dsh-think-translate 1.2.19 → 1.2.21

package/lib/runtime-state.js

@@ -1,25 +1,29 @@
-import { createHash } from 'node:crypto'
-
-/** Revision of translation-affecting settings, without disclosing key values. */
-export function translationNamespace(config, credentialRevision = 0) {
-  const providers = Object.keys(config.providers || {}).sort().map(id => {
-    const p = config.providers[id]
-    return [id, p.type, p.enabled, p.source, p.baseURL, p.model, p.apiKeyEnv, p.apiKeyRef]
-  })
-  return createHash('sha256').update(JSON.stringify(['protected-v3-glossary', config.chain, config.fallback, providers, config.glossary, credentialRevision])).digest('hex')
-}
-
-/** Effective Loader state, not mere installation, determines mutual exclusion. */
-export function officialTranslationState(ctx) {
-  const loader = ctx?.get?.('loader')
-  if (loader?.entries) {
-    const official = new Set(['@deepseek-ai/dsh-experimental-client-ui-cot-translation'])
-    for (const entry of loader.entries()) {
-      if (official.has(entry.options?.name) && !entry.disabled && ![3, 4, 5].includes(entry.fiber?.state)) {
-        return { officialEnabled: true, conflictDetection: 'available' }
-      }
-    }
-    return { officialEnabled: !!ctx.get('cotTranslation'), conflictDetection: 'available' }
-  }
-  return { officialEnabled: !!ctx?.get?.('cotTranslation'), conflictDetection: 'unavailable' }
-}
+import { createHash } from 'node:crypto'
+import { BUILTIN_GLOSSARY } from './glossary-presets.js'
+
+const builtinRevision = createHash('sha256').update(JSON.stringify(BUILTIN_GLOSSARY)).digest('hex')
+
+/** Revision of translation-affecting settings, without disclosing key values. */
+export function translationNamespace(config, credentialRevision = 0) {
+  const providers = Object.keys(config.providers || {}).sort().map(id => {
+    const p = config.providers[id]
+    return [id, p.type, p.enabled, p.source, p.baseURL, p.model, p.apiKeyEnv, p.apiKeyRef]
+  })
+  return createHash('sha256').update(JSON.stringify(['protected-v4-independent-glossary', config.chain, config.fallback, providers, config.glossary,
+    config.glossary?.builtin === true ? builtinRevision : null, credentialRevision])).digest('hex')
+}
+
+/** Effective Loader state, not mere installation, determines mutual exclusion. */
+export function officialTranslationState(ctx) {
+  const loader = ctx?.get?.('loader')
+  if (loader?.entries) {
+    const official = new Set(['@deepseek-ai/dsh-experimental-client-ui-cot-translation'])
+    for (const entry of loader.entries()) {
+      if (official.has(entry.options?.name) && !entry.disabled && ![3, 4, 5].includes(entry.fiber?.state)) {
+        return { officialEnabled: true, conflictDetection: 'available' }
+      }
+    }
+    return { officialEnabled: !!ctx.get('cotTranslation'), conflictDetection: 'available' }
+  }
+  return { officialEnabled: !!ctx?.get?.('cotTranslation'), conflictDetection: 'unavailable' }
+}

package/lib/text.js

@@ -1,111 +1,111 @@
-// Code spans keep their exact source offsets so neither translation nor batching
-// needs placeholders that a provider could translate, remove, or reorder.
-export function protectedTextParts(text) {
-  const parts = []
-  let proseStart = 0
-  let cursor = 0
-  function protect(start, end, complete) {
-    if (start > proseStart) parts.push({ text: text.slice(proseStart, start), start: proseStart, end: start, code: false, complete: true })
-    parts.push({ text: text.slice(start, end), start, end, code: true, complete })
-    proseStart = end
-    cursor = end
-  }
-  function lineEnd(start) {
-    const newline = text.indexOf('\n', start)
-    return newline < 0 ? text.length : newline + 1
-  }
-  function technicalSpan(start) {
-    const before = text[start - 1] || ''
-    if (before && /[A-Za-z0-9_\\/]/.test(before)) return 0
-    const rest = text.slice(start)
-    // Quotes are part of a path token only when its prefix is unambiguous.
-    const quoted = rest.match(/^("|'|“)(?:[A-Za-z]:[\\/]|\\\\|~?\/|\.\.?\/)/)
-    if (quoted) {
-      const close = quoted[1] === '“' ? '”' : quoted[1]
-      const end = text.indexOf(close, start + 1)
-      return end < 0 ? text.length - start : end + 1 - start
-    }
-    const token = rest.match(/^(?:https?:\/\/|file:\/\/|[A-Za-z]:[\\/]|\\\\|~?\/|\.\.?\/)[^\s<>"'`，。！？；：、]+/)
-      || rest.match(/^[\w@.-]+\.(?:[cm]?js|jsx|tsx?|jsonl?|ya?ml|toml|ini|env|md|py|sh|ps1|cmd|bat|exe|lock|go|rs|cpp|sql|html|css|txt|zstd|png|jpe?g|svg)(?![\w.])/i)
-    if (token) {
-      let value = token[0].replace(/[.,;:!?，。！？；：、]+$/, '')
-      // Keep balanced URL parentheses, but leave Markdown/prose delimiters out.
-      for (const [open, close] of [['(', ')'], ['[', ']'], ['{', '}']]) {
-        while (value.endsWith(close) && value.split(close).length > value.split(open).length) value = value.slice(0, -1)
-      }
-      return value.length
-    }
-    const command = rest.match(/^(?:(?:npm|pnpm|yarn)\s+(?:run\s+[\w:.-]+|test|ci|install|build|start)|git\s+(?:status|diff|log|fetch|pull|push|show|branch)|(?:python3?|node)\s+[\w./\\:-]+\.(?:py|[cm]?js))(?=$|[\s.,;:!?，。！？；：、)]|--)(?:[ \t]+(?:--?[\w-]+(?:=[\w./\\:@=-]+)?|--|"[^"\r\n]*"|'[^'\r\n]*'))*/)
-    return command ? command[0].length : 0
-  }
-  while (cursor < text.length) {
-    if (cursor === 0 || text[cursor - 1] === '\n') {
-      const end = lineEnd(cursor)
-      const line = text.slice(cursor, end)
-      // Treat indented/quoted fences conservatively as code as well.
-      const fence = line.match(/^[\t ]*(?:>[\t ]*)*(?:(?:[-+*]|[0-9]+[.)])[\t ]+)?(`{3,}|~{3,})[^\r\n]*(?:\r?\n)?$/)
-      if (fence) {
-        const close = new RegExp('^[\\t ]*(?:>[\\t ]*)*' + fence[1][0] + '{' + fence[1].length + ',}[\\t ]*(?:\\r?\\n)?$')
-        let stop = end
-        let complete = false
-        while (stop < text.length) {
-          const next = lineEnd(stop)
-          if (close.test(text.slice(stop, next))) { stop = next; complete = true; break }
-          stop = next
-        }
-        protect(cursor, stop, complete)
-        continue
-      }
-      if (/^(?: {4}|\t)/.test(line)) {
-        protect(cursor, end, end < text.length || line.endsWith('\n'))
-        continue
-      }
-      const shellLine = /^(?:\$\s*)?(?:(?:npm|pnpm|yarn)\s+(?:run|test|ci|install|add|remove|uninstall|build|start)\b|git\s+(?:status|diff|log|fetch|pull|push|show|branch|checkout|switch|commit|add)\b|(?:curl|wget)\s+-)/.test(line)
-      if (shellLine && !/\b(?:and|then|before|after|because|please|shows|prints|returns|failed|succeeded|should|will|can|is|are)\b/.test(line)) {
-        protect(cursor, end, end < text.length || line.endsWith('\n'))
-        continue
-      }
-    }
-    if (text[cursor] === '`') {
-      let slashes = 0
-      for (let i = cursor - 1; i >= 0 && text[i] === '\\'; i--) slashes++
-      if (slashes % 2) { cursor++; continue }
-      let after = cursor + 1
-      while (text[after] === '`') after++
-      const size = after - cursor
-      let next = after
-      let stop = text.length
-      let complete = false
-      while ((next = text.indexOf('`', next)) >= 0) {
-        let end = next + 1
-        while (text[end] === '`') end++
-        if (end - next === size) { stop = end; complete = true; break }
-        next = end
-      }
-      // An unfinished streaming code span must not expose its body as prose.
-      protect(cursor, stop, complete)
-      continue
-    }
-    const technicalLength = technicalSpan(cursor)
-    if (technicalLength) {
-      protect(cursor, cursor + technicalLength, true)
-      continue
-    }
-    cursor++
-  }
-  if (proseStart < text.length) parts.push({ text: text.slice(proseStart), start: proseStart, end: text.length, code: false, complete: true })
-  return parts
-}
-
-// Only prose reaches a provider; separators adjoining code stay byte-for-byte.
-export async function translateProtectedText(text, translate, split = value => [value]) {
-  const output = []
-  for (const part of protectedTextParts(text)) {
-    const match = !part.code && part.text.match(/^(\s*)([\s\S]*?)(\s*)$/)
-    if (!match || !/\p{L}/u.test(match[2])) { output.push(part.text); continue }
-    const chunks = []
-    for (const chunk of split(match[2])) chunks.push((await translate(chunk)).trim())
-    output.push(match[1] + chunks.join('\n\n') + match[3])
-  }
-  return output.join('')
-}
+// Code spans keep their exact source offsets so neither translation nor batching
+// needs placeholders that a provider could translate, remove, or reorder.
+export function protectedTextParts(text) {
+  const parts = []
+  let proseStart = 0
+  let cursor = 0
+  function protect(start, end, complete) {
+    if (start > proseStart) parts.push({ text: text.slice(proseStart, start), start: proseStart, end: start, code: false, complete: true })
+    parts.push({ text: text.slice(start, end), start, end, code: true, complete })
+    proseStart = end
+    cursor = end
+  }
+  function lineEnd(start) {
+    const newline = text.indexOf('\n', start)
+    return newline < 0 ? text.length : newline + 1
+  }
+  function technicalSpan(start) {
+    const before = text[start - 1] || ''
+    if (before && /[A-Za-z0-9_\\/]/.test(before)) return 0
+    const rest = text.slice(start)
+    // Quotes are part of a path token only when its prefix is unambiguous.
+    const quoted = rest.match(/^("|'|“)(?:[A-Za-z]:[\\/]|\\\\|~?\/|\.\.?\/)/)
+    if (quoted) {
+      const close = quoted[1] === '“' ? '”' : quoted[1]
+      const end = text.indexOf(close, start + 1)
+      return end < 0 ? text.length - start : end + 1 - start
+    }
+    const token = rest.match(/^(?:https?:\/\/|file:\/\/|[A-Za-z]:[\\/]|\\\\|~?\/|\.\.?\/)[^\s<>"'`，。！？；：、]+/)
+      || rest.match(/^[\w@.-]+\.(?:[cm]?js|jsx|tsx?|jsonl?|ya?ml|toml|ini|env|md|py|sh|ps1|cmd|bat|exe|lock|go|rs|cpp|sql|html|css|txt|zstd|png|jpe?g|svg)(?![\w.])/i)
+    if (token) {
+      let value = token[0].replace(/[.,;:!?，。！？；：、]+$/, '')
+      // Keep balanced URL parentheses, but leave Markdown/prose delimiters out.
+      for (const [open, close] of [['(', ')'], ['[', ']'], ['{', '}']]) {
+        while (value.endsWith(close) && value.split(close).length > value.split(open).length) value = value.slice(0, -1)
+      }
+      return value.length
+    }
+    const command = rest.match(/^(?:(?:npm|pnpm|yarn)\s+(?:run\s+[\w:.-]+|test|ci|install|build|start)|git\s+(?:status|diff|log|fetch|pull|push|show|branch)|(?:python3?|node)\s+[\w./\\:-]+\.(?:py|[cm]?js))(?=$|[\s.,;:!?，。！？；：、)]|--)(?:[ \t]+(?:--?[\w-]+(?:=[\w./\\:@=-]+)?|--|"[^"\r\n]*"|'[^'\r\n]*'))*/)
+    return command ? command[0].length : 0
+  }
+  while (cursor < text.length) {
+    if (cursor === 0 || text[cursor - 1] === '\n') {
+      const end = lineEnd(cursor)
+      const line = text.slice(cursor, end)
+      // Treat indented/quoted fences conservatively as code as well.
+      const fence = line.match(/^[\t ]*(?:>[\t ]*)*(?:(?:[-+*]|[0-9]+[.)])[\t ]+)?(`{3,}|~{3,})[^\r\n]*(?:\r?\n)?$/)
+      if (fence) {
+        const close = new RegExp('^[\\t ]*(?:>[\\t ]*)*' + fence[1][0] + '{' + fence[1].length + ',}[\\t ]*(?:\\r?\\n)?$')
+        let stop = end
+        let complete = false
+        while (stop < text.length) {
+          const next = lineEnd(stop)
+          if (close.test(text.slice(stop, next))) { stop = next; complete = true; break }
+          stop = next
+        }
+        protect(cursor, stop, complete)
+        continue
+      }
+      if (/^(?: {4}|\t)/.test(line)) {
+        protect(cursor, end, end < text.length || line.endsWith('\n'))
+        continue
+      }
+      const shellLine = /^(?:\$\s*)?(?:(?:npm|pnpm|yarn)\s+(?:run|test|ci|install|add|remove|uninstall|build|start)\b|git\s+(?:status|diff|log|fetch|pull|push|show|branch|checkout|switch|commit|add)\b|(?:curl|wget)\s+-)/.test(line)
+      if (shellLine && !/\b(?:and|then|before|after|because|please|shows|prints|returns|failed|succeeded|should|will|can|is|are)\b/.test(line)) {
+        protect(cursor, end, end < text.length || line.endsWith('\n'))
+        continue
+      }
+    }
+    if (text[cursor] === '`') {
+      let slashes = 0
+      for (let i = cursor - 1; i >= 0 && text[i] === '\\'; i--) slashes++
+      if (slashes % 2) { cursor++; continue }
+      let after = cursor + 1
+      while (text[after] === '`') after++
+      const size = after - cursor
+      let next = after
+      let stop = text.length
+      let complete = false
+      while ((next = text.indexOf('`', next)) >= 0) {
+        let end = next + 1
+        while (text[end] === '`') end++
+        if (end - next === size) { stop = end; complete = true; break }
+        next = end
+      }
+      // An unfinished streaming code span must not expose its body as prose.
+      protect(cursor, stop, complete)
+      continue
+    }
+    const technicalLength = technicalSpan(cursor)
+    if (technicalLength) {
+      protect(cursor, cursor + technicalLength, true)
+      continue
+    }
+    cursor++
+  }
+  if (proseStart < text.length) parts.push({ text: text.slice(proseStart), start: proseStart, end: text.length, code: false, complete: true })
+  return parts
+}
+
+// Only prose reaches a provider; separators adjoining code stay byte-for-byte.
+export async function translateProtectedText(text, translate, split = value => [value]) {
+  const output = []
+  for (const part of protectedTextParts(text)) {
+    const match = !part.code && part.text.match(/^(\s*)([\s\S]*?)(\s*)$/)
+    if (!match || !/\p{L}/u.test(match[2])) { output.push(part.text); continue }
+    const chunks = []
+    for (const chunk of split(match[2])) chunks.push((await translate(chunk)).trim())
+    output.push(match[1] + chunks.join('\n\n') + match[3])
+  }
+  return output.join('')
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-think-translate",
   "description": "Display-layer translation for the DSH web app: thinking chain, task cards, and optional answer text, in 9 target languages (local Ollama / any OpenAI-compatible or Anthropic endpoint / Google / Bing, fail-open)",
-  "version": "1.2.19",
+  "version": "1.2.21",
   "type": "module",
   "main": "lib/index.js",
   "exports": {
@@ -20,6 +20,7 @@
     "lib/credentials.js",
     "lib/runtime-state.js",
     "lib/glossary.js",
+    "lib/glossary-presets.js",
     "cordis.patch.yml",
     "docs/official-translation-comparison.md",
     "docs/first-release-evidence.json",