Function computeVisualWordOrder

lib/MiniBidi/BidiUtils.cpp:103–232

}

bool computeVisualWordOrder(const std::vector<std::string>& words, bool paragraphIsRtl,
                            std::vector<uint16_t>& visualOrder) {
  visualOrder.clear();
  const size_t nWords = words.size();
  if (nWords <= 1 || nWords > BIDI_MAX_LINE) return false;

  static bidi_char line[BIDI_MAX_LINE];
  int count = 0;
  bool truncated = false;

  for (size_t w = 0; w < nWords && !truncated; w++) {
    auto* p = reinterpret_cast<const unsigned char*>(words[w].c_str());
    while (*p) {
      if (count >= BIDI_MAX_LINE) {
        truncated = true;
        break;
      }
      const uint32_t cp = utf8NextCodepoint(&p);
      if (!cp || cp == REPLACEMENT_GLYPH) break;
      line[count].origwc = line[count].wc = cp;
      line[count].index = static_cast<uint16_t>(w);
      count++;
    }

    if (!truncated && w + 1 < nWords) {
      if (count >= BIDI_MAX_LINE) {
        truncated = true;
        break;
      }
      line[count].origwc = line[count].wc = ' ';
      line[count].index = static_cast<uint16_t>(nWords);
      count++;
    }
  }

  if (truncated || count == 0) return false;

  // Fast-path for homogeneous lines: skip UAX#9 if there's no mixing.
  bool hasL = false, hasR = false;
  for (int i = 0; i < count; i++) {
    uchar bc = bidi_class(line[i].wc);
    if (bc == L || bc == EN || bc == AN)
      hasL = true;
    else if (bc == R || bc == AL)
      hasR = true;
  }

  // Purely LTR line in RTL paragraph: identity order, but we might still need to reorder
  // if some characters are mirrored or neutral resolution differs.
  // Actually, UAX#9 rule L1/L2 says purely LTR in RTL para stays as is (identity).
  // Purely RTL line: just reverse the words.
  if (!hasL && hasR && paragraphIsRtl) {
    visualOrder.reserve(nWords);
    for (int i = static_cast<int>(nWords) - 1; i >= 0; i--) {
      visualOrder.push_back(static_cast<uint16_t>(i));
    }
    return true;
  }