
+      const low = ((code - 0x10000) % 0x400) + 0xdc00
+      output += `\\u${high}?\\u${low}?`
+    }
+  }
+  return output
+}
+
+export function buildRtf(doc) {
+  const groups = new Map()
+  for (const segment of doc.segments) {
+    const key = segment.paragraph ?? 0
+    if (!groups.has(key)) groups.set(key, [])
+    groups.get(key).push(String(segment.target || segment.source || '').trim())
+  }
+  const body = [...groups.values()]
+    .map((parts) => parts.filter(Boolean).join(' '))
+    .filter(Boolean)
+    .map((text) => `\\pard\\f0\\fs22 ${escapeRtf(text)}\\par`)
+    .join('\n')
+  return `{\\rtf1\\ansi\\deff0{\\fonttbl{\\f0 Calibri;}}\\viewkind4\\uc1\n${body}\n}\n`
+}
+