4 * Based on the flexible and economical utf8 decoder:
5 * http://bjoern.hoehrmann.de/utf-8/decoder/dfa/
7 * This is slightly more economical, the fastest way to decode utf8 is
8 * with a lookup table as in:
11 * if that fails, 2-byte lookup
12 * if that fails, 3-byte lookup
13 * if that fails, 4-byte lookup
15 * The following table can be generated with some interval trickery.
16 * consider an interval [a, b):
18 * a must be 0x80 or b must be 0xc0, lower 3 bits
20 * interval(a,b) = ((uint32_t)((a==0x80?0x40-b:-a)<<23))
22 * The failstate can be represented as interval(0x80,0x80), it's
23 * odd to see but this is a full state machine.
25 * The table than maps the corresponding sections as a serise of
28 * In this table the transition values are pre-multiplied with 16 to
29 * save a shift instruction for every byte, we throw away fillers
30 * which makes the table smaller.
32 * The first section of the table handles bytes with leading C
33 * The second section of the table handles bytes with leading D
34 * The third section of the table handles bytes with leading E
35 * The last section of the table handles bytes with leading F
37 * The values themselfs in the table are arranged so that when you
38 * left shift them by 6 to shift continuation characters into place, the
39 * new top bits tell you:
41 * 1 - if you keep going
42 * 2 - the range of valid values for the next byte
44 static const uint32_t utf8_tab[] = {
45 0xC0000002, 0xC0000003, 0xC0000004, 0xC0000005, 0xC0000006,
46 0xC0000007, 0xC0000008, 0xC0000009, 0xC000000A, 0xC000000B,
47 0xC000000C, 0xC000000D, 0xC000000E, 0xC000000F, 0xC0000010,
48 0xC0000011, 0xC0000012, 0xC0000013, 0xC0000014, 0xC0000015,
49 0xC0000016, 0xC0000017, 0xC0000018, 0xC0000019, 0xC000001A,
50 0xC000001B, 0xC000001C, 0xC000001D, 0xC000001E, 0xC000001F,
51 0xB3000000, 0xC3000001, 0xC3000002, 0xC3000003, 0xC3000004,
52 0xC3000005, 0xC3000006, 0xC3000007, 0xC3000008, 0xC3000009,
53 0xC300000A, 0xC300000B, 0xC300000C, 0xD300000D, 0xC300000E,
54 0xC300000F, 0xBB0C0000, 0xC30C0001, 0xC30C0002, 0xC30C0003,
58 int utf8_from(char *s, utf8ch_t ch) {
62 if ((unsigned)ch < 0x80) {
65 } else if ((unsigned)ch < 0x800) {
66 *s++ = 0xC0 | (ch >> 6);
67 *s = 0x80 | (ch & 0x3F);
69 } else if ((unsigned)ch < 0xD800 || (unsigned)ch - 0xE000 < 0x2000) {
70 *s++ = 0xE0 | (ch >> 12);
71 *s++ = 0x80 | ((ch >> 6) & 0x3F);
72 *s = 0x80 | (ch & 0x3F);
74 } else if ((unsigned)ch - 0x10000 < 0x100000) {
75 *s++ = 0xF0 | (ch >> 18);
76 *s++ = 0x80 | ((ch >> 12) & 0x3F);
77 *s++ = 0x80 | ((ch >> 6) & 0x3F);
78 *s = 0x80 | (ch & 0x3F);
84 int utf8_to(utf8ch_t *i, const unsigned char *s, size_t n) {
90 /* This is consistent with mbtowc behaviour. */
92 i = (utf8ch_t*)(void*)&i;
99 c = utf8_tab[*s++-0xC2U];
102 * Avoid excessive checks against n.
104 * When shifting state `n-1` times does not clear the high bit,
105 * then the value of `n` won't satisfy the condition to read a
106 * character as it will be insufficent.
108 if (n < 4 && ((c<<(6*n-6)) & (1U << 31)))
112 * The upper 6 state bits are negitive integer offset to a bound-check
113 * next byte equivlant to: ((b-0x80)+(b+offset))&~0x3f
115 if ((((*s>>3)-0x10)|((*s>>3)+((int32_t)c>>26))) & ~7)
118 for (j=2; j<3; j++) {
119 if (!((c = c<<6 | (*s++-0x80))&(1U<<31))) {
123 if (*s-0x80U >= 0x40)
127 *i = c<<6 | (*s++-0x80);