#include "simdutf.h"

#include <fstream>
#include <random>
#include <memory>
#include <vector>

#include <tests/reference/encode_utf8.h>
#include <tests/helpers/test.h>

static uint32_t seed = 123;
const size_t MAX_SIZE = 1025;

#ifndef SIMDUTF_TEST_FUZZER_TRIALS
  #error "SIMDUTF_TEST_FUZZER_TRIALS not set."
#endif
constexpr std::size_t trials = SIMDUTF_TEST_FUZZER_TRIALS;

std::vector<char> input;
std::pair<bool, bool> is_ok_utf8 = std::make_pair(false, false);
std::pair<bool, bool> is_ok_utf16 = std::make_pair(false, false);
std::pair<bool, bool> is_ok_utf32 = std::make_pair(false, false);

std::pair<bool, size_t> utf8_to_utf16 = std::make_pair(false, 0);
std::pair<bool, size_t> utf8_to_utf32 = std::make_pair(false, 0);
std::pair<bool, size_t> utf16_to_utf8 = std::make_pair(false, 0);
std::pair<bool, size_t> utf16_to_utf32 = std::make_pair(false, 0);
std::pair<bool, size_t> utf32_to_utf8 = std::make_pair(false, 0);
std::pair<bool, size_t> utf32_to_utf16 = std::make_pair(false, 0);

void reset() {
  is_ok_utf8.first = false;
  is_ok_utf16.first = false;
  is_ok_utf32.first = false;

  utf8_to_utf16.first = false;
  utf8_to_utf32.first = false;
  utf16_to_utf8.first = false;
  utf16_to_utf32.first = false;
  utf32_to_utf8.first = false;
  utf32_to_utf16.first = false;
}

extern "C" {
void __asan_on_error() {
  std::fstream log;
  log.open("fuzzer_log.txt", std::ios::app);
  const size_t buf_size = 4 * MAX_SIZE + 3;
  char buffer[buf_size];
  for (unsigned int i = 0; i < input.size(); i++) {
    SIMDUTF_PUSH_DISABLE_WARNINGS
    SIMDUTF_DISABLE_DEPRECATED_WARNING
    sprintf(buffer + 4 * i + 1, "\\x%02x", input[i]);
    SIMDUTF_POP_DISABLE_WARNINGS
  }
  buffer[0] = '"';
  buffer[buf_size - 2] = '"';
  buffer[buf_size - 1] = '\0';
  log << std::boolalpha;
  log << "Input: " << buffer << '\n';
  if (is_ok_utf8.first) {
    log << "validate_utf8:" << is_ok_utf8.second << '\n';
  }
  if (is_ok_utf16.first) {
    log << "validate_utf16le:" << is_ok_utf16.second << '\n';
  }
  if (is_ok_utf32.first) {
    log << "validate_utf32:" << is_ok_utf32.second << '\n';
  }
  if (utf8_to_utf16.first) {
    log << "convert_utf8_to_utf16le:" << utf8_to_utf16.second << '\n';
  }
  if (utf8_to_utf32.first) {
    log << "convert_utf8_to_utf32:" << utf8_to_utf32.second << '\n';
  }
  if (utf16_to_utf8.first) {
    log << "convert_utf16le_to_utf8:" << utf16_to_utf8.second << '\n';
  }
  if (utf16_to_utf32.first) {
    log << "convert_utf16le_to_utf32:" << utf16_to_utf32.second << '\n';
  }
  if (utf32_to_utf8.first) {
    log << "convert_utf32_to_utf8:" << utf32_to_utf8.second << '\n';
  }
  if (utf32_to_utf16.first) {
    log << "convert_utf32_to_utf16le:" << utf32_to_utf16.second << '\n';
  }
  log << '\n';
  log.close();
}
}

/**
 * We should be able to receive random data without any problem
 * when using the validating transcoder. It is difficult to test
 * extensively, but it is easy to try many thousands of random test
 * cases.
 */

namespace {
std::vector<size_t> input_size{7,   16,  12,  64,   67,
                               128, 129, 256, 1024, MAX_SIZE};
int weights[6][2] = {{10, 90}, {25, 75}, {50, 50}, {75, 25}, {90, 10}, {99, 1}};
} // namespace

//  Possible states.
//  Format: xxx_yyy where xxx is the number of bytes (in UTF-8) and yyy is the
//  error encoded (if any).
enum state {
  ONE_VALID = 0,
  ONE_TOO_LONG,
  ONE_TOO_LARGE,
  TWO_VALID, // 3
  TWO_HEADER,
  TWO_TOO_SHORT,
  TWO_TOO_LONG,
  TWO_OVERLONG,
  THREE_VALID, // 8
  THREE_HEADER,
  THREE_TOO_SHORT,
  THREE_TOO_LONG,
  THREE_OVERLONG,
  THREE_SURROGATE,
  FOUR_VALID, // 14
  FOUR_HEADER,
  FOUR_TOO_SHORT,
  FOUR_TOO_LONG,
  FOUR_OVERLONG,
  FOUR_TOO_LARGE
};

struct state_tracker {
private:
  enum state current_state;
  enum state first_error;
  double state_weights[20];
  std::mt19937 gen;

public:
  state_tracker(uint64_t seed, double onevalid, double onetoolong,
                double onetoolarge, double twovalid, double twoheader,
                double twotooshort, double twotoolong, double twooverlong,
                double threevalid, double threeheader, double threetooshort,
                double threetoolong, double threeoverlong,
                double threesurrogate, double fourvalid, double fourheader,
                double fourtooshort, double fourtoolong, double fouroverlong,
                double fourtoolarge) noexcept
      : state_weights{onevalid,      onetoolong,     onetoolarge,
                      twovalid,      twoheader,      twotooshort,
                      twotoolong,    twooverlong,    threevalid,
                      threeheader,   threetooshort,  threetoolong,
                      threeoverlong, threesurrogate, fourvalid,
                      fourheader,    fourtooshort,   fourtoolong,
                      fouroverlong,  fourtoolarge},
        gen(std::mt19937::result_type(seed)) {
    current_state = next_state();
    first_error = current_state;
  }

  // valid_weight is the weight of having a valid state (4 valid states) and
  // invalid_weight is the weight of having a invalid state (16 invalid states)
  state_tracker(uint64_t seed, double valid_weight, double invalid_weight)
      : state_tracker(
            seed, valid_weight / 4, invalid_weight / 16, invalid_weight / 16,
            valid_weight / 4, invalid_weight / 16, invalid_weight / 16,
            invalid_weight / 16, invalid_weight / 16, valid_weight / 4,
            invalid_weight / 16, invalid_weight / 16, invalid_weight / 16,
            invalid_weight / 16, invalid_weight / 16, valid_weight / 4,
            invalid_weight / 16, invalid_weight / 16, invalid_weight / 16,
            invalid_weight / 16, invalid_weight / 16) {}

  size_t next(std::vector<char> &output) {
    // Write current state to output
    auto consume = [&output](uint8_t byte) { output.push_back(byte); };
    size_t count{0};
    switch (current_state) {
    case ONE_VALID: {
      simdutf::tests::reference::utf8::encode(generate(0x0, 0x7f), consume);
      count = 1;
      break;
    }
    case ONE_TOO_LONG: {
      simdutf::tests::reference::utf8::encode(generate(0x0, 0x7f), consume);
      output.push_back(generate(0x80, 0xbf)); // Add random continuation byte
      count = 2;
      break;
    }
    case ONE_TOO_LARGE: {
      output.push_back(generate(0x80, 0xff));
      count = 1;
      break;
    }
    case TWO_VALID: {
      simdutf::tests::reference::utf8::encode(generate(0x80, 0x7ff), consume);
      count = 2;
      break;
    }
    case TWO_HEADER: {
      uint32_t codepoint = generate(0x80, 0x7ff);
      output.push_back(0xf8 | (codepoint >> 6)); // Corrupt leading byte
      output.push_back(0x80 | (codepoint & 0x3f));
      count = 2;
      break;
    }
    case TWO_TOO_SHORT: {
      output.push_back(
          generate(0xc1, 0xdf)); // Only produce normal leading byte
      count = 1;
      break;
    }
    case TWO_TOO_LONG: {
      simdutf::tests::reference::utf8::encode(generate(0x80, 0x7ff), consume);
      output.push_back(generate(0x80, 0xbf)); // Add random continuation byte
      count = 3;
      break;
    }
    case TWO_OVERLONG: {
      output.push_back(char(0xc0));           // Add "empty" leading byte
      output.push_back(generate(0x80, 0xbf)); // Add random continuation byte
      count = 2;
      break;
    }
    case THREE_VALID: {
      uint32_t codepoint = generate(0x800, 0xffff);
      // Hacky, but there is only a ~3.2% to generate a codepoint in the
      // forbidden range each time
      while (codepoint >= 0xd800 && codepoint <= 0xdfff) {
        codepoint = generate(0x800, 0xffff);
      }
      simdutf::tests::reference::utf8::encode(generate(0x80, 0x7ff), consume);
      count = 3;
      break;
    }
    case THREE_HEADER: {
      uint32_t codepoint = generate(0x800, 0xffff);
      while (codepoint >= 0xd800 && codepoint <= 0xdfff) {
        codepoint = generate(0x800, 0xffff);
      }
      output.push_back(0xf8 | (codepoint >> 12)); // Corrupt leading byte
      output.push_back(0x80 | ((codepoint >> 6) & 0x3f));
      output.push_back(0x80 | (codepoint & 0x3f));
      count = 3;
      break;
    }
    case THREE_TOO_SHORT: {
      uint32_t codepoint = generate(0x800, 0xffff);
      while (codepoint >= 0xd800 && codepoint <= 0xdfff) {
        codepoint = generate(0x800, 0xffff);
      }
      output.push_back(0xe0 | (codepoint >> 12)); // Corrupt leading byte
      output.push_back(0x80 | ((codepoint >> 6) & 0x3f));
      count = 2;
      break;
    }
    case THREE_TOO_LONG: {
      uint32_t codepoint = generate(0x800, 0xffff);
      while (codepoint >= 0xd800 && codepoint <= 0xdfff) {
        codepoint = generate(0x800, 0xffff);
      }
      simdutf::tests::reference::utf8::encode(generate(0x80, 0x7ff), consume);
      output.push_back(generate(0x80, 0xbf)); // Add random continuation byte
      count = 4;
      break;
    }
    case THREE_OVERLONG: {
      output.push_back(char(0xe0)); // Add "empty" leading byte
      output.push_back(generate(
          0x80, 0x9f)); // First continuation byte must start by 0x8_ or 0x9_
      output.push_back(generate(0x80, 0xbf)); // Add random continuation byte
      count = 3;
      break;
    }
    case THREE_SURROGATE: {
      simdutf::tests::reference::utf8::encode(generate(0xd800, 0xdfff),
                                              consume);
      count = 3;
      break;
    }
    case FOUR_VALID: {
      simdutf::tests::reference::utf8::encode(generate(0x10000, 0x10ffff),
                                              consume);
      count = 4;
      break;
    }
    case FOUR_HEADER: {
      uint32_t codepoint = generate(0x10000, 0x10ffff);
      output.push_back(0xf8 | (codepoint >> 18)); // Corrupt leading byte
      output.push_back(0x80 | ((codepoint >> 12) & 0x3f));
      output.push_back(0x80 | ((codepoint >> 6) & 0x3f));
      output.push_back(0x80 | (codepoint & 0x3f));
      count = 4;
      break;
    }
    case FOUR_TOO_SHORT: {
      uint32_t codepoint = generate(0x10000, 0x10ffff);
      output.push_back(0xf0 | (codepoint >> 18));
      output.push_back(0x80 | ((codepoint >> 12) & 0x3f));
      output.push_back(0x80 | ((codepoint >> 6) & 0x3f));
      count = 3;
      break;
    }
    case FOUR_TOO_LONG: {
      simdutf::tests::reference::utf8::encode(generate(0x10000, 0x10ffff),
                                              consume);
      output.push_back(generate(0x80, 0xbf)); // Add random continuation byte
      count = 5;
      break;
    }
    case FOUR_OVERLONG: {
      output.push_back(char(0xf0)); // Add "empty" leading byte
      output.push_back(generate(
          0x80, 0x8f)); // First continuation byte must have start by 0x8_
      output.push_back(
          generate(0x80, 0xbf)); // Add two random continuation bytes
      output.push_back(generate(0x80, 0xbf));
      count = 4;
      break;
    }
    case FOUR_TOO_LARGE: {
      simdutf::tests::reference::utf8::encode(generate(0x110000, 0x1fffff),
                                              consume);
      count = 4;
      break;
    }
    }
    if (first_error == ONE_VALID || first_error == TWO_VALID ||
        first_error == THREE_VALID || first_error == FOUR_VALID) {
      if (current_state != ONE_VALID && current_state != TWO_VALID &&
          current_state != THREE_VALID && current_state != FOUR_VALID) {
        first_error = current_state;
      }
    }
    // Move to next state
    current_state = next_state();

    return count;
  }

private:
  uint32_t generate(uint32_t lo, uint32_t hi) {
    return std::uniform_int_distribution<uint32_t>{lo, hi}(gen);
  }

  enum state next_state() {
    return static_cast<state>(std::discrete_distribution<>{
        state_weights[0],  state_weights[1],  state_weights[2],
        state_weights[3],  state_weights[4],  state_weights[5],
        state_weights[6],  state_weights[7],  state_weights[8],
        state_weights[9],  state_weights[10], state_weights[11],
        state_weights[12], state_weights[13], state_weights[14],
        state_weights[15], state_weights[16], state_weights[17],
        state_weights[18], state_weights[19]}(gen));
  }
};

TEST(garbage_utf8_fuzz_with_errors) {
  // Here we generate fully random inputs and try transcoding from UTF-8.
  // The inputs are almost certainly *NOT* valid UTF-8.
  std::mt19937 gen((std::mt19937::result_type)(seed));
  std::uniform_int_distribution<size_t> length_generator{1, 65};
  std::uniform_int_distribution<uint32_t> byte_generator{0, 256};

  for (size_t counter = 0; counter < trials; counter++) {
    if ((counter % 10000) == 0) {
      printf("-");
      fflush(NULL);
    }
    size_t length = length_generator(gen);
    std::unique_ptr<char[]> utf8_buffer(new char[length]);
    for (size_t i = 0; i < length; i++) {
      utf8_buffer.get()[i] = byte_generator(gen);
    }
    size_t expected_utf16_length =
        implementation.utf16_length_from_utf8(utf8_buffer.get(), length);
    std::unique_ptr<char16_t[]> utf16_buffer(
        new char16_t[expected_utf16_length]);
    auto r = implementation.convert_utf8_to_utf16le_with_errors(
        utf8_buffer.get(), length, utf16_buffer.get());
    // r.count: In case of error, indicates the position of the error in the
    // input. In case of success, indicates the number of code units
    // validated/written.
    if (r.error == simdutf::SUCCESS) {
      ASSERT_EQUAL(r.count, expected_utf16_length);
    } else {
      ASSERT_TRUE(r.count < length);
    }
    r = implementation.convert_utf8_to_utf16be_with_errors(
        utf8_buffer.get(), length, utf16_buffer.get());
    if (r.error == simdutf::SUCCESS) {
      ASSERT_EQUAL(r.count, expected_utf16_length);
    } else {
      ASSERT_TRUE(r.count < length);
    }
    size_t expected_utf32_length =
        implementation.utf32_length_from_utf8(utf8_buffer.get(), length);
    std::unique_ptr<char32_t[]> utf32_buffer(
        new char32_t[expected_utf32_length]);
    r = implementation.convert_utf8_to_utf32_with_errors(
        utf8_buffer.get(), length, utf32_buffer.get());
    if (r.error == simdutf::SUCCESS) {
      ASSERT_EQUAL(r.count, expected_utf32_length);
    } else {
      ASSERT_TRUE(r.count < length);
    }
  }
}

TEST(garbage_utf8_fuzz) {
  // Here we generate fully random inputs and try transcoding from UTF-8.
  // The inputs are almost certainly *NOT* valid UTF-8.
  std::mt19937 gen((std::mt19937::result_type)(seed));
  std::uniform_int_distribution<size_t> length_generator{1, 65};
  std::uniform_int_distribution<uint32_t> byte_generator{0, 256};

  for (size_t counter = 0; counter < trials; counter++) {
    if ((counter % 10000) == 0) {
      printf("-");
      fflush(NULL);
    }
    size_t length = length_generator(gen);
    std::unique_ptr<char[]> utf8_buffer(new char[length]);
    for (size_t i = 0; i < length; i++) {
      utf8_buffer.get()[i] = byte_generator(gen);
    }
    size_t expected_utf16_length =
        implementation.utf16_length_from_utf8(utf8_buffer.get(), length);
    std::unique_ptr<char16_t[]> utf16_buffer(
        new char16_t[expected_utf16_length]);
    auto r = implementation.convert_utf8_to_utf16le(utf8_buffer.get(), length,
                                                    utf16_buffer.get());
    if (r != 0) {
      ASSERT_EQUAL(r, expected_utf16_length);
    }
    r = implementation.convert_utf8_to_utf16be(utf8_buffer.get(), length,
                                               utf16_buffer.get());
    if (r != 0) {
      ASSERT_EQUAL(r, expected_utf16_length);
    }
    size_t expected_utf32_length =
        implementation.utf32_length_from_utf8(utf8_buffer.get(), length);
    std::unique_ptr<char32_t[]> utf32_buffer(
        new char32_t[expected_utf32_length]);
    r = implementation.convert_utf8_to_utf32(utf8_buffer.get(), length,
                                             utf32_buffer.get());
    if (r != 0) {
      ASSERT_EQUAL(r, expected_utf32_length);
    }
  }
}

TEST(overflow_fuzz) {
  size_t counter{0};
  for (int i = 0; i < 6; i++) {
    state_tracker tracker(seed, weights[i][0], weights[i][1]);
    while (counter < trials) {
      for (size_t size : input_size) {
        input.clear();
        std::vector<char> output(4 * size);
        while (input.size() < size) {
          tracker.next(input);
        }
        input.shrink_to_fit(); // make sure we don't have superfluous space
        counter++;
        if ((counter % 10000) == 0) {
          printf("-");
          fflush(NULL);
        }
        reset();
        is_ok_utf8.first = true;
        is_ok_utf8.second =
            implementation.validate_utf8(input.data(), input.size());
        is_ok_utf16.first = true;
        is_ok_utf16.second = implementation.validate_utf16le(
            reinterpret_cast<char16_t *>(input.data()),
            input.size() / sizeof(char16_t));
        is_ok_utf32.first = true;
        is_ok_utf32.second = implementation.validate_utf32(
            reinterpret_cast<char32_t *>(input.data()),
            input.size() / sizeof(char32_t));
        if (is_ok_utf8.second) {
          size_t expected_length =
              implementation.utf16_length_from_utf8(input.data(), input.size());
          output.resize(expected_length * sizeof(char16_t));
          output.shrink_to_fit(); // make sure we don't have superfluous space
          utf8_to_utf16.first = true;
          utf8_to_utf16.second = implementation.convert_utf8_to_utf16le(
              input.data(), input.size(),
              reinterpret_cast<char16_t *>(output.data()));
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char16_t) == output.size()) &&
                      expected_length == utf8_to_utf16.second);

          expected_length =
              implementation.utf32_length_from_utf8(input.data(), input.size());
          output.resize(expected_length * sizeof(char32_t));
          output.shrink_to_fit(); // make sure we don't have superfluous space
          utf8_to_utf32.first = true;
          utf8_to_utf32.second = implementation.convert_utf8_to_utf32(
              input.data(), input.size(),
              reinterpret_cast<char32_t *>(output.data()));
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char32_t) == output.size()) &&
                      expected_length == utf8_to_utf32.second);
        }
        if (is_ok_utf16.second) {
          size_t expected_length = implementation.utf8_length_from_utf16le(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t));
          output.resize(expected_length);
          output.shrink_to_fit(); // make sure we don't have superfluous space
          utf16_to_utf8.first = true;
          utf16_to_utf8.second = implementation.convert_utf16le_to_utf8(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t), output.data());
          ASSERT_TRUE(expected_length > 0 && expected_length == output.size() &&
                      expected_length == utf16_to_utf8.second);

          expected_length = implementation.utf32_length_from_utf16le(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t));
          output.resize(expected_length * sizeof(char32_t));

          output.shrink_to_fit(); // make sure we don't have superfluous space
          utf16_to_utf32.first = true;
          utf16_to_utf32.second = implementation.convert_utf16le_to_utf32(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t),
              reinterpret_cast<char32_t *>(output.data()));
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char32_t) == output.size()) &&
                      expected_length == utf16_to_utf32.second);
        }
        if (is_ok_utf32.second) {
          size_t expected_length = implementation.utf8_length_from_utf32(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t));
          output.resize(expected_length);
          output.shrink_to_fit(); // make sure we don't have superfluous space
          utf32_to_utf8.first = true;
          utf32_to_utf8.second = implementation.convert_utf32_to_utf8(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t), output.data());
          ASSERT_TRUE(expected_length > 0 && expected_length == output.size() &&
                      expected_length == utf32_to_utf8.second);

          expected_length = implementation.utf16_length_from_utf32(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t));
          output.resize(expected_length * sizeof(char16_t));
          output.shrink_to_fit(); // make sure we don't have superfluous space
          utf32_to_utf16.first = true;
          utf32_to_utf16.second = implementation.convert_utf32_to_utf16le(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t),
              reinterpret_cast<char16_t *>(output.data()));
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char16_t) == output.size()) &&
                      expected_length == utf32_to_utf16.second);
        }
      }
    }
  }
}

TEST(overflow_with_errors_fuzz) {
  size_t counter{0};
  for (int i = 0; i < 6; i++) {
    state_tracker tracker(seed, weights[i][0], weights[i][1]);
    while (counter < trials) {
      for (size_t size : input_size) {
        input.clear();
        std::vector<char> output(4 * size);
        while (input.size() < size) {
          tracker.next(input);
        }
        input.shrink_to_fit(); // make sure we don't have superfluous space
        counter++;
        if ((counter % 10000) == 0) {
          printf("-");
          fflush(NULL);
        }
        reset();
        is_ok_utf8.first = true;
        is_ok_utf8.second =
            implementation.validate_utf8(input.data(), input.size());
        is_ok_utf16.first = true;
        is_ok_utf16.second = implementation.validate_utf16le(
            reinterpret_cast<char16_t *>(input.data()),
            input.size() / sizeof(char16_t));
        is_ok_utf32.first = true;
        is_ok_utf32.second = implementation.validate_utf32(
            reinterpret_cast<char32_t *>(input.data()),
            input.size() / sizeof(char32_t));
        if (is_ok_utf8.second) {
          size_t expected_length =
              implementation.utf16_length_from_utf8(input.data(), input.size());
          output.resize(expected_length * sizeof(char16_t));
          output.shrink_to_fit(); // make sure we don't have superfluous space
          auto r = implementation.convert_utf8_to_utf16le_with_errors(
              input.data(), input.size(),
              reinterpret_cast<char16_t *>(output.data()));
          utf8_to_utf16.second = r.count;
          utf8_to_utf16.first = (r.error == simdutf::SUCCESS);
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char16_t) == output.size()) &&
                      expected_length == utf8_to_utf16.second);

          expected_length =
              implementation.utf32_length_from_utf8(input.data(), input.size());
          output.resize(expected_length * sizeof(char32_t));
          output.shrink_to_fit(); // make sure we don't have superfluous space
          r = implementation.convert_utf8_to_utf32_with_errors(
              input.data(), input.size(),
              reinterpret_cast<char32_t *>(output.data()));
          utf8_to_utf32.second = r.count;
          utf8_to_utf32.first = (r.error == simdutf::SUCCESS);
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char32_t) == output.size()) &&
                      expected_length == utf8_to_utf32.second);
        }
        if (is_ok_utf16.second) {
          size_t expected_length = implementation.utf8_length_from_utf16le(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t));
          output.resize(expected_length);
          output.shrink_to_fit(); // make sure we don't have superfluous space
          auto r = implementation.convert_utf16le_to_utf8_with_errors(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t), output.data());
          utf16_to_utf8.second = r.count;
          utf16_to_utf8.first = (r.error == simdutf::SUCCESS);
          ASSERT_TRUE(expected_length > 0 && expected_length == output.size() &&
                      expected_length == utf16_to_utf8.second);

          expected_length = implementation.utf32_length_from_utf16le(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t));
          output.resize(expected_length * sizeof(char32_t));

          output.shrink_to_fit(); // make sure we don't have superfluous space
          r = implementation.convert_utf16le_to_utf32_with_errors(
              reinterpret_cast<char16_t *>(input.data()),
              input.size() / sizeof(char16_t),
              reinterpret_cast<char32_t *>(output.data()));
          utf16_to_utf32.first = true;
          utf16_to_utf32.second = r.count;
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char32_t) == output.size()) &&
                      expected_length == utf16_to_utf32.second);
        }
        if (is_ok_utf32.second) {
          size_t expected_length = implementation.utf8_length_from_utf32(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t));
          output.resize(expected_length);
          output.shrink_to_fit(); // make sure we don't have superfluous space
          auto r = implementation.convert_utf32_to_utf8_with_errors(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t), output.data());
          utf32_to_utf8.first = (r.error == simdutf::SUCCESS);
          utf32_to_utf8.second = r.count;
          ASSERT_TRUE(expected_length > 0 && expected_length == output.size() &&
                      expected_length == utf32_to_utf8.second);

          expected_length = implementation.utf16_length_from_utf32(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t));
          output.resize(expected_length * sizeof(char16_t));
          output.shrink_to_fit(); // make sure we don't have superfluous space
          r = implementation.convert_utf32_to_utf16le_with_errors(
              reinterpret_cast<char32_t *>(input.data()),
              input.size() / sizeof(char32_t),
              reinterpret_cast<char16_t *>(output.data()));
          utf32_to_utf16.first = (r.error == simdutf::SUCCESS);
          utf32_to_utf16.second = r.count;
          ASSERT_TRUE(expected_length > 0 &&
                      (expected_length * sizeof(char16_t) == output.size()) &&
                      expected_length == utf32_to_utf16.second);
        }
      }
    }
  }
}

TEST(basic_fuzz) {
  size_t counter{0};
  for (int i = 0; i < 6; i++) {
    state_tracker tracker(seed, weights[i][0], weights[i][1]);
    while (counter < trials) {
      for (size_t size : input_size) {
        input.clear();
        std::vector<char> output(4 * size);
        while (input.size() < size) {
          tracker.next(input);
        }
        input.shrink_to_fit(); // make sure we don't have superfluous space
        counter++;
        if ((counter % 10000) == 0) {
          printf("-");
          fflush(NULL);
        }
        reset();
        is_ok_utf8.first = true;
        is_ok_utf8.second =
            implementation.validate_utf8(input.data(), input.size());
        is_ok_utf16.first = true;
        is_ok_utf16.second = implementation.validate_utf16le(
            reinterpret_cast<char16_t *>(input.data()),
            input.size() / sizeof(char16_t));
        is_ok_utf32.first = true;
        is_ok_utf32.second = implementation.validate_utf32(
            reinterpret_cast<char32_t *>(input.data()),
            input.size() / sizeof(char32_t));
        utf8_to_utf16.first = true;
        utf8_to_utf16.second = implementation.convert_utf8_to_utf16le(
            input.data(), input.size(),
            reinterpret_cast<char16_t *>(output.data()));
        utf8_to_utf32.first = true;
        utf8_to_utf32.second = implementation.convert_utf8_to_utf32(
            input.data(), input.size(),
            reinterpret_cast<char32_t *>(output.data()));
        utf16_to_utf8.first = true;
        utf16_to_utf8.second = implementation.convert_utf16le_to_utf8(
            reinterpret_cast<char16_t *>(input.data()),
            input.size() / sizeof(char16_t), output.data());
        utf16_to_utf32.first = true;
        utf16_to_utf32.second = implementation.convert_utf16le_to_utf32(
            reinterpret_cast<char16_t *>(input.data()),
            input.size() / sizeof(char16_t),
            reinterpret_cast<char32_t *>(output.data()));
        utf32_to_utf8.first = true;
        utf32_to_utf8.second = implementation.convert_utf32_to_utf8(
            reinterpret_cast<char32_t *>(input.data()),
            input.size() / sizeof(char32_t), output.data());
        utf32_to_utf16.first = true;
        utf32_to_utf16.second = implementation.convert_utf32_to_utf16le(
            reinterpret_cast<char32_t *>(input.data()),
            input.size() / sizeof(char32_t),
            reinterpret_cast<char16_t *>(output.data()));
        if (is_ok_utf8.second
                ? (utf8_to_utf16.second == 0 || utf8_to_utf32.second == 0)
                : (utf8_to_utf16.second > 0 || utf8_to_utf32.second > 0)) {
          printf("%s\n",
                 (is_ok_utf8.second ? "UTF-8 is ok" : "UTF-8 is not ok"));
          printf(" size = %zu\n", input.size());
          printf("  implementation.convert_utf8_to_utf16.second return %zu\n",
                 utf8_to_utf16.second);
          printf("  implementation.convert_utf8_to_utf32.second return %zu\n",
                 utf8_to_utf32.second);
        }
        ASSERT_TRUE(
            is_ok_utf8.second
                ? (utf8_to_utf16.second > 0 && utf8_to_utf32.second > 0)
                : (utf8_to_utf16.second == 0 && utf8_to_utf32.second == 0));
        if (is_ok_utf16.second
                ? (utf16_to_utf8.second == 0 || utf16_to_utf32.second == 0)
                : (utf16_to_utf8.second > 0 || utf16_to_utf32.second > 0)) {
          printf("%s\n",
                 (is_ok_utf16.second ? "UTF-16 is ok" : "UTF-16 is not ok"));
          printf(" size = %zu\n", input.size() / sizeof(char16_t));
          printf("  implementation.convert_utf16_to_utf8.second return %zu\n",
                 utf16_to_utf8.second);
          printf("  implementation.convert_utf16_to_utf32.second return %zu\n",
                 utf16_to_utf32.second);
        }
        ASSERT_TRUE(
            is_ok_utf16.second
                ? (utf16_to_utf8.second > 0 && utf16_to_utf32.second > 0)
                : (utf16_to_utf8.second == 0 && utf16_to_utf32.second == 0));
        if (is_ok_utf32.second
                ? (utf32_to_utf8.second == 0 || utf32_to_utf16.second == 0)
                : (utf32_to_utf8.second > 0 || utf32_to_utf16.second > 0)) {
          printf("%s\n",
                 (is_ok_utf32.second ? "UTF-32 is ok" : "UTF-32 is not ok"));
          printf(" size = %zu\n", input.size() / sizeof(char32_t));
          printf("  implementation.convert_utf32_to_utf8.second return %zu\n",
                 utf32_to_utf8.second);
          printf("  implementation.convert_utf32_to_utf16.second return %zu\n",
                 utf32_to_utf16.second);
        }
        ASSERT_TRUE(
            is_ok_utf32.second
                ? (utf32_to_utf8.second > 0 && utf32_to_utf16.second > 0)
                : (utf32_to_utf8.second == 0 && utf32_to_utf16.second == 0));
      }
    }
  }
}

int main(int argc, char *argv[]) {
  if (argc == 2) {
    try {
      seed = std::stoi(argv[1]);
    } catch (const std::exception &e) {
      printf("%s\n", e.what());
      return EXIT_FAILURE;
    }
  }
  std::mt19937 gen{seed};
  for (int i = 0; i < 20; i++) {
    input_size.push_back(std::uniform_int_distribution<uint32_t>{50, 800}(
        gen)); // Range must be less than max_size
  }
  return simdutf::test::main((argc == 2) ? 1 : argc, argv);
}