Function utf8_decode

include/fmt/format.h:579–614

A public domain branchless UTF-8 decoder by Christopher Wellons: https://github.com/skeeto/branchless-utf8 Decode the next character, c, from s, reporting errors in e. * * Since this is a branchless decoder, four bytes will be read from the * buffer regardless of the actual length of the next character. This * means the buffer _must_ have at least three bytes of zero padding * following the e

 * error, but it will always advance at least one byte.
 */
FMT_CONSTEXPR inline auto utf8_decode(const char* s, uint32_t* c, int* e)
    -> const char* {
  constexpr int masks[] = {0x00, 0x7f, 0x1f, 0x0f, 0x07};
  constexpr uint32_t mins[] = {4194304, 0, 128, 2048, 65536};
  constexpr int shiftc[] = {0, 18, 12, 6, 0};
  constexpr int shifte[] = {0, 6, 4, 2, 0};

  int len = "\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\0\0\0\0\0\0\0\0\2\2\2\2\3\3\4"
      [static_cast<unsigned char>(*s) >> 3];
  // Compute the pointer to the next character early so that the next
  // iteration can start working on the next character. Neither Clang
  // nor GCC figure out this reordering on their own.
  const char* next = s + len + !len;

  using uchar = unsigned char;

  // Assume a four-byte character and load four bytes. Unused bits are
  // shifted out.
  *c = uint32_t(uchar(s[0]) & masks[len]) << 18;
  *c |= uint32_t(uchar(s[1]) & 0x3f) << 12;
  *c |= uint32_t(uchar(s[2]) & 0x3f) << 6;
  *c |= uint32_t(uchar(s[3]) & 0x3f) << 0;
  *c >>= shiftc[len];

  // Accumulate the various error conditions.
  *e = (*c < mins[len]) << 6;       // non-canonical encoding
  *e |= ((*c >> 11) == 0x1b) << 7;  // surrogate half?
  *e |= (*c > 0x10FFFF) << 8;       // out of range?
  *e |= (uchar(s[1]) & 0xc0) >> 2;
  *e |= (uchar(s[2]) & 0xc0) >> 4;
  *e |= uchar(s[3]) >> 6;
  *e ^= 0x2a;  // top two bits of each tail byte correct?
  *e >>= shifte[len];

  return next;
}

constexpr FMT_INLINE_VARIABLE uint32_t invalid_code_point = ~uint32_t();