Logo Search packages:      
Sourcecode: icu version File versions

uscript.h File Reference


Detailed Description

C API: Unicode Script Information.

Definition in file uscript.h.

#include "unicode/utypes.h"

Go to the source code of this file.

Enumerations

enum  UScriptCode {
  USCRIPT_INVALID_CODE = -1, USCRIPT_COMMON = 0, USCRIPT_INHERITED = 1, USCRIPT_ARABIC = 2,
  USCRIPT_ARMENIAN = 3, USCRIPT_BENGALI = 4, USCRIPT_BOPOMOFO = 5, USCRIPT_CHEROKEE = 6,
  USCRIPT_COPTIC = 7, USCRIPT_CYRILLIC = 8, USCRIPT_DESERET = 9, USCRIPT_DEVANAGARI = 10,
  USCRIPT_ETHIOPIC = 11, USCRIPT_GEORGIAN = 12, USCRIPT_GOTHIC = 13, USCRIPT_GREEK = 14,
  USCRIPT_GUJARATI = 15, USCRIPT_GURMUKHI = 16, USCRIPT_HAN = 17, USCRIPT_HANGUL = 18,
  USCRIPT_HEBREW = 19, USCRIPT_HIRAGANA = 20, USCRIPT_KANNADA = 21, USCRIPT_KATAKANA = 22,
  USCRIPT_KHMER = 23, USCRIPT_LAO = 24, USCRIPT_LATIN = 25, USCRIPT_MALAYALAM = 26,
  USCRIPT_MONGOLIAN = 27, USCRIPT_MYANMAR = 28, USCRIPT_OGHAM = 29, USCRIPT_OLD_ITALIC = 30,
  USCRIPT_ORIYA = 31, USCRIPT_RUNIC = 32, USCRIPT_SINHALA = 33, USCRIPT_SYRIAC = 34,
  USCRIPT_TAMIL = 35, USCRIPT_TELUGU = 36, USCRIPT_THAANA = 37, USCRIPT_THAI = 38,
  USCRIPT_TIBETAN = 39, USCRIPT_CANADIAN_ABORIGINAL = 40, USCRIPT_UCAS = USCRIPT_CANADIAN_ABORIGINAL, USCRIPT_YI = 41,
  USCRIPT_TAGALOG = 42, USCRIPT_HANUNOO = 43, USCRIPT_BUHID = 44, USCRIPT_TAGBANWA = 45,
  USCRIPT_BRAILLE = 46, USCRIPT_CYPRIOT = 47, USCRIPT_LIMBU = 48, USCRIPT_LINEAR_B = 49,
  USCRIPT_OSMANYA = 50, USCRIPT_SHAVIAN = 51, USCRIPT_TAI_LE = 52, USCRIPT_UGARITIC = 53,
  USCRIPT_KATAKANA_OR_HIRAGANA = 54, USCRIPT_BUGINESE = 55, USCRIPT_GLAGOLITIC = 56, USCRIPT_KHAROSHTHI = 57,
  USCRIPT_SYLOTI_NAGRI = 58, USCRIPT_NEW_TAI_LUE = 59, USCRIPT_TIFINAGH = 60, USCRIPT_OLD_PERSIAN = 61,
  USCRIPT_BALINESE = 62, USCRIPT_BATAK = 63, USCRIPT_BLISSYMBOLS = 64, USCRIPT_BRAHMI = 65,
  USCRIPT_CHAM = 66, USCRIPT_CIRTH = 67, USCRIPT_OLD_CHURCH_SLAVONIC_CYRILLIC = 68, USCRIPT_DEMOTIC_EGYPTIAN = 69,
  USCRIPT_HIERATIC_EGYPTIAN = 70, USCRIPT_EGYPTIAN_HIEROGLYPHS = 71, USCRIPT_KHUTSURI = 72, USCRIPT_SIMPLIFIED_HAN = 73,
  USCRIPT_TRADITIONAL_HAN = 74, USCRIPT_PAHAWH_HMONG = 75, USCRIPT_OLD_HUNGARIAN = 76, USCRIPT_HARAPPAN_INDUS = 77,
  USCRIPT_JAVANESE = 78, USCRIPT_KAYAH_LI = 79, USCRIPT_LATIN_FRAKTUR = 80, USCRIPT_LATIN_GAELIC = 81,
  USCRIPT_LEPCHA = 82, USCRIPT_LINEAR_A = 83, USCRIPT_MANDAEAN = 84, USCRIPT_MAYAN_HIEROGLYPHS = 85,
  USCRIPT_MEROITIC = 86, USCRIPT_NKO = 87, USCRIPT_ORKHON = 88, USCRIPT_OLD_PERMIC = 89,
  USCRIPT_PHAGS_PA = 90, USCRIPT_PHOENICIAN = 91, USCRIPT_PHONETIC_POLLARD = 92, USCRIPT_RONGORONGO = 93,
  USCRIPT_SARATI = 94, USCRIPT_ESTRANGELO_SYRIAC = 95, USCRIPT_WESTERN_SYRIAC = 96, USCRIPT_EASTERN_SYRIAC = 97,
  USCRIPT_TENGWAR = 98, USCRIPT_VAI = 99, USCRIPT_VISIBLE_SPEECH = 100, USCRIPT_CUNEIFORM = 101,
  USCRIPT_UNWRITTEN_LANGUAGES = 102, USCRIPT_UNKNOWN = 103, USCRIPT_CARIAN = 104, USCRIPT_JAPANESE = 105,
  USCRIPT_LANNA = 106, USCRIPT_LYCIAN = 107, USCRIPT_LYDIAN = 108, USCRIPT_OL_CHIKI = 109,
  USCRIPT_REJANG = 110, USCRIPT_SAURASHTRA = 111, USCRIPT_SIGN_WRITING = 112, USCRIPT_SUNDANESE = 113,
  USCRIPT_MOON = 114, USCRIPT_MEITEI_MAYEK = 115, USCRIPT_IMPERIAL_ARAMAIC = 116, USCRIPT_AVESTAN = 117,
  USCRIPT_CHAKMA = 118, USCRIPT_KOREAN = 119, USCRIPT_KAITHI = 120, USCRIPT_MANICHAEAN = 121,
  USCRIPT_INSCRIPTIONAL_PAHLAVI = 122, USCRIPT_PSALTER_PAHLAVI = 123, USCRIPT_BOOK_PAHLAVI = 124, USCRIPT_INSCRIPTIONAL_PARTHIAN = 125,
  USCRIPT_SAMARITAN = 126, USCRIPT_TAI_VIET = 127, USCRIPT_MATHEMATICAL_NOTATION = 128, USCRIPT_SYMBOLS = 129,
  USCRIPT_CODE_LIMIT = 130
}

Functions

U_STABLE int32_t U_EXPORT2 uscript_getCode (const char *nameOrAbbrOrLocale, UScriptCode *fillIn, int32_t capacity, UErrorCode *err)
U_STABLE const char *U_EXPORT2 uscript_getName (UScriptCode scriptCode)
U_STABLE UScriptCode U_EXPORT2 uscript_getScript (UChar32 codepoint, UErrorCode *err)
U_STABLE const char *U_EXPORT2 uscript_getShortName (UScriptCode scriptCode)


Generated by  Doxygen 1.6.0   Back to index