// 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).
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 || '');

// Translate the prose pieces (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) {
  const parts = splitProse(text);
  const prose = parts.filter((p) => p.prose && hasWords(p.prose));
  if (!prose.length) return text;
  const outs = await translateMany(prose.map((p) => p.prose.trim()));
  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;
  });
  return parts.map((p) => p.keep ?? p.out ?? p.prose).join('');
}