1 files changed, 61 insertions, 0 deletions
diff --git a/names/vendor/translate-split.js b/names/vendor/translate-split.js
new file mode 100644
index 0000000..95f6468
--- /dev/null
+++ b/names/vendor/translate-split.js
@@ -0,0 +1,61 @@
+// What must come back from a translation exactly as it went in: links,
+// nostr: mentions, custom emoji shortcodes, hashtags, @handles, invoices and
+// other long codes. Translation models mangle them ("https: / / example.com
+// / fot.jpg"), so only the prose between them is translated and the text is
+// put back together around the untouched tokens. Shared by the registrar
+// (names/translate.js) and the browser's own translator (messages.js).
+import { scriptOf, LANG_SCRIPT } from './lang-guess.js';
+
+const KEEP = /(https?:\/\/\S+|nostr:[a-z0-9]+|:[a-z0-9_+-]{1,40}:|#[\p{L}\p{N}_]+|@[\w.-]+(?:@[\w.-]+)?|\b(?:ln|lightning:|cashu|bitcoin:|nevent|npub|nprofile|note|naddr)[a-z0-9]{20,}|\S{40,})/giu;
+
+export function splitProse(text) {
+  const parts = [];
+  let last = 0;
+  for (const m of String(text || '').matchAll(KEEP)) {
+    if (m.index > last) parts.push({ prose: text.slice(last, m.index) });
+    parts.push({ keep: m[0] });
+    last = m.index + m[0].length;
+  }
+  if (last < text.length) parts.push({ prose: text.slice(last) });
+  return parts;
+}
+// a piece worth translating: has words, not just spaces and punctuation
+export const hasWords = (s) => /\p{L}{2,}/u.test(s || '');
+
+// A hashtag stays a hashtag (it's a link to its topic), but one written in
+// another script than the reader's means nothing to them: such a tag gets
+// its translation in brackets after it, "#総合部絵チャ (general art chat)".
+const tagWords = (tag) => tag.slice(1).replace(/_/g, ' ');
+export function glossable(tag, target) {
+  if (!tag.startsWith('#') || !target) return false;
+  const s = scriptOf(tagWords(tag), 1);
+  return !!s && !!LANG_SCRIPT[target] && s !== LANG_SCRIPT[target];
+}
+// What a translator gets to see of a post: its prose, then the words of the
+// hashtags the reader can't read (also what language detection runs on).
+export function translatable(text, target) {
+  const parts = splitProse(text);
+  return {
+    parts,
+    prose: parts.filter((p) => p.prose && hasWords(p.prose)),
+    tags: parts.filter((p) => p.keep && glossable(p.keep, target)),
+  };
+}
+
+// Translate the prose pieces and the glossed tags (one call for all of them)
+// and reassemble. Each piece keeps its own leading/trailing whitespace —
+// line breaks matter.
+export async function translatePieces(text, translateMany, target = null) {
+  const { parts, prose, tags } = translatable(text, target);
+  if (!prose.length && !tags.length) return text;
+  const outs = await translateMany([...prose.map((p) => p.prose.trim()), ...tags.map((p) => tagWords(p.keep))]);
+  prose.forEach((p, i) => {
+    const lead = p.prose.match(/^\s*/)[0], trail = p.prose.match(/\s*$/)[0];
+    p.out = lead + String(outs[i] ?? p.prose.trim()) + trail;
+  });
+  tags.forEach((p, i) => {
+    const g = String(outs[prose.length + i] ?? '').trim();
+    p.out = g && g.toLowerCase() !== tagWords(p.keep).toLowerCase() ? p.keep + ' (' + g + ')' : p.keep;
+  });
+  return parts.map((p) => p.out ?? p.keep ?? p.prose).join('');
+}