#ifndef NBYTES_H
#define NBYTES_H
#include <algorithm>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <cstring>
#include <string>

namespace nbytes {

#if NBYTES_DEVELOPMENT_CHECKS
#define NBYTES_STR(x) #x
#define NBYTES_REQUIRE(EXPR) \
  {                          \
    if (!(EXPR) { abort(); }) }

#define NBYTES_FAIL(MESSAGE)                         \
  do {                                               \
    std::cerr << "FAIL: " << (MESSAGE) << std::endl; \
    abort();                                         \
  } while (0);
#define NBYTES_ASSERT_EQUAL(LHS, RHS, MESSAGE)                                 \
  do {                                                                         \
    if (LHS != RHS) {                                                          \
      std::cerr << "Mismatch: '" << LHS << "' - '" << RHS << "'" << std::endl; \
      NBYTES_FAIL(MESSAGE);                                                    \
    }                                                                          \
  } while (0);
#define NBYTES_ASSERT_TRUE(COND)                                            \
  do {                                                                      \
    if (!(COND)) {                                                          \
      std::cerr << "Assert at line " << __LINE__ << " of file " << __FILE__ \
                << std::endl;                                               \
      NBYTES_FAIL(NBYTES_STR(COND));                                        \
    }                                                                       \
  } while (0);
#else
#define NBYTES_FAIL(MESSAGE)
#define NBYTES_ASSERT_EQUAL(LHS, RHS, MESSAGE)
#define NBYTES_ASSERT_TRUE(COND)
#endif

[[noreturn]] inline void unreachable() {
#ifdef __GNUC__
  __builtin_unreachable();
#elif defined(_MSC_VER)
  __assume(false);
#else
#endif
}

// The nbytes (short for "node bytes") is a set of utility helpers for
// working with bytes that are extracted from Node.js' internals. The
// motivation for extracting these into a separate library is to make it
// easier for other projects to implement functionality that is compatible
// with Node.js' implementation of various byte manipulation functions.

// Round up a to the next highest multiple of b.
template <typename T>
constexpr T RoundUp(T a, T b) {
  return a % b != 0 ? a + b - (a % b) : a;
}

// Align ptr to an `alignment`-bytes boundary.
template <typename T, typename U>
constexpr T *AlignUp(T *ptr, U alignment) {
  return reinterpret_cast<T *>(
      RoundUp(reinterpret_cast<uintptr_t>(ptr), alignment));
}

template <typename T, typename U>
inline T AlignDown(T value, U alignment) {
  return reinterpret_cast<T>(
      (reinterpret_cast<uintptr_t>(value) & ~(alignment - 1)));
}

template <typename T>
inline T MultiplyWithOverflowCheck(T a, T b) {
  auto ret = a * b;
  if (a != 0) {
    NBYTES_ASSERT_TRUE(b == ret / a);
  }

  return ret;
}

void ForceAsciiSlow(const char *src, char *dst, size_t len);
void ForceAscii(const char *src, char *dst, size_t len);

// ============================================================================
// Byte Swapping

// Swaps bytes in place. nbytes is the number of bytes to swap and must be a
// multiple of the word size (checked by function).
bool SwapBytes16(char *data, size_t nbytes);
bool SwapBytes32(char *data, size_t nbytes);
bool SwapBytes64(char *data, size_t nbytes);

// ============================================================================
// Base64 (legacy)

#ifdef _MSC_VER
#pragma warning(push)
// MSVC C4003: not enough actual parameters for macro 'identifier'
#pragma warning(disable : 4003)
#endif

extern const int8_t unbase64_table[256];

template <typename TypeName>
bool Base64DecodeGroupSlow(char *const dst, const size_t dstlen,
                           const TypeName *const src, const size_t srclen,
                           size_t *const i, size_t *const k) {
  uint8_t hi;
  uint8_t lo;
#define V(expr)                                                        \
  for (;;) {                                                           \
    const uint8_t c = static_cast<uint8_t>(src[*i]);                   \
    lo = unbase64_table[c];                                            \
    *i += 1;                                                           \
    if (lo < 64) break;                         /* Legal character. */ \
    if (c == '=' || *i >= srclen) return false; /* Stop decoding. */   \
  }                                                                    \
  expr;                                                                \
  if (*i >= srclen) return false;                                      \
  if (*k >= dstlen) return false;                                      \
  hi = lo;
  V(/* Nothing. */);
  V(dst[(*k)++] = ((hi & 0x3F) << 2) | ((lo & 0x30) >> 4));
  V(dst[(*k)++] = ((hi & 0x0F) << 4) | ((lo & 0x3C) >> 2));
  V(dst[(*k)++] = ((hi & 0x03) << 6) | ((lo & 0x3F) >> 0));
#undef V
  return true;  // Continue decoding.
}

enum class Base64Mode { NORMAL, URL };

inline constexpr size_t Base64EncodedSize(
    size_t size, Base64Mode mode = Base64Mode::NORMAL) {
  return mode == Base64Mode::NORMAL ? ((size + 2) / 3 * 4)
                                    : static_cast<size_t>(std::ceil(
                                          static_cast<double>(size * 4) / 3));
}

// Doesn't check for padding at the end.  Can be 1-2 bytes over.
inline constexpr size_t Base64DecodedSizeFast(size_t size) {
  // 1-byte input cannot be decoded
  return size > 1 ? (size / 4) * 3 + (size % 4 + 1) / 2 : 0;
}

inline uint32_t ReadUint32BE(const unsigned char *p) {
  return static_cast<uint32_t>(p[0] << 24U) |
         static_cast<uint32_t>(p[1] << 16U) |
         static_cast<uint32_t>(p[2] << 8U) | static_cast<uint32_t>(p[3]);
}

template <typename TypeName>
size_t Base64DecodedSize(const TypeName *src, size_t size) {
  // 1-byte input cannot be decoded
  if (size < 2) return 0;

  if (src[size - 1] == '=') {
    size--;
    if (src[size - 1] == '=') size--;
  }
  return Base64DecodedSizeFast(size);
}

template <typename TypeName>
size_t Base64DecodeFast(char *const dst, const size_t dstlen,
                        const TypeName *const src, const size_t srclen,
                        const size_t decoded_size) {
  const size_t available = dstlen < decoded_size ? dstlen : decoded_size;
  const size_t max_k = available / 3 * 3;
  size_t max_i = srclen / 4 * 4;
  size_t i = 0;
  size_t k = 0;
  while (i < max_i && k < max_k) {
    const unsigned char txt[] = {
        static_cast<unsigned char>(
            unbase64_table[static_cast<uint8_t>(src[i + 0])]),
        static_cast<unsigned char>(
            unbase64_table[static_cast<uint8_t>(src[i + 1])]),
        static_cast<unsigned char>(
            unbase64_table[static_cast<uint8_t>(src[i + 2])]),
        static_cast<unsigned char>(
            unbase64_table[static_cast<uint8_t>(src[i + 3])]),
    };

    const uint32_t v = ReadUint32BE(txt);
    // If MSB is set, input contains whitespace or is not valid base64.
    if (v & 0x80808080) {
      if (!Base64DecodeGroupSlow(dst, dstlen, src, srclen, &i, &k)) return k;
      max_i = i + (srclen - i) / 4 * 4;  // Align max_i again.
    } else {
      dst[k + 0] = ((v >> 22) & 0xFC) | ((v >> 20) & 0x03);
      dst[k + 1] = ((v >> 12) & 0xF0) | ((v >> 10) & 0x0F);
      dst[k + 2] = ((v >> 2) & 0xC0) | ((v >> 0) & 0x3F);
      i += 4;
      k += 3;
    }
  }
  if (i < srclen && k < dstlen) {
    Base64DecodeGroupSlow(dst, dstlen, src, srclen, &i, &k);
  }
  return k;
}

template <typename TypeName>
size_t Base64Decode(char *const dst, const size_t dstlen,
                    const TypeName *const src, const size_t srclen) {
  const size_t decoded_size = Base64DecodedSize(src, srclen);
  return Base64DecodeFast(dst, dstlen, src, srclen, decoded_size);
}

#ifdef _MSC_VER
#pragma warning(pop)
#endif

// ============================================================================
// Hex (legacy)

extern const int8_t unhex_table[256];

template <typename TypeName>
static size_t HexDecode(char *buf, size_t len, const TypeName *src,
                        const size_t srcLen) {
  size_t i;
  for (i = 0; i < len && i * 2 + 1 < srcLen; ++i) {
    unsigned a = unhex_table[static_cast<uint8_t>(src[i * 2 + 0])];
    unsigned b = unhex_table[static_cast<uint8_t>(src[i * 2 + 1])];
    if (!~a || !~b) return i;
    buf[i] = (a << 4) | b;
  }

  return i;
}

size_t HexEncode(const char *src, size_t slen, char *dst, size_t dlen);

std::string HexEncode(const char *src, size_t slen);

// ============================================================================
// StringSearch

namespace stringsearch {

template <typename T>
class Vector {
 public:
  Vector(T *data, size_t length, bool isForward)
      : start_(data), length_(length), is_forward_(isForward) {
    CHECK(length > 0 && data != nullptr);
  }

  // Returns the start of the memory range.
  // For vector v this is NOT necessarily &v[0], see forward().
  const T *start() const { return start_; }

  // Returns the length of the vector, in characters.
  size_t length() const { return length_; }

  // Returns true if the Vector is front-to-back, false if back-to-front.
  // In the latter case, v[0] corresponds to the *end* of the memory range.
  bool forward() const { return is_forward_; }

  // Access individual vector elements - checks bounds in debug mode.
  T &operator[](size_t index) const {
    NBYTES_ASSERT_TRUE(index < length_);
    return start_[is_forward_ ? index : (length_ - index - 1)];
  }

 private:
  T *start_;
  size_t length_;
  bool is_forward_;
};

//---------------------------------------------------------------------
// String Search object.
//---------------------------------------------------------------------

// Class holding constants and methods that apply to all string search variants,
// independently of subject and pattern char size.
class StringSearchBase {
 protected:
  // Cap on the maximal shift in the Boyer-Moore implementation. By setting a
  // limit, we can fix the size of tables. For a needle longer than this limit,
  // search will not be optimal, since we only build tables for a suffix
  // of the string, but it is a safe approximation.
  static const int kBMMaxShift = 250;

  // Reduce alphabet to this size.
  // One of the tables used by Boyer-Moore and Boyer-Moore-Horspool has size
  // proportional to the input alphabet. We reduce the alphabet size by
  // equating input characters modulo a smaller alphabet size. This gives
  // a potentially less efficient searching, but is a safe approximation.
  // For needles using only characters in the same Unicode 256-code point page,
  // there is no search speed degradation.
  static const int kLatin1AlphabetSize = 256;
  static const int kUC16AlphabetSize = 256;

  // Bad-char shift table stored in the state. It's length is the alphabet size.
  // For patterns below this length, the skip length of Boyer-Moore is too short
  // to compensate for the algorithmic overhead compared to simple brute force.
  static const int kBMMinPatternLength = 8;

  // Store for the BoyerMoore(Horspool) bad char shift table.
  int bad_char_shift_table_[kUC16AlphabetSize];
  // Store for the BoyerMoore good suffix shift table.
  int good_suffix_shift_table_[kBMMaxShift + 1];
  // Table used temporarily while building the BoyerMoore good suffix
  // shift table.
  int suffix_table_[kBMMaxShift + 1];
};

template <typename Char>
class StringSearch : private StringSearchBase {
 public:
  typedef stringsearch::Vector<const Char> Vector;

  explicit StringSearch(Vector pattern) : pattern_(pattern), start_(0) {
    if (pattern.length() >= kBMMaxShift) {
      start_ = pattern.length() - kBMMaxShift;
    }

    size_t pattern_length = pattern_.length();
    NBYTES_ASSERT_TRUE(pattern_length > 0);
    if (pattern_length < kBMMinPatternLength) {
      if (pattern_length == 1) {
        strategy_ = SearchStrategy::kSingleChar;
        return;
      }
      strategy_ = SearchStrategy::kLinear;
      return;
    }
    strategy_ = SearchStrategy::kInitial;
  }

  size_t Search(Vector subject, size_t index) {
    switch (strategy_) {
      case kBoyerMooreHorspool:
        return BoyerMooreHorspoolSearch(subject, index);
      case kBoyerMoore:
        return BoyerMooreSearch(subject, index);
      case kInitial:
        return InitialSearch(subject, index);
      case kLinear:
        return LinearSearch(subject, index);
      case kSingleChar:
        return SingleCharSearch(subject, index);
    }
    unreachable();
  }

  static inline int AlphabetSize() {
    if (sizeof(Char) == 1) {
      // Latin1 needle.
      return kLatin1AlphabetSize;
    } else {
      // UC16 needle.
      return kUC16AlphabetSize;
    }

    static_assert(
        sizeof(Char) == sizeof(uint8_t) || sizeof(Char) == sizeof(uint16_t),
        "sizeof(Char) == sizeof(uint16_t) || sizeof(uint8_t)");
  }

 private:
  typedef size_t (StringSearch::*SearchFunction)(Vector, size_t);
  size_t SingleCharSearch(Vector subject, size_t start_index);
  size_t LinearSearch(Vector subject, size_t start_index);
  size_t InitialSearch(Vector subject, size_t start_index);
  size_t BoyerMooreHorspoolSearch(Vector subject, size_t start_index);
  size_t BoyerMooreSearch(Vector subject, size_t start_index);

  void PopulateBoyerMooreHorspoolTable();

  void PopulateBoyerMooreTable();

  static inline int CharOccurrence(int *bad_char_occurrence, Char char_code) {
    if (sizeof(Char) == 1) {
      return bad_char_occurrence[static_cast<int>(char_code)];
    }
    // Both pattern and subject are UC16. Reduce character to equivalence class.
    int equiv_class = char_code % kUC16AlphabetSize;
    return bad_char_occurrence[equiv_class];
  }

  enum SearchStrategy {
    kBoyerMooreHorspool,
    kBoyerMoore,
    kInitial,
    kLinear,
    kSingleChar,
  };

  // The pattern to search for.
  Vector pattern_;
  SearchStrategy strategy_;
  // Cache value of Max(0, pattern_length() - kBMMaxShift)
  size_t start_;
};

inline uint8_t GetHighestValueByte(uint16_t character) {
  return std::max(static_cast<uint8_t>(character & 0xFF),
                  static_cast<uint8_t>(character >> 8));
}

inline uint8_t GetHighestValueByte(uint8_t character) { return character; }

// Searches for a byte value in a memory buffer, back to front.
// Uses memrchr(3) on systems which support it, for speed.
// Falls back to a vanilla for loop on non-GNU systems such as Windows.
inline const void *MemrchrFill(const void *haystack, uint8_t needle,
                               size_t haystack_len) {
#ifdef _GNU_SOURCE
  return memrchr(haystack, needle, haystack_len);
#else
  const uint8_t *haystack8 = static_cast<const uint8_t *>(haystack);
  for (size_t i = haystack_len - 1; i != static_cast<size_t>(-1); i--) {
    if (haystack8[i] == needle) {
      return haystack8 + i;
    }
  }
  return nullptr;
#endif
}

// Finds the first occurrence of *two-byte* character pattern[0] in the string
// `subject`. Does not check that the whole pattern matches.
template <typename Char>
inline size_t FindFirstCharacter(Vector<const Char> pattern,
                                 Vector<const Char> subject, size_t index) {
  const Char pattern_first_char = pattern[0];
  const size_t max_n = (subject.length() - pattern.length() + 1);

  // For speed, search for the more `rare` of the two bytes in pattern[0]
  // using memchr / memrchr (which are much faster than a simple for loop).
  const uint8_t search_byte = GetHighestValueByte(pattern_first_char);
  size_t pos = index;
  do {
    const size_t bytes_to_search = (max_n - pos) * sizeof(Char);
    const void *void_pos;
    if (subject.forward()) {
      // Assert that bytes_to_search won't overflow
      NBYTES_ASSERT_TRUE(pos <= max_n);
      NBYTES_ASSERT_TRUE(max_n - pos <= SIZE_MAX / sizeof(Char));
      void_pos = memchr(subject.start() + pos, search_byte, bytes_to_search);
    } else {
      NBYTES_ASSERT_TRUE(pos <= subject.length());
      NBYTES_ASSERT_TRUE(subject.length() - pos <= SIZE_MAX / sizeof(Char));
      void_pos = MemrchrFill(subject.start() + pattern.length() - 1,
                             search_byte, bytes_to_search);
    }
    const Char *char_pos = static_cast<const Char *>(void_pos);
    if (char_pos == nullptr) return subject.length();

    // Then, for each match, verify that the full two bytes match pattern[0].
    char_pos = AlignDown(char_pos, sizeof(Char));
    size_t raw_pos = static_cast<size_t>(char_pos - subject.start());
    pos = subject.forward() ? raw_pos : (subject.length() - raw_pos - 1);
    if (subject[pos] == pattern_first_char) {
      // Match found, hooray.
      return pos;
    }
    // Search byte matched, but the other byte of pattern[0] didn't. Keep going.
  } while (++pos < max_n);

  return subject.length();
}

// Finds the first occurrence of the byte pattern[0] in string `subject`.
// Does not verify that the whole pattern matches.
template <>
inline size_t FindFirstCharacter(Vector<const uint8_t> pattern,
                                 Vector<const uint8_t> subject, size_t index) {
  const uint8_t pattern_first_char = pattern[0];
  const size_t subj_len = subject.length();
  const size_t max_n = (subject.length() - pattern.length() + 1);

  const void *pos;
  if (subject.forward()) {
    pos = memchr(subject.start() + index, pattern_first_char, max_n - index);
  } else {
    pos = MemrchrFill(subject.start() + pattern.length() - 1,
                      pattern_first_char, max_n - index);
  }
  const uint8_t *char_pos = static_cast<const uint8_t *>(pos);
  if (char_pos == nullptr) {
    return subj_len;
  }

  size_t raw_pos = static_cast<size_t>(char_pos - subject.start());
  return subject.forward() ? raw_pos : (subj_len - raw_pos - 1);
}

//---------------------------------------------------------------------
// Single Character Pattern Search Strategy
//---------------------------------------------------------------------

template <typename Char>
size_t StringSearch<Char>::SingleCharSearch(Vector subject, size_t index) {
  NBYTES_ASSERT_TRUE(1 == pattern_.length());
  return FindFirstCharacter(pattern_, subject, index);
}

//---------------------------------------------------------------------
// Linear Search Strategy
//---------------------------------------------------------------------

// Simple linear search for short patterns. Never bails out.
template <typename Char>
size_t StringSearch<Char>::LinearSearch(Vector subject, size_t index) {
  NBYTES_ASSERT_TRUE(pattern_.length() > 1);
  const size_t n = subject.length() - pattern_.length();
  for (size_t i = index; i <= n; i++) {
    i = FindFirstCharacter(pattern_, subject, i);
    if (i == subject.length()) return subject.length();
    NBYTES_ASSERT_TRUE(i <= n);

    bool matches = true;
    for (size_t j = 1; j < pattern_.length(); j++) {
      if (pattern_[j] != subject[i + j]) {
        matches = false;
        break;
      }
    }
    if (matches) {
      return i;
    }
  }
  return subject.length();
}

//---------------------------------------------------------------------
// Boyer-Moore string search
//---------------------------------------------------------------------

template <typename Char>
size_t StringSearch<Char>::BoyerMooreSearch(Vector subject,
                                            size_t start_index) {
  const size_t subject_length = subject.length();
  const size_t pattern_length = pattern_.length();
  // Only preprocess at most kBMMaxShift last characters of pattern.
  size_t start = start_;

  int *bad_char_occurrence = bad_char_shift_table_;
  int *good_suffix_shift = good_suffix_shift_table_ - start_;

  Char last_char = pattern_[pattern_length - 1];
  size_t index = start_index;
  // Continue search from i.
  while (index <= subject_length - pattern_length) {
    size_t j = pattern_length - 1;
    int c;
    while (last_char != (c = subject[index + j])) {
      int shift = j - CharOccurrence(bad_char_occurrence, c);
      index += shift;
      if (index > subject_length - pattern_length) {
        return subject.length();
      }
    }
    while (pattern_[j] == (c = subject[index + j])) {
      if (j == 0) {
        return index;
      }
      j--;
    }
    if (j < start) {
      // we have matched more than our tables allow us to be smart about.
      // Fall back on BMH shift.
      index +=
          pattern_length - 1 - CharOccurrence(bad_char_occurrence, last_char);
    } else {
      int gs_shift = good_suffix_shift[j + 1];
      int bc_occ = CharOccurrence(bad_char_occurrence, c);
      int shift = j - bc_occ;
      if (gs_shift > shift) {
        shift = gs_shift;
      }
      index += shift;
    }
  }

  return subject.length();
}

template <typename Char>
void StringSearch<Char>::PopulateBoyerMooreTable() {
  const size_t pattern_length = pattern_.length();
  // Only look at the last kBMMaxShift characters of pattern (from start_
  // to pattern_length).
  const size_t start = start_;
  const size_t length = pattern_length - start;

  // Biased tables so that we can use pattern indices as table indices,
  // even if we only cover the part of the pattern from offset start.
  int *shift_table = good_suffix_shift_table_ - start_;
  int *suffix_table = suffix_table_ - start_;

  // Initialize table.
  for (size_t i = start; i < pattern_length; i++) {
    shift_table[i] = length;
  }
  shift_table[pattern_length] = 1;
  suffix_table[pattern_length] = pattern_length + 1;

  if (pattern_length <= start) {
    return;
  }

  // Find suffixes.
  Char last_char = pattern_[pattern_length - 1];
  size_t suffix = pattern_length + 1;
  {
    size_t i = pattern_length;
    while (i > start) {
      Char c = pattern_[i - 1];
      while (suffix <= pattern_length && c != pattern_[suffix - 1]) {
        if (static_cast<size_t>(shift_table[suffix]) == length) {
          shift_table[suffix] = suffix - i;
        }
        suffix = suffix_table[suffix];
      }
      suffix_table[--i] = --suffix;
      if (suffix == pattern_length) {
        // No suffix to extend, so we check against last_char only.
        while ((i > start) && (pattern_[i - 1] != last_char)) {
          if (static_cast<size_t>(shift_table[pattern_length]) == length) {
            shift_table[pattern_length] = pattern_length - i;
          }
          suffix_table[--i] = pattern_length;
        }
        if (i > start) {
          suffix_table[--i] = --suffix;
        }
      }
    }
  }
  // Build shift table using suffixes.
  if (suffix < pattern_length) {
    for (size_t i = start; i <= pattern_length; i++) {
      if (static_cast<size_t>(shift_table[i]) == length) {
        shift_table[i] = suffix - start;
      }
      if (i == suffix) {
        suffix = suffix_table[suffix];
      }
    }
  }
}

//---------------------------------------------------------------------
// Boyer-Moore-Horspool string search.
//---------------------------------------------------------------------

template <typename Char>
size_t StringSearch<Char>::BoyerMooreHorspoolSearch(Vector subject,
                                                    size_t start_index) {
  const size_t subject_length = subject.length();
  const size_t pattern_length = pattern_.length();
  int *char_occurrences = bad_char_shift_table_;
  int64_t badness = -static_cast<int64_t>(pattern_length);

  // How bad we are doing without a good-suffix table.
  Char last_char = pattern_[pattern_length - 1];
  int last_char_shift =
      pattern_length - 1 - CharOccurrence(char_occurrences, last_char);

  // Perform search
  size_t index = start_index;  // No matches found prior to this index.
  while (index <= subject_length - pattern_length) {
    size_t j = pattern_length - 1;
    int subject_char;
    while (last_char != (subject_char = subject[index + j])) {
      int bc_occ = CharOccurrence(char_occurrences, subject_char);
      int shift = j - bc_occ;
      index += shift;
      badness += 1 - shift;  // at most zero, so badness cannot increase.
      if (index > subject_length - pattern_length) {
        return subject_length;
      }
    }
    j--;
    while (pattern_[j] == (subject[index + j])) {
      if (j == 0) {
        return index;
      }
      j--;
    }
    index += last_char_shift;
    // Badness increases by the number of characters we have
    // checked, and decreases by the number of characters we
    // can skip by shifting. It's a measure of how we are doing
    // compared to reading each character exactly once.
    badness += (pattern_length - j) - last_char_shift;
    if (badness > 0) {
      PopulateBoyerMooreTable();
      strategy_ = SearchStrategy::kBoyerMoore;
      return BoyerMooreSearch(subject, index);
    }
  }
  return subject.length();
}

template <typename Char>
void StringSearch<Char>::PopulateBoyerMooreHorspoolTable() {
  const size_t pattern_length = pattern_.length();

  int *bad_char_occurrence = bad_char_shift_table_;

  // Only preprocess at most kBMMaxShift last characters of pattern.
  const size_t start = start_;
  // Run forwards to populate bad_char_table, so that *last* instance
  // of character equivalence class is the one registered.
  // Notice: Doesn't include the last character.
  const size_t table_size = AlphabetSize();
  if (start == 0) {
    // All patterns less than kBMMaxShift in length.
    memset(bad_char_occurrence, -1, table_size * sizeof(*bad_char_occurrence));
  } else {
    for (size_t i = 0; i < table_size; i++) {
      bad_char_occurrence[i] = start - 1;
    }
  }
  for (size_t i = start; i < pattern_length - 1; i++) {
    Char c = pattern_[i];
    int bucket = (sizeof(Char) == 1) ? c : c % AlphabetSize();
    bad_char_occurrence[bucket] = i;
  }
}

//---------------------------------------------------------------------
// Linear string search with bailout to BMH.
//---------------------------------------------------------------------

// Simple linear search for short patterns, which bails out if the string
// isn't found very early in the subject. Upgrades to BoyerMooreHorspool.
template <typename Char>
size_t StringSearch<Char>::InitialSearch(Vector subject, size_t index) {
  const size_t pattern_length = pattern_.length();
  // Badness is a count of how much work we have done.  When we have
  // done enough work we decide it's probably worth switching to a better
  // algorithm.
  int64_t badness = -10 - (pattern_length << 2);

  // We know our pattern is at least 2 characters, we cache the first so
  // the common case of the first character not matching is faster.
  for (size_t i = index, n = subject.length() - pattern_length; i <= n; i++) {
    badness++;
    if (badness <= 0) {
      i = FindFirstCharacter(pattern_, subject, i);
      if (i == subject.length()) return subject.length();
      NBYTES_ASSERT_TRUE(i <= n);
      size_t j = 1;
      do {
        if (pattern_[j] != subject[i + j]) {
          break;
        }
        j++;
      } while (j < pattern_length);
      if (j == pattern_length) {
        return i;
      }
      badness += j;
    } else {
      PopulateBoyerMooreHorspoolTable();
      strategy_ = SearchStrategy::kBoyerMooreHorspool;
      return BoyerMooreHorspoolSearch(subject, i);
    }
  }
  return subject.length();
}

// Perform a single stand-alone search.
// If searching multiple times for the same pattern, a search
// object should be constructed once and the Search function then called
// for each search.
template <typename Char>
size_t SearchString(Vector<const Char> subject, Vector<const Char> pattern,
                    size_t start_index) {
  StringSearch<Char> search(pattern);
  return search.Search(subject, start_index);
}
}  // namespace stringsearch

template <typename Char>
size_t SearchString(const Char *haystack, size_t haystack_length,
                    const Char *needle, size_t needle_length,
                    size_t start_index, bool is_forward) {
  if (haystack_length < needle_length) return haystack_length;
  // To do a reverse search (lastIndexOf instead of indexOf) without redundant
  // code, create two vectors that are reversed views into the input strings.
  // For example, v_needle[0] would return the *last* character of the needle.
  // So we're searching for the first instance of rev(needle) in rev(haystack)
  stringsearch::Vector<const Char> v_needle(needle, needle_length, is_forward);
  stringsearch::Vector<const Char> v_haystack(haystack, haystack_length,
                                              is_forward);
  size_t diff = haystack_length - needle_length;
  size_t relative_start_index;
  if (is_forward) {
    relative_start_index = start_index;
  } else if (diff < start_index) {
    relative_start_index = 0;
  } else {
    relative_start_index = diff - start_index;
  }
  size_t pos =
      stringsearch::SearchString(v_haystack, v_needle, relative_start_index);
  if (pos == haystack_length) {
    // not found
    return pos;
  }
  return is_forward ? pos : (haystack_length - needle_length - pos);
}

template <size_t N>
size_t SearchString(const char *haystack, size_t haystack_length,
                    const char (&needle)[N]) {
  return SearchString(
      reinterpret_cast<const uint8_t *>(haystack), haystack_length,
      reinterpret_cast<const uint8_t *>(needle), N - 1, 0, true);
}

// ============================================================================
// Version metadata
#define NBYTES_VERSION "0.1.1"

enum {
  NBYTES_VERSION_MAJOR = 0,
  NBYTES_VERSION_MINOR = 1,
  NBYTES_VERSION_REVISION = 1,
};

}  // namespace nbytes

#endif  // NBYTES_H