Server IP : 66.29.132.122 / Your IP : 18.221.61.135 Web Server : LiteSpeed System : Linux business142.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64 User : admazpex ( 531) PHP Version : 7.2.34 Disable Function : NONE MySQL : OFF | cURL : ON | WGET : ON | Perl : ON | Python : ON | Sudo : OFF | Pkexec : OFF Directory : /proc/self/root/proc/thread-self/root/proc/thread-self/root/opt/alt/libicu65/usr/share/doc/alt-libicu65-devel/samples/ustring/ |
Upload File : |
/* ******************************************************************************* * * © 2016 and later: Unicode, Inc. and others. * License & terms of use: http://www.unicode.org/copyright.html#License * ******************************************************************************* ******************************************************************************* * * Copyright (C) 2000-2014, International Business Machines * Corporation and others. All Rights Reserved. * ******************************************************************************* * file name: ustring.c * encoding: UTF-8 * tab size: 8 (not used) * indentation:4 * * created on: 2000aug15 * created by: Markus W. Scherer * * This file contains sample code that illustrates the use of Unicode strings * with ICU. */ #define __STDC_FORMAT_MACROS 1 #include <inttypes.h> #include <stdio.h> #include "unicode/utypes.h" #include "unicode/uchar.h" #include "unicode/locid.h" #include "unicode/ustring.h" #include "unicode/ucnv.h" #include "unicode/unistr.h" using namespace icu; #ifndef UPRV_LENGTHOF #define UPRV_LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0])) #endif // helper functions -------------------------------------------------------- *** // default converter for the platform encoding static UConverter *cnv=NULL; static void printUString(const char *announce, const UChar *s, int32_t length) { static char out[200]; UChar32 c; int32_t i; UErrorCode errorCode=U_ZERO_ERROR; /* * Convert to the "platform encoding". See notes in printUnicodeString(). * ucnv_fromUChars(), like most ICU APIs understands length==-1 * to mean that the string is NUL-terminated. */ ucnv_fromUChars(cnv, out, sizeof(out), s, length, &errorCode); if(U_FAILURE(errorCode) || errorCode==U_STRING_NOT_TERMINATED_WARNING) { printf("%sproblem converting string from Unicode: %s\n", announce, u_errorName(errorCode)); return; } printf("%s%s {", announce, out); /* output the code points (not code units) */ if(length>=0) { /* s is not NUL-terminated */ for(i=0; i<length; /* U16_NEXT post-increments */) { U16_NEXT(s, i, length, c); printf(" %04x", c); } } else { /* s is NUL-terminated */ for(i=0; /* condition in loop body */; /* U16_NEXT post-increments */) { U16_NEXT(s, i, length, c); if(c==0) { break; } printf(" %04x", c); } } printf(" }\n"); } static void printUnicodeString(const char *announce, const UnicodeString &s) { static char out[200]; int32_t i, length; // output the string, converted to the platform encoding // Note for Windows: The "platform encoding" defaults to the "ANSI codepage", // which is different from the "OEM codepage" in the console window. // However, if you pipe the output into a file and look at it with Notepad // or similar, then "ANSI" characters will show correctly. // Production code should be aware of what encoding is required, // and use a UConverter or at least a charset name explicitly. out[s.extract(0, 99, out)]=0; printf("%s%s {", announce, out); // output the code units (not code points) length=s.length(); for(i=0; i<length; ++i) { printf(" %04x", s.charAt(i)); } printf(" }\n"); } // sample code for utf.h macros -------------------------------------------- *** static void demo_utf_h_macros() { static UChar input[]={ 0x0061, 0xd800, 0xdc00, 0xdbff, 0xdfff, 0x0062 }; UChar32 c; int32_t i; UBool isError; printf("\n* demo_utf_h_macros() -------------- ***\n\n"); printUString("iterate forward through: ", input, UPRV_LENGTHOF(input)); for(i=0; i<UPRV_LENGTHOF(input); /* U16_NEXT post-increments */) { /* Iterating forwards Codepoint at offset 0: U+0061 Codepoint at offset 1: U+10000 Codepoint at offset 3: U+10ffff Codepoint at offset 5: U+0062 */ printf("Codepoint at offset %d: U+", i); U16_NEXT(input, i, UPRV_LENGTHOF(input), c); printf("%04x\n", c); } puts(""); isError=FALSE; i=1; /* write position, gets post-incremented so needs to be in an l-value */ U16_APPEND(input, i, UPRV_LENGTHOF(input), 0x0062, isError); printUString("iterate backward through: ", input, UPRV_LENGTHOF(input)); for(i=UPRV_LENGTHOF(input); i>0; /* U16_PREV pre-decrements */) { U16_PREV(input, 0, i, c); /* Iterating backwards Codepoint at offset 5: U+0062 Codepoint at offset 3: U+10ffff Codepoint at offset 2: U+dc00 -- unpaired surrogate because lead surr. overwritten Codepoint at offset 1: U+0062 -- by this BMP code point Codepoint at offset 0: U+0061 */ printf("Codepoint at offset %d: U+%04x\n", i, c); } } // sample code for Unicode strings in C ------------------------------------ *** static void demo_C_Unicode_strings() { printf("\n* demo_C_Unicode_strings() --------- ***\n\n"); static const UChar text[]={ 0x41, 0x42, 0x43, 0 }; /* "ABC" */ static const UChar appendText[]={ 0x61, 0x62, 0x63, 0 }; /* "abc" */ static const UChar cmpText[]={ 0x61, 0x53, 0x73, 0x43, 0 }; /* "aSsC" */ UChar buffer[32]; int32_t compare; int32_t length=u_strlen(text); /* length=3 */ /* simple ANSI C-style functions */ buffer[0]=0; /* empty, NUL-terminated string */ u_strncat(buffer, text, 1); /* append just n=1 character ('A') */ u_strcat(buffer, appendText); /* buffer=="Aabc" */ length=u_strlen(buffer); /* length=4 */ printUString("should be \"Aabc\": ", buffer, -1); /* bitwise comparing buffer with text */ compare=u_strcmp(buffer, text); if(compare<=0) { printf("String comparison error, expected \"Aabc\" > \"ABC\"\n"); } /* Build "A<sharp s>C" in the buffer... */ u_strcpy(buffer, text); buffer[1]=0xdf; /* sharp s, case-compares equal to "ss" */ printUString("should be \"A<sharp s>C\": ", buffer, -1); /* Compare two strings case-insensitively using full case folding */ compare=u_strcasecmp(buffer, cmpText, U_FOLD_CASE_DEFAULT); if(compare!=0) { printf("String case insensitive comparison error, expected \"AbC\" to be equal to \"ABC\"\n"); } } // sample code for case mappings with C APIs -------------------------------- *** static void demoCaseMapInC() { /* * input= * "aB<capital sigma>" * "iI<small dotless i><capital dotted I> " * "<sharp s> <small lig. ffi>" * "<small final sigma><small sigma><capital sigma>" */ static const UChar input[]={ 0x61, 0x42, 0x3a3, 0x69, 0x49, 0x131, 0x130, 0x20, 0xdf, 0x20, 0xfb03, 0x3c2, 0x3c3, 0x3a3, 0 }; UChar buffer[32]; UErrorCode errorCode; UChar32 c; int32_t i, j, length; UBool isError; printf("\n* demoCaseMapInC() ----------------- ***\n\n"); /* * First, use simple case mapping functions which provide * 1:1 code point mappings without context/locale ID. * * Note that some mappings will not be "right" because some "real" * case mappings require context, depend on the locale ID, * and/or result in a change in the number of code points. */ printUString("input string: ", input, -1); /* uppercase */ isError=FALSE; for(i=j=0; j<UPRV_LENGTHOF(buffer) && !isError; /* U16_NEXT post-increments */) { U16_NEXT(input, i, INT32_MAX, c); /* without length because NUL-terminated */ if(c==0) { break; /* stop at terminating NUL, no need to terminate buffer */ } c=u_toupper(c); U16_APPEND(buffer, j, UPRV_LENGTHOF(buffer), c, isError); } printUString("simple-uppercased: ", buffer, j); /* lowercase */ isError=FALSE; for(i=j=0; j<UPRV_LENGTHOF(buffer) && !isError; /* U16_NEXT post-increments */) { U16_NEXT(input, i, INT32_MAX, c); /* without length because NUL-terminated */ if(c==0) { break; /* stop at terminating NUL, no need to terminate buffer */ } c=u_tolower(c); U16_APPEND(buffer, j, UPRV_LENGTHOF(buffer), c, isError); } printUString("simple-lowercased: ", buffer, j); /* titlecase */ isError=FALSE; for(i=j=0; j<UPRV_LENGTHOF(buffer) && !isError; /* U16_NEXT post-increments */) { U16_NEXT(input, i, INT32_MAX, c); /* without length because NUL-terminated */ if(c==0) { break; /* stop at terminating NUL, no need to terminate buffer */ } c=u_totitle(c); U16_APPEND(buffer, j, UPRV_LENGTHOF(buffer), c, isError); } printUString("simple-titlecased: ", buffer, j); /* case-fold/default */ isError=FALSE; for(i=j=0; j<UPRV_LENGTHOF(buffer) && !isError; /* U16_NEXT post-increments */) { U16_NEXT(input, i, INT32_MAX, c); /* without length because NUL-terminated */ if(c==0) { break; /* stop at terminating NUL, no need to terminate buffer */ } c=u_foldCase(c, U_FOLD_CASE_DEFAULT); U16_APPEND(buffer, j, UPRV_LENGTHOF(buffer), c, isError); } printUString("simple-case-folded/default: ", buffer, j); /* case-fold/Turkic */ isError=FALSE; for(i=j=0; j<UPRV_LENGTHOF(buffer) && !isError; /* U16_NEXT post-increments */) { U16_NEXT(input, i, INT32_MAX, c); /* without length because NUL-terminated */ if(c==0) { break; /* stop at terminating NUL, no need to terminate buffer */ } c=u_foldCase(c, U_FOLD_CASE_EXCLUDE_SPECIAL_I); U16_APPEND(buffer, j, UPRV_LENGTHOF(buffer), c, isError); } printUString("simple-case-folded/Turkic: ", buffer, j); /* * Second, use full case mapping functions which provide * 1:n code point mappings (n can be 0!) and are sensitive to context and locale ID. * * Note that lower/upper/titlecasing take a locale ID while case-folding * has bit flag options instead, by design of the Unicode SpecialCasing.txt UCD file. * * Also, string titlecasing requires a BreakIterator to find starts of words. * The sample code here passes in a NULL pointer; u_strToTitle() will open and close a default * titlecasing BreakIterator automatically. * For production code where many strings are titlecased it would be more efficient * to open a BreakIterator externally and pass it in. */ printUString("\ninput string: ", input, -1); /* lowercase/English */ errorCode=U_ZERO_ERROR; length=u_strToLower(buffer, UPRV_LENGTHOF(buffer), input, -1, "en", &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-lowercased/en: ", buffer, length); } else { printf("error in u_strToLower(en)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* lowercase/Turkish */ errorCode=U_ZERO_ERROR; length=u_strToLower(buffer, UPRV_LENGTHOF(buffer), input, -1, "tr", &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-lowercased/tr: ", buffer, length); } else { printf("error in u_strToLower(tr)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* uppercase/English */ errorCode=U_ZERO_ERROR; length=u_strToUpper(buffer, UPRV_LENGTHOF(buffer), input, -1, "en", &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-uppercased/en: ", buffer, length); } else { printf("error in u_strToUpper(en)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* uppercase/Turkish */ errorCode=U_ZERO_ERROR; length=u_strToUpper(buffer, UPRV_LENGTHOF(buffer), input, -1, "tr", &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-uppercased/tr: ", buffer, length); } else { printf("error in u_strToUpper(tr)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* titlecase/English */ errorCode=U_ZERO_ERROR; length=u_strToTitle(buffer, UPRV_LENGTHOF(buffer), input, -1, NULL, "en", &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-titlecased/en: ", buffer, length); } else { printf("error in u_strToTitle(en)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* titlecase/Turkish */ errorCode=U_ZERO_ERROR; length=u_strToTitle(buffer, UPRV_LENGTHOF(buffer), input, -1, NULL, "tr", &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-titlecased/tr: ", buffer, length); } else { printf("error in u_strToTitle(tr)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* case-fold/default */ errorCode=U_ZERO_ERROR; length=u_strFoldCase(buffer, UPRV_LENGTHOF(buffer), input, -1, U_FOLD_CASE_DEFAULT, &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-case-folded/default: ", buffer, length); } else { printf("error in u_strFoldCase(default)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } /* case-fold/Turkic */ errorCode=U_ZERO_ERROR; length=u_strFoldCase(buffer, UPRV_LENGTHOF(buffer), input, -1, U_FOLD_CASE_EXCLUDE_SPECIAL_I, &errorCode); if(U_SUCCESS(errorCode)) { printUString("full-case-folded/Turkic: ", buffer, length); } else { printf("error in u_strFoldCase(Turkic)=%" PRId32 " error=%s\n", length, u_errorName(errorCode)); } } // sample code for case mappings with C++ APIs ------------------------------ *** static void demoCaseMapInCPlusPlus() { /* * input= * "aB<capital sigma>" * "iI<small dotless i><capital dotted I> " * "<sharp s> <small lig. ffi>" * "<small final sigma><small sigma><capital sigma>" */ static const UChar input[]={ 0x61, 0x42, 0x3a3, 0x69, 0x49, 0x131, 0x130, 0x20, 0xdf, 0x20, 0xfb03, 0x3c2, 0x3c3, 0x3a3, 0 }; printf("\n* demoCaseMapInCPlusPlus() --------- ***\n\n"); UnicodeString s(input), t; const Locale &en=Locale::getEnglish(); Locale tr("tr"); /* * Full case mappings as in demoCaseMapInC(), using UnicodeString functions. * These functions modify the string object itself. * Since we want to keep the input string around, we copy it each time * and case-map the copy. */ printUnicodeString("input string: ", s); /* lowercase/English */ printUnicodeString("full-lowercased/en: ", (t=s).toLower(en)); /* lowercase/Turkish */ printUnicodeString("full-lowercased/tr: ", (t=s).toLower(tr)); /* uppercase/English */ printUnicodeString("full-uppercased/en: ", (t=s).toUpper(en)); /* uppercase/Turkish */ printUnicodeString("full-uppercased/tr: ", (t=s).toUpper(tr)); /* titlecase/English */ printUnicodeString("full-titlecased/en: ", (t=s).toTitle(NULL, en)); /* titlecase/Turkish */ printUnicodeString("full-titlecased/tr: ", (t=s).toTitle(NULL, tr)); /* case-folde/default */ printUnicodeString("full-case-folded/default: ", (t=s).foldCase(U_FOLD_CASE_DEFAULT)); /* case-folde/Turkic */ printUnicodeString("full-case-folded/Turkic: ", (t=s).foldCase(U_FOLD_CASE_EXCLUDE_SPECIAL_I)); } // sample code for UnicodeString storage models ----------------------------- *** static const UChar readonly[]={ 0x61, 0x31, 0x20ac }; static UChar writeable[]={ 0x62, 0x32, 0xdbc0, 0xdc01 // includes a surrogate pair for a supplementary code point }; static char out[100]; static void demoUnicodeStringStorage() { // These sample code lines illustrate how to use UnicodeString, and the // comments tell what happens internally. There are no APIs to observe // most of this programmatically, except for stepping into the code // with a debugger. // This is by design to hide such details from the user. int32_t i; printf("\n* demoUnicodeStringStorage() ------- ***\n\n"); // * UnicodeString with internally stored contents // instantiate a UnicodeString from a single code point // the few (2) UChars will be stored in the object itself UnicodeString one((UChar32)0x24001); // this copies the few UChars into the "two" object UnicodeString two=one; printf("length of short string copy: %d\n", two.length()); // set "one" to contain the 3 UChars from readonly // this setTo() variant copies the characters one.setTo(readonly, UPRV_LENGTHOF(readonly)); // * UnicodeString with allocated contents // build a longer string that will not fit into the object's buffer one+=UnicodeString(writeable, UPRV_LENGTHOF(writeable)); one+=one; one+=one; printf("length of longer string: %d\n", one.length()); // copying will use the same allocated buffer and increment the reference // counter two=one; printf("length of longer string copy: %d\n", two.length()); // * UnicodeString using readonly-alias to a const UChar array // construct a string that aliases a readonly buffer UnicodeString three(FALSE, readonly, UPRV_LENGTHOF(readonly)); printUnicodeString("readonly-alias string: ", three); // copy-on-write: any modification to the string results in // a copy to either the internal buffer or to a newly allocated one three.setCharAt(1, 0x39); printUnicodeString("readonly-aliasing string after modification: ", three); // the aliased array is not modified for(i=0; i<three.length(); ++i) { printf("readonly buffer[%d] after modifying its string: 0x%" PRId32 "\n", i, readonly[i]); } // setTo() readonly alias one.setTo(FALSE, writeable, UPRV_LENGTHOF(writeable)); // copying the readonly-alias object with fastCopyFrom() (new in ICU 2.4) // will readonly-alias the same buffer two.fastCopyFrom(one); printUnicodeString("fastCopyFrom(readonly alias of \"writeable\" array): ", two); printf("verify that a fastCopyFrom(readonly alias) uses the same buffer pointer: %d (should be 1)\n", one.getBuffer()==two.getBuffer()); // a normal assignment will clone the contents (new in ICU 2.4) two=one; printf("verify that a regular copy of a readonly alias uses a different buffer pointer: %d (should be 0)\n", one.getBuffer()==two.getBuffer()); // * UnicodeString using writeable-alias to a non-const UChar array UnicodeString four(writeable, UPRV_LENGTHOF(writeable), UPRV_LENGTHOF(writeable)); printUnicodeString("writeable-alias string: ", four); // a modification writes through to the buffer four.setCharAt(1, 0x39); for(i=0; i<four.length(); ++i) { printf("writeable-alias backing buffer[%d]=0x%" PRId32 " " "after modification\n", i, writeable[i]); } // a copy will not alias any more; // instead, it will get a copy of the contents into allocated memory two=four; two.setCharAt(1, 0x21); for(i=0; i<two.length(); ++i) { printf("writeable-alias backing buffer[%d]=0x%" PRId32 " after " "modification of string copy\n", i, writeable[i]); } // setTo() writeable alias, capacity==length one.setTo(writeable, UPRV_LENGTHOF(writeable), UPRV_LENGTHOF(writeable)); // grow the string - it will not fit into the backing buffer any more // and will get copied before modification one.append((UChar)0x40); // shrink it back so it would fit one.truncate(one.length()-1); // we still operate on the copy one.setCharAt(1, 0x25); printf("string after growing too much and then shrinking[1]=0x%" PRId32 "\n" " backing store for this[1]=0x%" PRId32 "\n", one.charAt(1), writeable[1]); // if we need it in the original buffer, then extract() to it // extract() does not do anything if the string aliases that same buffer // i=min(one.length(), length of array) if(one.length()<UPRV_LENGTHOF(writeable)) { i=one.length(); } else { i=UPRV_LENGTHOF(writeable); } one.extract(0, i, writeable); for(i=0; i<UPRV_LENGTHOF(writeable); ++i) { printf("writeable-alias backing buffer[%d]=0x%" PRId32 " after re-extract\n", i, writeable[i]); } } // sample code for UnicodeString instantiations ----------------------------- *** static void demoUnicodeStringInit() { // *** Make sure to read about invariant characters in utypes.h! *** // Initialization of Unicode strings from C literals works _only_ for // invariant characters! printf("\n* demoUnicodeStringInit() ---------- ***\n\n"); // the string literal is 32 chars long - this must be counted for the macro UnicodeString invariantOnly=UNICODE_STRING("such characters are safe 123 %-.", 32); /* * In C, we need two macros: one to declare the UChar[] array, and * one to populate it; the second one is a noop on platforms where * wchar_t is compatible with UChar and ASCII-based. * The length of the string literal must be counted for both macros. */ /* declare the invString array for the string */ U_STRING_DECL(invString, "such characters are safe 123 %-.", 32); /* populate it with the characters */ U_STRING_INIT(invString, "such characters are safe 123 %-.", 32); // compare the C and C++ strings printf("C and C++ Unicode strings are equal: %d\n", invariantOnly==UnicodeString(TRUE, invString, 32)); /* * convert between char * and UChar * strings that * contain only invariant characters */ static const char *cs1="such characters are safe 123 %-."; static UChar us1[40]; static char cs2[40]; u_charsToUChars(cs1, us1, 33); /* include the terminating NUL */ u_UCharsToChars(us1, cs2, 33); printf("char * -> UChar * -> char * with only " "invariant characters: \"%s\"\n", cs2); // initialize a UnicodeString from a string literal that contains // escape sequences written with invariant characters // do not forget to duplicate the backslashes for ICU to see them // then, count each double backslash only once! UnicodeString german=UNICODE_STRING( "Sch\\u00f6nes Auto: \\u20ac 11240.\\fPrivates Zeichen: \\U00102345\\n", 64). unescape(); printUnicodeString("german UnicodeString from unescaping:\n ", german); /* * C: convert and unescape a char * string with only invariant * characters to fill a UChar * string */ UChar buffer[200]; int32_t length; length=u_unescape( "Sch\\u00f6nes Auto: \\u20ac 11240.\\fPrivates Zeichen: \\U00102345\\n", buffer, UPRV_LENGTHOF(buffer)); printf("german C Unicode string from char * unescaping: (length %d)\n ", length); printUnicodeString("", UnicodeString(buffer)); } extern int main(int argc, const char *argv[]) { UErrorCode errorCode=U_ZERO_ERROR; // Note: Using a global variable for any object is not exactly thread-safe... // You can change this call to e.g. ucnv_open("UTF-8", &errorCode) if you pipe // the output to a file and look at it with a Unicode-capable editor. // This will currently affect only the printUString() function, see the code above. // printUnicodeString() could use this, too, by changing to an extract() overload // that takes a UConverter argument. cnv=ucnv_open(NULL, &errorCode); if(U_FAILURE(errorCode)) { fprintf(stderr, "error %s opening the default converter\n", u_errorName(errorCode)); return errorCode; } ucnv_setFromUCallBack(cnv, UCNV_FROM_U_CALLBACK_ESCAPE, UCNV_ESCAPE_C, NULL, NULL, &errorCode); if(U_FAILURE(errorCode)) { fprintf(stderr, "error %s setting the escape callback in the default converter\n", u_errorName(errorCode)); ucnv_close(cnv); return errorCode; } demo_utf_h_macros(); demo_C_Unicode_strings(); demoCaseMapInC(); demoCaseMapInCPlusPlus(); demoUnicodeStringStorage(); demoUnicodeStringInit(); ucnv_close(cnv); return 0; }