VVVVVV/desktop_version/src/UTF8.c

203 lines
4.8 KiB
C

#include "UTF8.h"
#define STARTS_0(byte) ((byte & 0x80) == 0x00)
#define STARTS_10(byte) ((byte & 0xC0) == 0x80)
#define STARTS_110(byte) ((byte & 0xE0) == 0xC0)
#define STARTS_1110(byte) ((byte & 0xF0) == 0xE0)
#define STARTS_11110(byte) ((byte & 0xF8) == 0xF0)
#define TAKE(byte, nbits) (byte & ((1 << nbits)-1))
static inline bool is_illegal(uint32_t codepoint)
{
return (codepoint >= 0xD800 && codepoint <= 0xDFFF) || codepoint > 0x10FFFF;
}
uint32_t UTF8_peek_next(const char* s_str, uint8_t* codepoint_nbytes)
{
/* Get the next codepoint from a string, but instead of advancing the
* pointer, give the number of bytes the index will need to advance. */
if (s_str == NULL)
{
return 0;
}
// Pointer conversion to avoid all those brilliant signedness plot twists...
const unsigned char* str = (const unsigned char*) s_str;
uint32_t codepoint;
*codepoint_nbytes = 1;
if (STARTS_0(str[0]))
{
// 0xxx xxxx - ASCII
codepoint = str[0];
}
else if (STARTS_10(str[0]))
{
// 10xx xxxx - unexpected continuation byte
codepoint = 0xFFFD;
}
else if (STARTS_110(str[0]))
{
// 110x xxxx - 2-byte sequence
if (!STARTS_10(str[1]))
{
codepoint = 0xFFFD;
}
else
{
codepoint =
(TAKE(str[0], 5) << 6) |
(TAKE(str[1], 6));
*codepoint_nbytes = 2;
}
}
else if (STARTS_1110(str[0]))
{
// 1110 xxxx - 3-byte sequence
if (!STARTS_10(str[1]) || !STARTS_10(str[2]))
{
codepoint = 0xFFFD;
}
else
{
codepoint =
(TAKE(str[0], 4) << 12) |
(TAKE(str[1], 6) << 6) |
(TAKE(str[2], 6));
*codepoint_nbytes = 3;
}
}
else if (STARTS_11110(str[0]))
{
// 1111 0xxx - 4-byte sequence
if (!STARTS_10(str[1]) || !STARTS_10(str[2]) || !STARTS_10(str[3]))
{
codepoint = 0xFFFD;
}
else
{
codepoint =
(TAKE(str[0], 3) << 18) |
(TAKE(str[1], 6) << 12) |
(TAKE(str[2], 6) << 6) |
(TAKE(str[3], 6));
*codepoint_nbytes = 4;
}
}
else
{
// 1111 1xxx - invalid
codepoint = 0xFFFD;
}
// Overlong sequence?
if (
(codepoint <= 0x7F && *codepoint_nbytes > 1) ||
(codepoint > 0x7F && codepoint <= 0x7FF && *codepoint_nbytes > 2) ||
(codepoint > 0x7FF && codepoint <= 0xFFFF && *codepoint_nbytes > 3)
) {
codepoint = 0xFFFD;
}
// UTF-16 surrogates are invalid, so are codepoints after 10FFFF
if (is_illegal(codepoint))
{
codepoint = 0xFFFD;
}
return codepoint;
}
uint32_t UTF8_next(const char** p_str)
{
/* Get the next codepoint from a string, and advance the pointer.
* Example usage:
*
* const char* str = "asdf";
* uint32_t codepoint;
* while ((codepoint = UTF8_next(&str)))
* {
* // you have a codepoint congrats
* }
*/
if (p_str == NULL)
{
return 0;
}
uint8_t codepoint_nbytes;
uint32_t codepoint = UTF8_peek_next(*p_str, &codepoint_nbytes);
*p_str += codepoint_nbytes;
return codepoint;
}
UTF8_encoding UTF8_encode(uint32_t codepoint)
{
UTF8_encoding enc = {0};
// Pretend the bytes array is unsigned...
unsigned char* bytes = (unsigned char*) &enc.bytes;
if (is_illegal(codepoint))
{
codepoint = 0xFFFD;
enc.error = true;
}
if (codepoint <= 0x7F)
{
enc.nbytes = 1;
bytes[0] = codepoint;
}
else if (codepoint <= 0x7FF)
{
enc.nbytes = 2;
bytes[0] = 0xC0 | (codepoint >> 6);
bytes[1] = 0x80 | (codepoint & 0x3F);
}
else if (codepoint <= 0xFFFF)
{
enc.nbytes = 3;
bytes[0] = 0xE0 | (codepoint >> 12);
bytes[1] = 0x80 | ((codepoint >> 6) & 0x3F);
bytes[2] = 0x80 | (codepoint & 0x3F);
}
else
{
enc.nbytes = 4;
bytes[0] = 0xF0 | (codepoint >> 18);
bytes[1] = 0x80 | ((codepoint >> 12) & 0x3F);
bytes[2] = 0x80 | ((codepoint >> 6) & 0x3F);
bytes[3] = 0x80 | (codepoint & 0x3F);
}
return enc;
}
size_t UTF8_total_codepoints(const char* str)
{
size_t total = 0;
while (UTF8_next(&str))
{
total++;
}
return total;
}
size_t UTF8_backspace(const char* str, size_t len)
{
/* Given a string of length len,
* give the new length after removing the last character.
* In other words, the index at which to write a \0 byte. */
for (len -= 1; len > 0; len--)
{
if (!STARTS_10(str[len]))
{
break;
}
}
return len;
}