Current File : //usr/local/apps/dovecot/include/dovecot/unichar.h
#ifndef UNICHAR_H
#define UNICHAR_H

/* Character used to replace invalid input. */
#define UNICODE_REPLACEMENT_CHAR 0xfffd
#define UNICODE_REPLACEMENT_CHAR_UTF8 "\xEF\xBF\xBD"
#define UNICODE_REPLACEMENT_CHAR_UTF8_LEN \
	(sizeof(UNICODE_REPLACEMENT_CHAR_UTF8) - 1);
/* Horizontal ellipsis character ('...') */
#define UNICODE_HORIZONTAL_ELLIPSIS_CHAR 0x2026
#define UNICODE_HORIZONTAL_ELLIPSIS_CHAR_UTF8 "\xE2\x80\xA6"
#define UNICODE_HORIZONTAL_ELLIPSIS_CHAR_UTF8_LEN \
	(sizeof(UNICODE_HORIZONTAL_ELLIPSIS_CHAR_UTF8) - 1);

/* Characters >= base require surrogates */
#define UTF16_SURROGATE_BASE 0x10000

#define UTF16_SURROGATE_SHIFT 10
#define UTF16_SURROGATE_MASK 0x03ff
#define UTF16_SURROGATE_HIGH_FIRST 0xd800
#define UTF16_SURROGATE_HIGH_LAST 0xdbff
#define UTF16_SURROGATE_HIGH_MAX 0xdfff
#define UTF16_SURROGATE_LOW_FIRST 0xdc00
#define UTF16_SURROGATE_LOW_LAST 0xdfff

#define UTF16_SURROGATE_HIGH(chr) \
	(UTF16_SURROGATE_HIGH_FIRST + \
	 (((chr) - UTF16_SURROGATE_BASE) >> UTF16_SURROGATE_SHIFT))
#define UTF16_SURROGATE_LOW(chr) \
	(UTF16_SURROGATE_LOW_FIRST + \
	 (((chr) - UTF16_SURROGATE_BASE) & UTF16_SURROGATE_MASK))

/* Returns TRUE if given byte is ASCII character or the beginning of a
   multibyte UTF-8 sequence */
#define UTF8_IS_START_SEQ(b) \
	(((b) & 0x80) == 0 || ((b) & 0xC0) == 0xC0)

#define UTF8_REPLACEMENT_CHAR_LEN 3

#define UNICHAR_T_MAX 0x10ffff

#define UTF16_VALID_HIGH_SURROGATE(chr) (((chr) & 0xfffc00) == UTF16_SURROGATE_HIGH_FIRST)
#define UTF16_VALID_LOW_SURROGATE(chr) (((chr) & 0xfffc00) == UTF16_SURROGATE_LOW_FIRST)

typedef uint32_t unichar_t;
ARRAY_DEFINE_TYPE(unichars, unichar_t);

/* Normalize UTF8 input and append it to output buffer.
   Returns 0 if ok, -1 if input was invalid. Even if input was invalid,
   as much as possible should be added to output. */
typedef int normalizer_func_t(const void *input, size_t size,
			      buffer_t *output);

extern const unsigned char utf8_replacement_char[UTF8_REPLACEMENT_CHAR_LEN];
extern const uint8_t *const uni_utf8_non1_bytes;

static inline bool ATTR_PURE uni_is_valid_ucs4(unichar_t chr)
{
	return (!UTF16_VALID_HIGH_SURROGATE(chr) &&
		!UTF16_VALID_LOW_SURROGATE(chr) &&
		chr <= UNICHAR_T_MAX);
};

/* Returns number of characters in a NUL-terminated unicode string */
unsigned int uni_strlen(const unichar_t *str) ATTR_PURE;
/* Translates UTF-8 input to UCS-4 output. Returns 0 if ok, -1 if input was
   invalid */
int uni_utf8_to_ucs4(const char *input, ARRAY_TYPE(unichars) *output);
int uni_utf8_to_ucs4_n(const unsigned char *input, size_t size,
		       ARRAY_TYPE(unichars) *output);
/* Translates UCS-4 input to UTF-8 output. */
void uni_ucs4_to_utf8(const unichar_t *input, size_t len, buffer_t *output);
void uni_ucs4_to_utf8_c(unichar_t chr, buffer_t *output);

/* Returns char_bytes (>0) if *chr_r is set, 0 for incomplete trailing character,
   -1 for invalid input. */
int uni_utf8_get_char(const char *input, unichar_t *chr_r);
int uni_utf8_get_char_n(const void *input, size_t max_len, unichar_t *chr_r);
/* Returns number of characters in UTF-8 string. */
unsigned int uni_utf8_strlen(const char *input) ATTR_PURE;
/* Returns number of characters in UTF-8 input of specified size. */
unsigned int uni_utf8_strlen_n(const void *input, size_t size) ATTR_PURE;
/* Same as uni_utf8_strlen_n(), but if input ends with a partial UTF-8
   character, don't include it in the return value and set partial_pos_r to
   where the character begins. Otherwise partial_pos_r is set to the end
   of the input. */
unsigned int uni_utf8_partial_strlen_n(const void *input, size_t size,
				       size_t *partial_pos_r);

/* Returns the number of bytes belonging to this UTF-8 character. The given
   parameter is the first byte of the UTF-8 sequence. Invalid input is
   returned with length 1. */
static inline unsigned int ATTR_CONST
uni_utf8_char_bytes(unsigned char chr)
{
	/* 0x00 .. 0x7f are ASCII. 0x80 .. 0xC1 are invalid. */
	if (chr < (192 + 2))
		return 1;
	return uni_utf8_non1_bytes[chr - (192 + 2)];
}

/* Return given character in titlecase. */
unichar_t uni_ucs4_to_titlecase(unichar_t chr) ATTR_CONST;

/* Convert UTF-8 input to titlecase and decompose the titlecase characters to
   output buffer. Returns 0 if ok, -1 if input was invalid. This generates
   output that's compatible with i;unicode-casemap comparator. Invalid input
   is replaced with unicode replacement character (0xfffd). */
int uni_utf8_to_decomposed_titlecase(const void *input, size_t size,
				     buffer_t *output);

/* If input contains only valid UTF-8 characters, return TRUE without updating
   buf. If input contains invalid UTF-8 characters, replace them with unicode
   replacement character (0xfffd), write the output to buf and return FALSE. */
bool uni_utf8_get_valid_data(const unsigned char *input, size_t size,
			     buffer_t *buf) ATTR_WARN_UNUSED_RESULT;
/* Returns TRUE if string is valid UTF-8 input. */
bool uni_utf8_str_is_valid(const char *str);
/* Returns TRUE if data contains only valid UTF-8 input. */
bool uni_utf8_data_is_valid(const unsigned char *data, size_t size);
/* Returns the size of the data when truncated to be less than or equal to
   max_new_size, making sure UTF-8 character boundaries are respected. This only
   looks at the last character at the new boundary. */
size_t uni_utf8_data_truncate(const unsigned char *data, size_t old_size,
			      size_t max_new_size);

/* surrogate handling */
static inline unichar_t uni_join_surrogate(unichar_t high, unichar_t low)
{
	i_assert(UTF16_VALID_HIGH_SURROGATE(high) &&
		 UTF16_VALID_LOW_SURROGATE(low));

	return ((high - UTF16_SURROGATE_HIGH_FIRST)<<10) +
		(low - UTF16_SURROGATE_LOW_FIRST) +
		UTF16_SURROGATE_BASE;
}

static inline void uni_split_surrogate(unichar_t chr, unichar_t *high_r, unichar_t *low_r)
{
	i_assert(chr >= UTF16_SURROGATE_BASE && chr <= UNICHAR_T_MAX);
	i_assert(high_r != NULL && low_r != NULL);
	*high_r = UTF16_SURROGATE_HIGH(chr);
	*low_r = UTF16_SURROGATE_LOW(chr);
}
#endif