зеркало из https://github.com/github/putty.git
154 строки
4.2 KiB
C
154 строки
4.2 KiB
C
/*
|
|
* charset.h - header file for general character set conversion
|
|
* routines.
|
|
*/
|
|
|
|
#ifndef charset_charset_h
|
|
#define charset_charset_h
|
|
|
|
#include <stddef.h>
|
|
|
|
/*
|
|
* Enumeration that lists all the multibyte or single-byte
|
|
* character sets known to this library.
|
|
*/
|
|
typedef enum {
|
|
CS_NONE, /* used for reporting errors, etc */
|
|
CS_ISO8859_1,
|
|
CS_ISO8859_1_X11, /* X font encoding with VT100 glyphs */
|
|
CS_ISO8859_2,
|
|
CS_ISO8859_3,
|
|
CS_ISO8859_4,
|
|
CS_ISO8859_5,
|
|
CS_ISO8859_6,
|
|
CS_ISO8859_7,
|
|
CS_ISO8859_8,
|
|
CS_ISO8859_9,
|
|
CS_ISO8859_10,
|
|
CS_ISO8859_11,
|
|
CS_ISO8859_13,
|
|
CS_ISO8859_14,
|
|
CS_ISO8859_15,
|
|
CS_ISO8859_16,
|
|
CS_CP437,
|
|
CS_CP850,
|
|
CS_CP1250,
|
|
CS_CP1251,
|
|
CS_CP1252,
|
|
CS_CP1253,
|
|
CS_CP1254,
|
|
CS_CP1255,
|
|
CS_CP1256,
|
|
CS_CP1257,
|
|
CS_CP1258,
|
|
CS_KOI8_R,
|
|
CS_KOI8_U,
|
|
CS_MAC_ROMAN,
|
|
CS_MAC_TURKISH,
|
|
CS_MAC_CROATIAN,
|
|
CS_MAC_ICELAND,
|
|
CS_MAC_ROMANIAN,
|
|
CS_MAC_GREEK,
|
|
CS_MAC_CYRILLIC,
|
|
CS_MAC_THAI,
|
|
CS_MAC_CENTEURO,
|
|
CS_MAC_SYMBOL,
|
|
CS_MAC_DINGBATS,
|
|
CS_MAC_ROMAN_OLD,
|
|
CS_MAC_CROATIAN_OLD,
|
|
CS_MAC_ICELAND_OLD,
|
|
CS_MAC_ROMANIAN_OLD,
|
|
CS_MAC_GREEK_OLD,
|
|
CS_MAC_CYRILLIC_OLD,
|
|
CS_MAC_UKRAINE,
|
|
CS_MAC_VT100,
|
|
CS_MAC_VT100_OLD,
|
|
CS_VISCII,
|
|
CS_HP_ROMAN8,
|
|
CS_DEC_MCS,
|
|
CS_UTF8
|
|
} charset_t;
|
|
|
|
typedef struct {
|
|
unsigned long s0;
|
|
} charset_state;
|
|
|
|
/*
|
|
* Routine to convert a MB/SB character set to Unicode.
|
|
*
|
|
* This routine accepts some number of bytes, updates a state
|
|
* variable, and outputs some number of Unicode characters. There
|
|
* are no guarantees. You can't even guarantee that at most one
|
|
* Unicode character will be output per byte you feed in; for
|
|
* example, suppose you're reading UTF-8, you've seen E1 80, and
|
|
* then you suddenly see FE. Now you need to output _two_ error
|
|
* characters - one for the incomplete sequence E1 80, and one for
|
|
* the completely invalid UTF-8 byte FE.
|
|
*
|
|
* Returns the number of wide characters output; will never output
|
|
* more than the size of the buffer (as specified on input).
|
|
* Advances the `input' pointer and decrements `inlen', to indicate
|
|
* how far along the input string it got.
|
|
*
|
|
* The sequence of `errlen' wide characters pointed to by `errstr'
|
|
* will be used to indicate a conversion error. If `errstr' is
|
|
* NULL, `errlen' will be ignored, and the library will choose
|
|
* something sensible to do on its own. For Unicode, this will be
|
|
* U+FFFD (REPLACEMENT CHARACTER).
|
|
*/
|
|
|
|
int charset_to_unicode(char **input, int *inlen, wchar_t *output, int outlen,
|
|
int charset, charset_state *state,
|
|
const wchar_t *errstr, int errlen);
|
|
|
|
/*
|
|
* Routine to convert Unicode to an MB/SB character set.
|
|
*
|
|
* This routine accepts some number of Unicode characters, updates
|
|
* a state variable, and outputs some number of bytes.
|
|
*
|
|
* Returns the number of bytes characters output; will never output
|
|
* more than the size of the buffer (as specified on input), and
|
|
* will never output a partial MB character. Advances the `input'
|
|
* pointer and decrements `inlen', to indicate how far along the
|
|
* input string it got.
|
|
*
|
|
* The sequence of `errlen' characters pointed to by `errstr' will
|
|
* be used to indicate a conversion error. If `errstr' is NULL,
|
|
* `errlen' will be ignored, and the library will choose something
|
|
* sensible to do on its own (which will vary depending on the
|
|
* output charset).
|
|
*/
|
|
|
|
int charset_from_unicode(wchar_t **input, int *inlen, char *output, int outlen,
|
|
int charset, charset_state *state,
|
|
const char *errstr, int errlen);
|
|
|
|
/*
|
|
* Convert X11 encoding names to and from our charset identifiers.
|
|
*/
|
|
const char *charset_to_xenc(int charset);
|
|
int charset_from_xenc(const char *name);
|
|
|
|
/*
|
|
* Convert MIME encoding names to and from our charset identifiers.
|
|
*/
|
|
const char *charset_to_mimeenc(int charset);
|
|
int charset_from_mimeenc(const char *name);
|
|
|
|
/*
|
|
* Convert our own encoding names to and from our charset
|
|
* identifiers.
|
|
*/
|
|
const char *charset_to_localenc(int charset);
|
|
int charset_from_localenc(const char *name);
|
|
int charset_localenc_nth(int n);
|
|
|
|
/*
|
|
* Convert Mac OS script/region/font to our charset identifiers.
|
|
*/
|
|
int charset_from_macenc(int script, int region, int sysvers,
|
|
const char *fontname);
|
|
|
|
#endif /* charset_charset_h */
|