Function parse_url_impl

src/parser.cpp:14–927

template <class result_type, bool store_values>
result_type parse_url_impl(std::string_view user_input,
                           const result_type* base_url) {
  // We can specialize the implementation per type.
  // Important: result_type_is_ada_url is evaluated at *compile time*. This
  // means that doing if constexpr(result_type_is_ada_url) { something } else {
  // something else } is free (at runtime). This means that ada::url_aggregator
  // and ada::url **do not have to support the exact same API**.
  constexpr bool result_type_is_ada_url = std::is_same_v<url, result_type>;
  constexpr bool result_type_is_ada_url_aggregator =
      std::is_same_v<url_aggregator, result_type>;
  static_assert(result_type_is_ada_url ||
                result_type_is_ada_url_aggregator);  // We don't support
                                                     // anything else for now.

  ada_log("ada::parser::parse_url('", user_input, "' [", user_input.size(),
          " bytes],", (base_url != nullptr ? base_url->to_string() : "null"),
          ")");

  state state = state::SCHEME_START;
  result_type url{};

  // We refuse to parse URL strings that exceed 4GB. Such strings are almost
  // surely the result of a bug or are otherwise a security concern.
  if (user_input.size() > std::numeric_limits<uint32_t>::max()) [[unlikely]] {
    url.is_valid = false;
  }
  // Going forward, user_input.size() is in [0,
  // std::numeric_limits<uint32_t>::max). If we are provided with an invalid
  // base, or the optional_url was invalid, we must return.
  if (base_url != nullptr) {
    url.is_valid &= base_url->is_valid;
  }
  if (!url.is_valid) {
    return url;
  }
  if constexpr (result_type_is_ada_url_aggregator && store_values) {
    // Most of the time, we just need user_input.size().
    // In some instances, we may need a bit more.
    ///////////////////////////
    // This is *very* important. This line should *not* be removed
    // hastily. There are principled reasons why reserve is important
    // for performance. If you have a benchmark with small inputs,
    // it may not matter, but in other instances, it could.
    ////
    // This rounds up to the next power of two.
    // We know that user_input.size() is in [0,
    // std::numeric_limits<uint32_t>::max).
    uint32_t reserve_capacity =
        (0xFFFFFFFF >>
         helpers::leading_zeroes(uint32_t(1 | user_input.size()))) +
        1;
    url.reserve(reserve_capacity);
  }
  std::string tmp_buffer;
  std::string_view url_data;
  if (unicode::has_tabs_or_newline(user_input)) [[unlikely]] {
    tmp_buffer = user_input;
    // Optimization opportunity: Instead of copying and then pruning, we could