dsh-cait 1.0.3 → 1.0.4

package/lib/pptx.js

@@ -1,89 +1,89 @@
-import { readZipEntries, writeZip } from './docx.js'
-import { CatError } from './errors.js'
-import { decodeXml, escapeXml } from './xml.js'
-
-export const PPTX_MIME = 'application/vnd.openxmlformats-officedocument.presentationml.presentation'
-
-function fail(code, message) {
-  return new CatError(code, message)
-}
-
-function slideNames(entries) {
-  return entries
-    .map((entry) => entry.name)
-    .filter((name) => /^ppt\/slides\/slide\d+\.xml$/i.test(name))
-    .sort((left, right) => Number(left.match(/(\d+)\.xml$/)[1]) - Number(right.match(/(\d+)\.xml$/)[1]))
-}
-
-function paragraphText(paragraph) {
-  return [...String(paragraph).matchAll(/<a:t\b[^>]*>([\s\S]*?)<\/a:t>/gi)]
-    .map((match) => decodeXml(match[1]))
-    .join('')
-}
-
-export function parsePptxDetailed(buffer) {
-  let entries
-  try {
-    entries = readZipEntries(buffer)
-  } catch (error) {
-    throw fail('PPTX_PARSE', `PPTX 文件不是有效的 ZIP/PowerPoint 包: ${error.message}`)
-  }
-  const names = slideNames(entries)
-  if (!names.length) throw fail('PPTX_PARSE', 'PPTX 中找不到幻灯片。')
-  const segments = []
-  for (const name of names) {
-    const xml = entries.find((entry) => entry.name === name)?.data.toString('utf8') ?? ''
-    let index = 0
-    for (const match of xml.matchAll(/<a:p\b[^>]*>([\s\S]*?)<\/a:p>/gi)) {
-      const source = paragraphText(match[1]).trim()
-      const slideParagraph = index
-      index += 1
-      if (!source) continue
-      segments.push({
-        source,
-        target: '',
-        paragraph: segments.length,
-        meta: { format: 'pptx', pptxSlide: name, pptxParagraph: slideParagraph },
-      })
-    }
-  }
-  if (!segments.length) throw fail('PPTX_NO_UNITS', 'PPTX 中没有可用的正文文本。')
-  return { format: 'pptx', segments, warnings: [] }
-}
-
-function patchParagraph(paragraph, text) {
-  let replaced = false
-  return String(paragraph).replace(/<a:t\b([^>]*)>([\s\S]*?)<\/a:t>/gi, (all, attrs) => {
-    if (replaced) return `<a:t${attrs}></a:t>`
-    replaced = true
-    return `<a:t${attrs}>${escapeXml(text)}</a:t>`
-  })
-}
-
-function patchSlideXml(xml, replacements) {
-  let index = 0
-  return String(xml).replace(/<a:p\b[^>]*>[\s\S]*?<\/a:p>/gi, (paragraph) => {
-    const current = index
-    index += 1
-    if (!replacements.has(current)) return paragraph
-    return patchParagraph(paragraph, replacements.get(current))
-  })
-}
-
-export function buildPptx(template, replacements) {
-  let entries
-  try {
-    entries = readZipEntries(template)
-  } catch (error) {
-    throw fail('PPTX_TEMPLATE_INVALID', `PPTX 源模板损坏: ${error.message}`)
-  }
-  const patched = entries.map((entry) => {
-    const byParagraph = replacements.get(entry.name)
-    if (!byParagraph) return entry
-    return {
-      ...entry,
-      data: Buffer.from(patchSlideXml(entry.data.toString('utf8'), byParagraph), 'utf8'),
-    }
-  })
-  return writeZip(patched)
-}
+import { readZipEntries, writeZip } from './docx.js'
+import { CatError } from './errors.js'
+import { decodeXml, escapeXml } from './xml.js'
+
+export const PPTX_MIME = 'application/vnd.openxmlformats-officedocument.presentationml.presentation'
+
+function fail(code, message) {
+  return new CatError(code, message)
+}
+
+function slideNames(entries) {
+  return entries
+    .map((entry) => entry.name)
+    .filter((name) => /^ppt\/slides\/slide\d+\.xml$/i.test(name))
+    .sort((left, right) => Number(left.match(/(\d+)\.xml$/)[1]) - Number(right.match(/(\d+)\.xml$/)[1]))
+}
+
+function paragraphText(paragraph) {
+  return [...String(paragraph).matchAll(/<a:t\b[^>]*>([\s\S]*?)<\/a:t>/gi)]
+    .map((match) => decodeXml(match[1]))
+    .join('')
+}
+
+export function parsePptxDetailed(buffer) {
+  let entries
+  try {
+    entries = readZipEntries(buffer)
+  } catch (error) {
+    throw fail('PPTX_PARSE', `PPTX 文件不是有效的 ZIP/PowerPoint 包: ${error.message}`)
+  }
+  const names = slideNames(entries)
+  if (!names.length) throw fail('PPTX_PARSE', 'PPTX 中找不到幻灯片。')
+  const segments = []
+  for (const name of names) {
+    const xml = entries.find((entry) => entry.name === name)?.data.toString('utf8') ?? ''
+    let index = 0
+    for (const match of xml.matchAll(/<a:p\b[^>]*>([\s\S]*?)<\/a:p>/gi)) {
+      const source = paragraphText(match[1]).trim()
+      const slideParagraph = index
+      index += 1
+      if (!source) continue
+      segments.push({
+        source,
+        target: '',
+        paragraph: segments.length,
+        meta: { format: 'pptx', pptxSlide: name, pptxParagraph: slideParagraph },
+      })
+    }
+  }
+  if (!segments.length) throw fail('PPTX_NO_UNITS', 'PPTX 中没有可用的正文文本。')
+  return { format: 'pptx', segments, warnings: [] }
+}
+
+function patchParagraph(paragraph, text) {
+  let replaced = false
+  return String(paragraph).replace(/<a:t\b([^>]*)>([\s\S]*?)<\/a:t>/gi, (all, attrs) => {
+    if (replaced) return `<a:t${attrs}></a:t>`
+    replaced = true
+    return `<a:t${attrs}>${escapeXml(text)}</a:t>`
+  })
+}
+
+function patchSlideXml(xml, replacements) {
+  let index = 0
+  return String(xml).replace(/<a:p\b[^>]*>[\s\S]*?<\/a:p>/gi, (paragraph) => {
+    const current = index
+    index += 1
+    if (!replacements.has(current)) return paragraph
+    return patchParagraph(paragraph, replacements.get(current))
+  })
+}
+
+export function buildPptx(template, replacements) {
+  let entries
+  try {
+    entries = readZipEntries(template)
+  } catch (error) {
+    throw fail('PPTX_TEMPLATE_INVALID', `PPTX 源模板损坏: ${error.message}`)
+  }
+  const patched = entries.map((entry) => {
+    const byParagraph = replacements.get(entry.name)
+    if (!byParagraph) return entry
+    return {
+      ...entry,
+      data: Buffer.from(patchSlideXml(entry.data.toString('utf8'), byParagraph), 'utf8'),
+    }
+  })
+  return writeZip(patched)
+}

package/lib/segmenter.js

@@ -1,84 +1,84 @@
-/**
- * Paragraph-aware sentence segmentation.
- *
- * Paragraphs are separated by blank lines. Within a paragraph we split on
- * sentence-final punctuation while protecting decimals and a small set of
- * common abbreviations. CJK text usually has no space after the full stop,
- * so a boundary is also recognised when the next character is CJK.
- */
-
-const CJK = /[\u3400-\u9fff\u3040-\u30ff\uac00-\ud7af]/
-
-function isAbbreviation(text, dotIndex) {
-  const before = text.slice(Math.max(0, dotIndex - 12), dotIndex)
-  return /(?:^|[\s(])(?:e\.g|i\.e|etc|Mr|Mrs|Ms|Dr|Prof|vs|No)$/i.test(before)
-}
-
-function splitSentences(text) {
-  const out = []
-  let start = 0
-
-  for (let i = 0; i < text.length; i += 1) {
-    const char = text[i]
-    const isCjkStop = char === '。' || char === '！' || char === '？' || char === '…'
-    const isLatinStop = char === '.' || char === '!' || char === '?'
-    if (!isCjkStop && !isLatinStop) continue
-
-    if (isCjkStop && char === '…' && text[i + 1] === '…') continue
-    if (isLatinStop && char === '.') {
-      const previous = text[i - 1] || ''
-      const next = text[i + 1] || ''
-      if (/\d/.test(previous) && /\d/.test(next)) continue
-      if (isAbbreviation(text, i)) continue
-    }
-
-    let end = i + 1
-    while (end < text.length && /["'”’）)\]]/.test(text[end])) end += 1
-
-    if (isLatinStop) {
-      const next = text[end] || ''
-      if (!/\s/.test(next) && !CJK.test(next)) continue
-    }
-
-    const part = text.slice(start, end).trim()
-    if (part) out.push(part)
-    start = end
-    i = end - 1
-  }
-
-  const tail = text.slice(start).trim()
-  if (tail) out.push(tail)
-  return out
-}
-
-export function splitIntoSegments(input) {
-  const text = String(input ?? '').replace(/\r\n?/g, '\n').trim()
-  if (!text) return []
-
-  const segments = []
-  let paragraph = 0
-
-  for (const rawBlock of text.split(/\n{2,}/)) {
-    const block = rawBlock.trim()
-    if (!block) continue
-
-    if (/^#{1,6}\s+/.test(block)) {
-      const heading = block
-        .replace(/^#{1,6}\s+/, '')
-        .replace(/\s+#+\s*$/, '')
-        .replace(/\s*\n\s*/g, ' ')
-        .trim()
-      if (heading) segments.push({ source: heading, paragraph })
-      paragraph += 1
-      continue
-    }
-
-    const joined = block.replace(/\s*\n\s*/g, ' ').replace(/[ \t]{2,}/g, ' ').trim()
-    for (const sentence of splitSentences(joined)) {
-      segments.push({ source: sentence, paragraph })
-    }
-    paragraph += 1
-  }
-
-  return segments
-}
+/**
+ * Paragraph-aware sentence segmentation.
+ *
+ * Paragraphs are separated by blank lines. Within a paragraph we split on
+ * sentence-final punctuation while protecting decimals and a small set of
+ * common abbreviations. CJK text usually has no space after the full stop,
+ * so a boundary is also recognised when the next character is CJK.
+ */
+
+const CJK = /[\u3400-\u9fff\u3040-\u30ff\uac00-\ud7af]/
+
+function isAbbreviation(text, dotIndex) {
+  const before = text.slice(Math.max(0, dotIndex - 12), dotIndex)
+  return /(?:^|[\s(])(?:e\.g|i\.e|etc|Mr|Mrs|Ms|Dr|Prof|vs|No)$/i.test(before)
+}
+
+function splitSentences(text) {
+  const out = []
+  let start = 0
+
+  for (let i = 0; i < text.length; i += 1) {
+    const char = text[i]
+    const isCjkStop = char === '。' || char === '！' || char === '？' || char === '…'
+    const isLatinStop = char === '.' || char === '!' || char === '?'
+    if (!isCjkStop && !isLatinStop) continue
+
+    if (isCjkStop && char === '…' && text[i + 1] === '…') continue
+    if (isLatinStop && char === '.') {
+      const previous = text[i - 1] || ''
+      const next = text[i + 1] || ''
+      if (/\d/.test(previous) && /\d/.test(next)) continue
+      if (isAbbreviation(text, i)) continue
+    }
+
+    let end = i + 1
+    while (end < text.length && /["'”’）)\]]/.test(text[end])) end += 1
+
+    if (isLatinStop) {
+      const next = text[end] || ''
+      if (!/\s/.test(next) && !CJK.test(next)) continue
+    }
+
+    const part = text.slice(start, end).trim()
+    if (part) out.push(part)
+    start = end
+    i = end - 1
+  }
+
+  const tail = text.slice(start).trim()
+  if (tail) out.push(tail)
+  return out
+}
+
+export function splitIntoSegments(input) {
+  const text = String(input ?? '').replace(/\r\n?/g, '\n').trim()
+  if (!text) return []
+
+  const segments = []
+  let paragraph = 0
+
+  for (const rawBlock of text.split(/\n{2,}/)) {
+    const block = rawBlock.trim()
+    if (!block) continue
+
+    if (/^#{1,6}\s+/.test(block)) {
+      const heading = block
+        .replace(/^#{1,6}\s+/, '')
+        .replace(/\s+#+\s*$/, '')
+        .replace(/\s*\n\s*/g, ' ')
+        .trim()
+      if (heading) segments.push({ source: heading, paragraph })
+      paragraph += 1
+      continue
+    }
+
+    const joined = block.replace(/\s*\n\s*/g, ' ').replace(/[ \t]{2,}/g, ' ').trim()
+    for (const sentence of splitSentences(joined)) {
+      segments.push({ source: sentence, paragraph })
+    }
+    paragraph += 1
+  }
+
+  return segments
+}

package/lib/similarity.js

@@ -1,113 +1,113 @@
-/**
- * Text normalisation + fuzzy similarity used by the translation memory.
- * Deliberately dependency-free: character Dice gives robust short-segment
- * scoring, token Jaccard keeps word order changes from overstating a match.
- */
-
-export function normalizeText(value) {
-  return String(value ?? '')
-    .normalize('NFKC')
-    .replace(/\s+/g, ' ')
-    .trim()
-    .toLowerCase()
-}
-
-export function normalizeSource(value) {
-  return String(value ?? '')
-    .normalize('NFKC')
-    .replace(/\s+/g, ' ')
-    .trim()
-}
-
-/**
- * Near-exact comparison key: numbers, punctuation and whitespace are ignored.
- * This is intentionally separate from `similarity()`: punctuation-only edits can
- * score 1.0 while a digit-only edit can score below 0.7, so no score cutoff can
- * identify this class reliably.
- */
-export function nearExactKey(value) {
-  return String(value ?? '')
-    .normalize('NFKC')
-    .toLowerCase()
-    .replace(/[\p{N}\p{P}\s]+/gu, '')
-}
-
-/**
- * memoQ-style match grades for TM results.
- *
- * Boundaries are based on the observed score distribution:
- *   - one word / word-order edits land around 0.85-0.93;
- *   - a changed word or inflection lands around 0.75-0.85;
- *   - larger rewrites fall below 0.75.
- * The default TM retrieval floor is 0.55, so low_fuzzy is normally the
- * 0.55-0.7499 band rather than the whole range.
- *
- * `near_exact` is gated by `nearExactKey()` equality, never by `score`.
- */
-export function classifyMatchType({ exact = false, score = 0, query = '', source = '' } = {}) {
-  if (exact) return 'exact'
-  const queryKey = nearExactKey(query)
-  if (queryKey && queryKey === nearExactKey(source)) return 'near_exact'
-  if (score >= 0.85) return 'high_fuzzy'
-  if (score >= 0.75) return 'medium_fuzzy'
-  return 'low_fuzzy'
-}
-
-function bigrams(value) {
-  const chars = Array.from(value.replace(/[\s\p{P}\p{S}]+/gu, ''))
-  if (chars.length <= 1) return chars
-  const out = []
-  for (let i = 0; i < chars.length - 1; i += 1) out.push(chars[i] + chars[i + 1])
-  return out
-}
-
-function diceCoefficient(a, b) {
-  const left = bigrams(a)
-  const right = bigrams(b)
-  if (!left.length || !right.length) return a === b ? 1 : 0
-  const counts = new Map()
-  for (const gram of left) counts.set(gram, (counts.get(gram) || 0) + 1)
-  let overlap = 0
-  for (const gram of right) {
-    const count = counts.get(gram) || 0
-    if (count > 0) {
-      overlap += 1
-      counts.set(gram, count - 1)
-    }
-  }
-  return (2 * overlap) / (left.length + right.length)
-}
-
-function tokenSet(value) {
-  const latin = value.match(/[a-z0-9]+(?:['’-][a-z0-9]+)*/g) || []
-  const cjk = value.match(/[\u3400-\u9fff\u3040-\u30ff\uac00-\ud7af]/g) || []
-  return new Set([...latin, ...cjk])
-}
-
-function jaccard(a, b) {
-  const left = tokenSet(a)
-  const right = tokenSet(b)
-  if (!left.size || !right.size) return 0
-  let intersection = 0
-  for (const token of left) if (right.has(token)) intersection += 1
-  return intersection / (left.size + right.size - intersection)
-}
-
-export function similarity(a, b) {
-  const left = normalizeText(a)
-  const right = normalizeText(b)
-  if (!left || !right) return 0
-  if (left === right) return 1
-  const score = 0.72 * diceCoefficient(left, right) + 0.28 * jaccard(left, right)
-  return Math.round(Math.min(1, Math.max(0, score)) * 10000) / 10000
-}
-
-export function clamp(value, min, max) {
-  return Math.min(max, Math.max(min, value))
-}
-
-export function clampInt(value, min, max, fallback) {
-  const number = Number(value)
-  if (!Number.isFinite(number)) return fallback
-  return Math.round(clamp(number, min, max))
-}
+/**
+ * Text normalisation + fuzzy similarity used by the translation memory.
+ * Deliberately dependency-free: character Dice gives robust short-segment
+ * scoring, token Jaccard keeps word order changes from overstating a match.
+ */
+
+export function normalizeText(value) {
+  return String(value ?? '')
+    .normalize('NFKC')
+    .replace(/\s+/g, ' ')
+    .trim()
+    .toLowerCase()
+}
+
+export function normalizeSource(value) {
+  return String(value ?? '')
+    .normalize('NFKC')
+    .replace(/\s+/g, ' ')
+    .trim()
+}
+
+/**
+ * Near-exact comparison key: numbers, punctuation and whitespace are ignored.
+ * This is intentionally separate from `similarity()`: punctuation-only edits can
+ * score 1.0 while a digit-only edit can score below 0.7, so no score cutoff can
+ * identify this class reliably.
+ */
+export function nearExactKey(value) {
+  return String(value ?? '')
+    .normalize('NFKC')
+    .toLowerCase()
+    .replace(/[\p{N}\p{P}\s]+/gu, '')
+}
+
+/**
+ * memoQ-style match grades for TM results.
+ *
+ * Boundaries are based on the observed score distribution:
+ *   - one word / word-order edits land around 0.85-0.93;
+ *   - a changed word or inflection lands around 0.75-0.85;
+ *   - larger rewrites fall below 0.75.
+ * The default TM retrieval floor is 0.55, so low_fuzzy is normally the
+ * 0.55-0.7499 band rather than the whole range.
+ *
+ * `near_exact` is gated by `nearExactKey()` equality, never by `score`.
+ */
+export function classifyMatchType({ exact = false, score = 0, query = '', source = '' } = {}) {
+  if (exact) return 'exact'
+  const queryKey = nearExactKey(query)
+  if (queryKey && queryKey === nearExactKey(source)) return 'near_exact'
+  if (score >= 0.85) return 'high_fuzzy'
+  if (score >= 0.75) return 'medium_fuzzy'
+  return 'low_fuzzy'
+}
+
+function bigrams(value) {
+  const chars = Array.from(value.replace(/[\s\p{P}\p{S}]+/gu, ''))
+  if (chars.length <= 1) return chars
+  const out = []
+  for (let i = 0; i < chars.length - 1; i += 1) out.push(chars[i] + chars[i + 1])
+  return out
+}
+
+function diceCoefficient(a, b) {
+  const left = bigrams(a)
+  const right = bigrams(b)
+  if (!left.length || !right.length) return a === b ? 1 : 0
+  const counts = new Map()
+  for (const gram of left) counts.set(gram, (counts.get(gram) || 0) + 1)
+  let overlap = 0
+  for (const gram of right) {
+    const count = counts.get(gram) || 0
+    if (count > 0) {
+      overlap += 1
+      counts.set(gram, count - 1)
+    }
+  }
+  return (2 * overlap) / (left.length + right.length)
+}
+
+function tokenSet(value) {
+  const latin = value.match(/[a-z0-9]+(?:['’-][a-z0-9]+)*/g) || []
+  const cjk = value.match(/[\u3400-\u9fff\u3040-\u30ff\uac00-\ud7af]/g) || []
+  return new Set([...latin, ...cjk])
+}
+
+function jaccard(a, b) {
+  const left = tokenSet(a)
+  const right = tokenSet(b)
+  if (!left.size || !right.size) return 0
+  let intersection = 0
+  for (const token of left) if (right.has(token)) intersection += 1
+  return intersection / (left.size + right.size - intersection)
+}
+
+export function similarity(a, b) {
+  const left = normalizeText(a)
+  const right = normalizeText(b)
+  if (!left || !right) return 0
+  if (left === right) return 1
+  const score = 0.72 * diceCoefficient(left, right) + 0.28 * jaccard(left, right)
+  return Math.round(Math.min(1, Math.max(0, score)) * 10000) / 10000
+}
+
+export function clamp(value, min, max) {
+  return Math.min(max, Math.max(min, value))
+}
+
+export function clampInt(value, min, max, fallback) {
+  const number = Number(value)
+  if (!Number.isFinite(number)) return fallback
+  return Math.round(clamp(number, min, max))
+}