/******************************************************************** * COPYRIGHT: * Copyright (c) 1997-2010, International Business Machines Corporation * and others. All Rights Reserved. ********************************************************************/ /******************************************************************************* * * File CNMDPTST.C * * Madhu Katragadda Creation * Modification History: * * Date Name Description * 06/24/99 helena Integrated Alan's NF enhancements and Java2 bug fixes ******************************************************************************* */ /* C DEPTH TEST FOR NUMBER FORMAT */ #include "unicode/utypes.h" #if !UCONFIG_NO_FORMATTING #include "unicode/ucurr.h" #include "unicode/uloc.h" #include "unicode/unum.h" #include "unicode/ustring.h" #include "cintltst.h" #include "cnmdptst.h" #include "cmemory.h" #include "cstring.h" #include "ulist.h" #define CHECK(status,str) if (U_FAILURE(status)) { log_err("FAIL: %s\n", str); return; } void addNumFrDepTest(TestNode** root); static void TestCurrencyPreEuro(void); static void TestCurrencyObject(void); void addNumFrDepTest(TestNode** root) { addTest(root, &TestPatterns, "tsformat/cnmdptst/TestPatterns"); addTest(root, &TestQuotes, "tsformat/cnmdptst/TestQuotes"); addTest(root, &TestExponential, "tsformat/cnmdptst/TestExponential"); addTest(root, &TestCurrencySign, "tsformat/cnmdptst/TestCurrencySign"); addTest(root, &TestCurrency, "tsformat/cnmdptst/TestCurrency"); addTest(root, &TestCurrencyPreEuro, "tsformat/cnmdptst/TestCurrencyPreEuro"); addTest(root, &TestCurrencyObject, "tsformat/cnmdptst/TestCurrencyObject"); addTest(root, &TestRounding487, "tsformat/cnmdptst/TestRounding487"); addTest(root, &TestDoubleAttribute, "tsformat/cnmdptst/TestDoubleAttribute"); addTest(root, &TestSecondaryGrouping, "tsformat/cnmdptst/TestSecondaryGrouping"); addTest(root, &TestCurrencyKeywords, "tsformat/cnmdptst/TestCurrencyKeywords"); addTest(root, &TestRounding5350, "tsformat/cnmdptst/TestRounding5350"); addTest(root, &TestGetKeywordValuesForLocale, "tsformat/cnmdptst/TestGetKeywordValuesForLocale"); } /*Test Various format patterns*/ static void TestPatterns(void) { int32_t pat_length, i, lneed; UNumberFormat *fmt; UChar upat[5]; UChar unewpat[5]; UChar unum[5]; UChar *unewp=NULL; UChar *str=NULL; UErrorCode status = U_ZERO_ERROR; const char* pat[] = { "#.#", "#.", ".#", "#" }; const char* newpat[] = { "#0.#", "#0.", "#.0", "#" }; const char* num[] = { "0", "0.", ".0", "0" }; log_verbose("\nTesting different format patterns\n"); pat_length = sizeof(pat) / sizeof(pat[0]); for (i=0; i < pat_length; ++i) { status = U_ZERO_ERROR; u_uastrcpy(upat, pat[i]); fmt= unum_open(UNUM_IGNORE,upat, u_strlen(upat), "en_US",NULL, &status); if (U_FAILURE(status)) { log_err_status(status, "FAIL: Number format constructor failed for pattern %s -> %s\n", pat[i], u_errorName(status)); continue; } lneed=0; lneed=unum_toPattern(fmt, FALSE, NULL, lneed, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status= U_ZERO_ERROR; unewp=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_toPattern(fmt, FALSE, unewp, lneed+1, &status); } if(U_FAILURE(status)){ log_err("FAIL: Number format extracting the pattern failed for %s\n", pat[i]); } u_uastrcpy(unewpat, newpat[i]); if(u_strcmp(unewp, unewpat) != 0) log_err("FAIL: Pattern %s should be transmute to %s; %s seen instead\n", pat[i], newpat[i], austrdup(unewp) ); lneed=0; lneed=unum_format(fmt, 0, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_format(fmt, 0, str, lneed+1, NULL, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); } u_uastrcpy(unum, num[i]); if (u_strcmp(str, unum) != 0) { log_err("FAIL: Pattern %s should format zero as %s; %s Seen instead\n", pat[i], num[i], austrdup(str) ); } free(unewp); free(str); unum_close(fmt); } } /* Test the handling of quotes*/ static void TestQuotes(void) { int32_t lneed; UErrorCode status=U_ZERO_ERROR; UChar pat[15]; UChar res[15]; UChar *str=NULL; UNumberFormat *fmt; char tempBuf[256]; log_verbose("\nTestting the handling of quotes in number format\n"); u_uastrcpy(pat, "a'fo''o'b#"); fmt =unum_open(UNUM_IGNORE,pat, u_strlen(pat), "en_US",NULL, &status); if(U_FAILURE(status)){ log_err_status(status, "Error in number format costruction using pattern \"a'fo''o'b#\" -> %s\n", u_errorName(status)); } lneed=0; lneed=unum_format(fmt, 123, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_format(fmt, 123, str, lneed+1, NULL, &status); } if(U_FAILURE(status) || !str) { log_err_status(status, "Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); return; } log_verbose("Pattern \"%s\" \n", u_austrcpy(tempBuf, pat) ); log_verbose("Format 123 -> %s\n", u_austrcpy(tempBuf, str) ); u_uastrcpy(res, "afo'ob123"); if(u_strcmp(str, res) != 0) log_err("FAIL: Expected afo'ob123"); free(str); unum_close(fmt); u_uastrcpy(pat, ""); u_uastrcpy(pat, "a''b#"); fmt =unum_open(UNUM_IGNORE,pat, u_strlen(pat), "en_US",NULL, &status); if(U_FAILURE(status)){ log_err("Error in number format costruction using pattern \"a''b#\"\n"); } lneed=0; lneed=unum_format(fmt, 123, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_format(fmt, 123, str, lneed+1, NULL, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); } log_verbose("Pattern \"%s\" \n", u_austrcpy(tempBuf, pat) ); log_verbose("Format 123 -> %s\n", u_austrcpy(tempBuf, str) ); u_uastrcpy(res, ""); u_uastrcpy(res, "a'b123"); if(u_strcmp(str, res) != 0) log_err("FAIL: Expected a'b123\n"); free(str); unum_close(fmt); } /* Test exponential pattern*/ static void TestExponential(void) { int32_t pat_length, val_length, lval_length; int32_t ival, ilval, p, v, lneed; UNumberFormat *fmt; int32_t ppos; UChar *upat; UChar pattern[20]; UChar *str=NULL; UChar uvalfor[20], ulvalfor[20]; char tempMsgBug[256]; double a; UErrorCode status = U_ZERO_ERROR; #ifdef OS390 static const double val[] = { 0.01234, 123456789, 1.23e75, -3.141592653e-78 }; #else static const double val[] = { 0.01234, 123456789, 1.23e300, -3.141592653e-271 }; #endif static const char* pat[] = { "0.####E0", "00.000E00", "##0.######E000", "0.###E0;[0.###E0]" }; static const int32_t lval[] = { 0, -1, 1, 123456789 }; static const char* valFormat[] = { "1.234E-2", "1.2346E8", "1.23E300", "-3.1416E-271", "12.340E-03", "12.346E07", "12.300E299", "-31.416E-272", "12.34E-003", "123.4568E006", "1.23E300", "-314.1593E-273", "1.234E-2", "1.235E8", "1.23E300", "[3.142E-271]" }; static const char* lvalFormat[] = { "0E0", "-1E0", "1E0", "1.2346E8", "00.000E00", "-10.000E-01", "10.000E-01", "12.346E07", "0E000", "-1E000", "1E000", "123.4568E006", "0E0", "[1E0]", "1E0", "1.235E8" }; static const double valParse[] = { #ifdef OS390 0.01234, 123460000, 1.23E75, -3.1416E-78, 0.01234, 123460000, 1.23E75, -3.1416E-78, 0.01234, 123456800, 1.23E75, -3.141593E-78, 0.01234, 123500000, 1.23E75, -3.142E-78 #else /* We define the whole IEEE 754 number in the 4th column because Visual Age 7 has a bug in rounding numbers. */ 0.01234, 123460000, 1.23E300, -3.1415999999999999E-271, 0.01234, 123460000, 1.23E300, -3.1415999999999999E-271, 0.01234, 123456800, 1.23E300, -3.1415929999999999E-271, 0.01234, 123500000, 1.23E300, -3.1420000000000001E-271 #endif }; static const int32_t lvalParse[] = { 0, -1, 1, 123460000, 0, -1, 1, 123460000, 0, -1, 1, 123456800, 0, -1, 1, 123500000 }; pat_length = sizeof(pat) / sizeof(pat[0]); val_length = sizeof(val) / sizeof(val[0]); lval_length = sizeof(lval) / sizeof(lval[0]); ival = 0; ilval = 0; for (p=0; p < pat_length; ++p) { upat=(UChar*)malloc(sizeof(UChar) * (strlen(pat[p])+1) ); u_uastrcpy(upat, pat[p]); fmt=unum_open(UNUM_IGNORE,upat, u_strlen(upat), "en_US",NULL, &status); if (U_FAILURE(status)) { log_err_status(status, "FAIL: Bad status returned by Number format construction with pattern %s -> %s\n", pat[p], u_errorName(status)); continue; } lneed= u_strlen(upat) + 1; unum_toPattern(fmt, FALSE, pattern, lneed, &status); log_verbose("Pattern \" %s \" -toPattern-> \" %s \" \n", upat, u_austrcpy(tempMsgBug, pattern) ); for (v=0; v<val_length; ++v) { /*format*/ lneed=0; lneed=unum_formatDouble(fmt, val[v], NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_formatDouble(fmt, val[v], str, lneed+1, NULL, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); } u_uastrcpy(uvalfor, valFormat[v+ival]); if(u_strcmp(str, uvalfor) != 0) log_verbose("FAIL: Expected %s ( %s )\n", valFormat[v+ival], u_austrcpy(tempMsgBug, uvalfor) ); /*parsing*/ ppos=0; a=unum_parseDouble(fmt, str, u_strlen(str), &ppos, &status); if (ppos== u_strlen(str)) { if (a != valParse[v+ival]) log_err("FAIL: Expected: %e, Got: %g\n", valParse[v+ival], a); } else log_err(" FAIL: Partial parse ( %d chars ) -> %e\n", ppos, a); free(str); } for (v=0; v<lval_length; ++v) { /*format*/ lneed=0; lneed=unum_formatDouble(fmt, lval[v], NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_formatDouble(fmt, lval[v], str, lneed+1, NULL, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); } /*printf(" Format %e -> %s\n", lval[v], austrdup(str) );*/ u_uastrcpy(ulvalfor, lvalFormat[v+ilval]); if(u_strcmp(str, ulvalfor) != 0) log_err("FAIL: Expected %s ( %s )\n", valFormat[v+ilval], austrdup(ulvalfor) ); /*parsing*/ ppos=0; a=unum_parseDouble(fmt, str, u_strlen(str), &ppos, &status); if (ppos== u_strlen(str)) { /*printf(" Parse -> %e\n", a);*/ if (a != lvalParse[v+ilval]) log_err("FAIL: Expected : %e\n", valParse[v+ival]); } else log_err(" FAIL: Partial parse ( %d chars ) -> %e\n", ppos, a); free(str); } ival += val_length; ilval += lval_length; unum_close(fmt); free(upat); } } /** * Test the handling of the currency symbol in patterns. */ static void TestCurrencySign(void) { int32_t lneed; UNumberFormat *fmt; UChar *pattern=NULL; UChar *str=NULL; UChar *pat=NULL; UChar *res=NULL; UErrorCode status = U_ZERO_ERROR; char tempBuf[256]; pattern=(UChar*)malloc(sizeof(UChar) * (strlen("*#,##0.00;-*#,##0.00") + 1) ); u_uastrcpy(pattern, "*#,##0.00;-*#,##0.00"); pattern[0]=pattern[11]=0xa4; /* insert latin-1 currency symbol */ fmt = unum_open(UNUM_IGNORE,pattern, u_strlen(pattern), "en_US",NULL, &status); if(U_FAILURE(status)){ log_err_status(status, "Error in number format construction with pattern \"\\xA4#,##0.00;-\\xA4#,##0.00\\\" -> %s\n", u_errorName(status)); } lneed=0; lneed=unum_formatDouble(fmt, 1234.56, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_formatDouble(fmt, 1234.56, str, lneed+1, NULL, &status); } if(U_FAILURE(status)) { log_err_status(status, "Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); } lneed=0; lneed=unum_toPattern(fmt, FALSE, NULL, lneed, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; pat=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_formatDouble(fmt, FALSE, pat, lneed+1, NULL, &status); } log_verbose("Pattern \" %s \" \n", u_austrcpy(tempBuf, pat)); log_verbose("Format 1234.56 -> %s\n", u_austrcpy(tempBuf, str) ); if(U_SUCCESS(status) && str) { res=(UChar*)malloc(sizeof(UChar) * (strlen("$1,234.56")+1) ); u_uastrcpy(res, "$1,234.56"); if (u_strcmp(str, res) !=0) log_data_err("FAIL: Expected $1,234.56\n"); } else { log_err_status(status, "Error formatting -> %s\n", u_errorName(status)); } free(str); free(res); free(pat); lneed=0; lneed=unum_formatDouble(fmt, -1234.56, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); unum_formatDouble(fmt, -1234.56, str, lneed+1, NULL, &status); } if(U_FAILURE(status)) { log_err_status(status, "Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); } if(str) { res=(UChar*)malloc(sizeof(UChar) * (strlen("-$1,234.56")+1) ); u_uastrcpy(res, "-$1,234.56"); if (u_strcmp(str, res) != 0) log_data_err("FAIL: Expected -$1,234.56\n"); free(str); free(res); } unum_close(fmt); free(pattern); } /** * Test localized currency patterns. */ static void TestCurrency(void) { UNumberFormat *currencyFmt; UChar *str; int32_t lneed, i; UFieldPosition pos; UChar res[100]; UErrorCode status = U_ZERO_ERROR; const char* locale[]={"fr_CA", "de_DE_PREEURO", "fr_FR_PREEURO"}; const char* result[]={"1,50\\u00a0$", "1,50\\u00a0DM", "1,50\\u00a0F"}; log_verbose("\nTesting the number format with different currency patterns\n"); for(i=0; i < 3; i++) { str=NULL; currencyFmt = unum_open(UNUM_CURRENCY, NULL,0,locale[i],NULL, &status); if(U_FAILURE(status)){ log_data_err("Error in the construction of number format with style currency: %s (Are you missing data?)\n", myErrorName(status)); } else { lneed=0; lneed= unum_formatDouble(currencyFmt, 1.50, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); pos.field = 0; unum_formatDouble(currencyFmt, 1.50, str, lneed+1, &pos, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_formatDouble(.....): %s\n", myErrorName(status) ); } else { u_unescape(result[i], res, (int32_t)strlen(result[i])+1); if (u_strcmp(str, res) != 0){ log_err("FAIL: Expected %s Got: %s for locale: %s\n", result[i], aescstrdup(str, -1), locale[i]); } } } unum_close(currencyFmt); free(str); } } /** * Test localized currency patterns for PREEURO variants. */ static void TestCurrencyPreEuro(void) { UNumberFormat *currencyFmt; UChar *str=NULL, *res=NULL; int32_t lneed, i; UFieldPosition pos; UErrorCode status = U_ZERO_ERROR; const char* locale[]={ "ca_ES_PREEURO", "de_LU_PREEURO", "en_IE_PREEURO", "fi_FI_PREEURO", "fr_LU_PREEURO", "it_IT_PREEURO", "pt_PT_PREEURO", "de_AT_PREEURO", "el_GR_PREEURO", "es_ES_PREEURO", "fr_BE_PREEURO", "ga_IE_PREEURO", "nl_BE_PREEURO", "de_DE_PREEURO", "en_BE_PREEURO", "eu_ES_PREEURO", "fr_FR_PREEURO", "gl_ES_PREEURO", "nl_NL_PREEURO", }; const char* result[]={ "2\\u00A0\\u20A7", "2\\u00A0F", "IR\\u00A31.50", "1,50\\u00A0mk", "2\\u00A0F", "IT\\u20A4\\u00A02", "1$50\\u00A0Esc.", "\\u00F6S\\u00A01,50", "1,50\\u00A0\\u0394\\u03C1\\u03C7", "\\u20A7\\u00A02", "1,50\\u00A0FB", "IR\\u00a31.50", "1,50\\u00A0BF", "1,50\\u00A0DM", "1,50\\u00A0BF", "2\\u00A0\\u20A7", "1,50\\u00A0F", "2\\u00A0\\u20A7", "fl\\u00A01,50" }; log_verbose("\nTesting the number format with different currency patterns\n"); for(i=0; i < 19; i++) { char curID[256] = {0}; uloc_canonicalize(locale[i], curID, 256, &status); if(U_FAILURE(status)){ log_data_err("Could not canonicalize %s. Error: %s (Are you missing data?)\n", locale[i], u_errorName(status)); continue; } currencyFmt = unum_open(UNUM_CURRENCY, NULL,0,curID,NULL, &status); if(U_FAILURE(status)){ log_data_err("Error in the construction of number format with style currency: %s (Are you missing data?)\n", myErrorName(status)); } else { lneed=0; lneed= unum_formatDouble(currencyFmt, 1.50, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); pos.field = 0; unum_formatDouble(currencyFmt, 1.50, str, lneed+1, &pos, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_formatDouble(.....): %s\n", myErrorName(status) ); } else { res=(UChar*)malloc(sizeof(UChar) * (strlen(result[i])+1) ); u_unescape(result[i],res,(int32_t)(strlen(result[i])+1)); if (u_strcmp(str, res) != 0){ log_err("FAIL: Expected %s Got: %s for locale: %s\n", result[i],aescstrdup(str, -1),locale[i]); } } } unum_close(currencyFmt); free(str); free(res); } } /** * Test currency "object" (we use this name to match the other C++ * test name and the Jave name). Actually, test ISO currency code * support in the C API. */ static void TestCurrencyObject(void) { UNumberFormat *currencyFmt; UChar *str=NULL, *res=NULL; int32_t lneed, i; UFieldPosition pos; UErrorCode status = U_ZERO_ERROR; const char* locale[]={ "fr_FR", "fr_FR", }; const char* currency[]={ "", "JPY", }; const char* result[]={ "1\\u00A0234,56\\u00A0\\u20AC", "1\\u00A0235\\u00A0\\u00A5JP", }; log_verbose("\nTesting the number format with different currency codes\n"); for(i=0; i < 2; i++) { char cStr[20]={0}; UChar isoCode[16]={0}; currencyFmt = unum_open(UNUM_CURRENCY, NULL,0,locale[i],NULL, &status); if(U_FAILURE(status)){ log_data_err("Error in the construction of number format with style currency: %s (Are you missing data?)\n", myErrorName(status)); } else { if (*currency[i]) { u_uastrcpy(isoCode, currency[i]); unum_setTextAttribute(currencyFmt, UNUM_CURRENCY_CODE, isoCode, u_strlen(isoCode), &status); if(U_FAILURE(status)) { log_err("FAIL: can't set currency code %s\n", myErrorName(status) ); } } unum_getTextAttribute(currencyFmt, UNUM_CURRENCY_CODE, isoCode, sizeof(isoCode), &status); if(U_FAILURE(status)) { log_err("FAIL: can't get currency code %s\n", myErrorName(status) ); } u_UCharsToChars(isoCode,cStr,u_strlen(isoCode)); log_verbose("ISO code %s\n", cStr); if (*currency[i] && uprv_strcmp(cStr, currency[i])) { log_err("FAIL: currency should be %s, but is %s\n", currency[i], cStr); } lneed=0; lneed= unum_formatDouble(currencyFmt, 1234.56, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; str=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); pos.field = 0; unum_formatDouble(currencyFmt, 1234.56, str, lneed+1, &pos, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_formatDouble(.....): %s\n", myErrorName(status) ); } else { res=(UChar*)malloc(sizeof(UChar) * (strlen(result[i])+1) ); u_unescape(result[i],res, (int32_t)(strlen(result[i])+1)); if (u_strcmp(str, res) != 0){ log_err("FAIL: Expected %s Got: %s for locale: %s\n", result[i],aescstrdup(str, -1),locale[i]); } } } unum_close(currencyFmt); free(str); free(res); } } /** * Test proper rounding by the format method. */ static void TestRounding487(void) { UNumberFormat *nnf; UErrorCode status = U_ZERO_ERROR; /* this is supposed to open default date format, but later on it treats it like it is "en_US" - very bad if you try to run the tests on machine where default locale is NOT "en_US" */ /* nnf = unum_open(UNUM_DEFAULT, NULL, &status); */ nnf = unum_open(UNUM_DEFAULT, NULL,0,"en_US",NULL, &status); if(U_FAILURE(status)){ log_data_err("FAIL: failure in the construction of number format: %s (Are you missing data?)\n", myErrorName(status)); } else { roundingTest(nnf, 0.00159999, 4, "0.0016"); roundingTest(nnf, 0.00995, 4, "0.01"); roundingTest(nnf, 12.3995, 3, "12.4"); roundingTest(nnf, 12.4999, 0, "12"); roundingTest(nnf, - 19.5, 0, "-20"); } unum_close(nnf); } /*-------------------------------------*/ static void roundingTest(UNumberFormat* nf, double x, int32_t maxFractionDigits, const char* expected) { UChar *out = NULL; UChar *res; UFieldPosition pos; UErrorCode status; int32_t lneed; status=U_ZERO_ERROR; unum_setAttribute(nf, UNUM_MAX_FRACTION_DIGITS, maxFractionDigits); lneed=0; lneed=unum_formatDouble(nf, x, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; out=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); pos.field=0; unum_formatDouble(nf, x, out, lneed+1, &pos, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_formatDouble(.....): %s\n", myErrorName(status) ); } /*Need to use log_verbose here. Problem with the float*/ /*printf("%f format with %d fraction digits to %s\n", x, maxFractionDigits, austrdup(out) );*/ res=(UChar*)malloc(sizeof(UChar) * (strlen(expected)+1) ); u_uastrcpy(res, expected); if (u_strcmp(out, res) != 0) log_err("FAIL: Expected: %s or %s\n", expected, austrdup(res) ); free(res); if(out != NULL) { free(out); } } /* * Testing unum_getDoubleAttribute and unum_setDoubleAttribute() */ static void TestDoubleAttribute(void) { double mydata[] = { 1.11, 22.22, 333.33, 4444.44, 55555.55, 666666.66, 7777777.77, 88888888.88, 999999999.99}; double dvalue; int i; UErrorCode status=U_ZERO_ERROR; UNumberFormatAttribute attr; UNumberFormatStyle style= UNUM_DEFAULT; UNumberFormat *def; log_verbose("\nTesting get and set DoubleAttributes\n"); def=unum_open(style, NULL,0,NULL,NULL, &status); if (U_FAILURE(status)) { log_data_err("Fail: error creating a default number formatter -> %s (Are you missing data?)\n", u_errorName(status)); } else { attr=UNUM_ROUNDING_INCREMENT; dvalue=unum_getDoubleAttribute(def, attr); for (i = 0; i<9 ; i++) { dvalue = mydata[i]; unum_setDoubleAttribute(def, attr, dvalue); if(unum_getDoubleAttribute(def,attr)!=mydata[i]) log_err("Fail: error in setting and getting double attributes for UNUM_ROUNDING_INCREMENT\n"); else log_verbose("Pass: setting and getting double attributes for UNUM_ROUNDING_INCREMENT works fine\n"); } } unum_close(def); } /** * Test the functioning of the secondary grouping value. */ static void TestSecondaryGrouping(void) { UErrorCode status = U_ZERO_ERROR; UNumberFormat *f = NULL, *g= NULL; UNumberFormat *us = unum_open(UNUM_DECIMAL,NULL,0, "en_US", NULL,&status); UFieldPosition pos; UChar resultBuffer[512]; int32_t l = 1876543210L; UBool ok = TRUE; UChar buffer[512]; int32_t i; UBool expectGroup = FALSE, isGroup = FALSE; u_uastrcpy(buffer, "#,##,###"); f = unum_open(UNUM_IGNORE,buffer, -1, "en_US",NULL, &status); if (U_FAILURE(status)) { log_data_err("Error DecimalFormat ct -> %s (Are you missing data?)\n", u_errorName(status)); return; } pos.field = 0; unum_format(f, (int32_t)123456789L, resultBuffer, 512 , &pos, &status); u_uastrcpy(buffer, "12,34,56,789"); if ((u_strcmp(resultBuffer, buffer) != 0) || U_FAILURE(status)) { log_err("Fail: Formatting \"#,##,###\" pattern with 123456789 got %s, expected %s\n", resultBuffer, "12,34,56,789"); } if (pos.beginIndex != 0 && pos.endIndex != 12) { log_err("Fail: Formatting \"#,##,###\" pattern pos = (%d, %d) expected pos = (0, 12)\n", pos.beginIndex, pos.endIndex); } memset(resultBuffer,0, sizeof(UChar)*512); unum_toPattern(f, FALSE, resultBuffer, 512, &status); u_uastrcpy(buffer, "#,##,###"); if ((u_strcmp(resultBuffer, buffer) != 0) || U_FAILURE(status)) { log_err("Fail: toPattern() got %s, expected %s\n", resultBuffer, "#,##,###"); } memset(resultBuffer,0, sizeof(UChar)*512); u_uastrcpy(buffer, "#,###"); unum_applyPattern(f, FALSE, buffer, -1,NULL,NULL); if (U_FAILURE(status)) { log_err("Fail: applyPattern call failed\n"); } unum_setAttribute(f, UNUM_SECONDARY_GROUPING_SIZE, 4); unum_format(f, (int32_t)123456789L, resultBuffer, 512 , &pos, &status); u_uastrcpy(buffer, "12,3456,789"); if ((u_strcmp(resultBuffer, buffer) != 0) || U_FAILURE(status)) { log_err("Fail: Formatting \"#,###\" pattern with 123456789 got %s, expected %s\n", resultBuffer, "12,3456,789"); } memset(resultBuffer,0, sizeof(UChar)*512); unum_toPattern(f, FALSE, resultBuffer, 512, &status); u_uastrcpy(buffer, "#,####,###"); if ((u_strcmp(resultBuffer, buffer) != 0) || U_FAILURE(status)) { log_err("Fail: toPattern() got %s, expected %s\n", resultBuffer, "#,####,###"); } memset(resultBuffer,0, sizeof(UChar)*512); g = unum_open(UNUM_DECIMAL, NULL,0,"hi_IN",NULL, &status); if (U_FAILURE(status)) { log_err("Fail: Cannot create UNumberFormat for \"hi_IN\" locale.\n"); } unum_format(g, l, resultBuffer, 512, &pos, &status); unum_close(g); /* expect "1,87,65,43,210", but with Hindi digits */ /* 01234567890123 */ if (u_strlen(resultBuffer) != 14) { ok = FALSE; } else { for (i=0; i<u_strlen(resultBuffer); ++i) { expectGroup = FALSE; switch (i) { case 1: case 4: case 7: case 10: expectGroup = TRUE; break; } /* Later -- fix this to get the actual grouping */ /* character from the resource bundle. */ isGroup = (UBool)(resultBuffer[i] == 0x002C); if (isGroup != expectGroup) { ok = FALSE; break; } } } if (!ok) { log_err("FAIL Expected %s x hi_IN -> \"1,87,65,43,210\" (with Hindi digits), got %s\n", "1876543210L", resultBuffer); } unum_close(f); unum_close(us); } static void TestCurrencyKeywords(void) { static const char * const currencies[] = { "ADD", "ADP", "AED", "AFA", "AFN", "AIF", "ALK", "ALL", "ALV", "ALX", "AMD", "ANG", "AOA", "AOK", "AON", "AOR", "AOS", "ARA", "ARM", "ARP", "ARS", "ATS", "AUD", "AUP", "AWG", "AZM", "BAD", "BAM", "BAN", "BBD", "BDT", "BEC", "BEF", "BEL", "BGL", "BGM", "BGN", "BGO", "BGX", "BHD", "BIF", "BMD", "BMP", "BND", "BOB", "BOL", "BOP", "BOV", "BRB", "BRC", "BRE", "BRL", "BRN", "BRR", "BRZ", "BSD", "BSP", "BTN", "BTR", "BUK", "BUR", "BWP", "BYB", "BYL", "BYR", "BZD", "BZH", "CAD", "CDF", "CDG", "CDL", "CFF", "CHF", "CKD", "CLC", "CLE", "CLF", "CLP", "CMF", "CNP", "CNX", "CNY", "COB", "COF", "COP", "CRC", "CSC", "CSK", "CUP", "CUX", "CVE", "CWG", "CYP", "CZK", "DDM", "DEM", "DES", "DJF", "DKK", "DOP", "DZD", "DZF", "DZG", "ECS", "ECV", "EEK", "EGP", "ERN", "ESP", "ETB", "ETD", "EUR", "FIM", "FIN", "FJD", "FJP", "FKP", "FOK", "FRF", "FRG", "GAF", "GBP", "GEK", "GEL", "GHC", "GHO", "GHP", "GHR", "GIP", "GLK", "GMD", "GMP", "GNF", "GNI", "GNS", "GPF", "GQE", "GQF", "GQP", "GRD", "GRN", "GTQ", "GUF", "GWE", "GWM", "GWP", "GYD", "HKD", "HNL", "HRD", "HRK", "HTG", "HUF", "IBP", "IDG", "IDJ", "IDN", "IDR", "IEP", "ILL", "ILP", "ILS", "IMP", "INR", "IQD", "IRR", "ISK", "ITL", "JEP", "JMD", "JMP", "JOD", "JPY", "KES", "KGS", "KHO", "KHR", "KID", "KMF", "KPP", "KPW", "KRH", "KRO", "KRW", "KWD", "KYD", "KZR", "KZT", "LAK", "LBP", "LIF", "LKR", "LNR", "LRD", "LSL", "LTL", "LTT", "LUF", "LVL", "LVR", "LYB", "LYD", "LYP", "MAD", "MAF", "MCF", "MCG", "MDC", "MDL", "MDR", "MGA", "MGF", "MHD", "MKD", "MKN", "MLF", "MMK", "MMX", "MNT", "MOP", "MQF", "MRO", "MTL", "MTP", "MUR", "MVP", "MVR", "MWK", "MWP", "MXN", "MXP", "MXV", "MYR", "MZE", "MZM", "NAD", "NCF", "NGN", "NGP", "NHF", "NIC", "NIG", "NIO", "NLG", "NOK", "NPR", "NZD", "NZP", "OMR", "OMS", "PAB", "PDK", "PDN", "PDR", "PEI", "PEN", "PES", "PGK", "PHP", "PKR", "PLN", "PLX", "PLZ", "PSP", "PTC", "PTE", "PYG", "QAR", "REF", "ROL", "RON", "RUB", "RUR", "RWF", "SAR", "SAS", "SBD", "SCR", "SDD", "SDP", "SEK", "SGD", "SHP", "SIB", "SIT", "SKK", "SLL", "SML", "SOS", "SQS", "SRG", "SSP", "STD", "STE", "SUN", "SUR", "SVC", "SYP", "SZL", "TCC", "TDF", "THB", "TJR", "TJS", "TMM", "TND", "TOP", "TOS", "TPE", "TPP", "TRL", "TTD", "TTO", "TVD", "TWD", "TZS", "UAH", "UAK", "UGS", "UGX", "USD", "USN", "USS", "UYF", "UYP", "UYU", "UZC", "UZS", "VAL", "VDD", "VDN", "VDP", "VEB", "VGD", "VND", "VNN", "VNR", "VNS", "VUV", "WSP", "WST", "XAD", "XAF", "XAM", "XAU", "XBA", "XBB", "XBC", "XBD", "XCD", "XCF", "XDR", "XEF", "XEU", "XFO", "XFU", "XID", "XMF", "XNF", "XOF", "XPF", "XPS", "XSS", "XTR", "YDD", "YEI", "YER", "YUD", "YUF", "YUG", "YUM", "YUN", "YUO", "YUR", "ZAL", "ZAP", "ZAR", "ZMK", "ZMP", "ZRN", "ZRZ", "ZWD" }; UErrorCode status = U_ZERO_ERROR; int32_t i = 0, j = 0; int32_t noLocales = uloc_countAvailable(); char locale[256]; char currLoc[256]; UChar result[4]; UChar currBuffer[256]; for(i = 0; i < noLocales; i++) { strcpy(currLoc, uloc_getAvailable(i)); for(j = 0; j < sizeof(currencies)/sizeof(currencies[0]); j++) { strcpy(locale, currLoc); strcat(locale, "@currency="); strcat(locale, currencies[j]); ucurr_forLocale(locale, result, 4, &status); u_charsToUChars(currencies[j], currBuffer, 3); currBuffer[3] = 0; if(u_strcmp(currBuffer, result) != 0) { log_err("Didn't get the right currency for %s\n", locale); } } } } static void TestGetKeywordValuesForLocale(void) { #define PREFERRED_SIZE 13 #define MAX_NUMBER_OF_KEYWORDS 3 const char *PREFERRED[PREFERRED_SIZE][MAX_NUMBER_OF_KEYWORDS] = { { "root", "USD", NULL }, { "und", "USD", NULL }, { "und_ZZ", "USD", NULL }, { "en_US", "USD", NULL }, { "en_029", "USD", NULL }, { "en_TH", "THB", NULL }, { "de", "EUR", NULL }, { "de_DE", "EUR", NULL }, { "ar", "EGP", NULL }, { "ar_PS", "JOD", "ILS" }, { "en@currency=CAD", "USD", NULL }, { "fr@currency=zzz", "EUR", NULL }, { "de_DE@currency=DEM", "EUR", NULL }, }; const int32_t EXPECTED_SIZE[PREFERRED_SIZE] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1 }; UErrorCode status = U_ZERO_ERROR; int32_t i, j, size; UEnumeration *pref, *all; const char *loc = NULL; UBool matchPref, matchAll; const char *value = NULL; int32_t valueLength = 0; UList *ALLList = NULL; UEnumeration *ALL = ucurr_getKeywordValuesForLocale("currency", uloc_getDefault(), FALSE, &status); if (ALL == NULL) { log_err_status(status, "ERROR getting keyword value for default locale. -> %s\n", u_errorName(status)); return; } for (i = 0; i < PREFERRED_SIZE; i++) { pref = NULL; all = NULL; loc = PREFERRED[i][0]; pref = ucurr_getKeywordValuesForLocale("currency", loc, TRUE, &status); matchPref = FALSE; matchAll = FALSE; size = uenum_count(pref, &status); if (size == EXPECTED_SIZE[i]) { matchPref = TRUE; for (j = 0; j < size; j++) { if ((value = uenum_next(pref, &valueLength, &status)) != NULL && U_SUCCESS(status)) { if (uprv_strcmp(value, PREFERRED[i][j+1]) != 0) { log_err("ERROR: locale %s got keywords #%d %s expected %s\n", loc, j, value, PREFERRED[i][j+1]); matchPref = FALSE; break; } } else { matchPref = FALSE; log_err("ERROR getting keyword value for locale \"%s\"\n", loc); break; } } } else { log_err("FAIL: size of locale \"%s\" %d does not match expected size %d\n", loc, size, EXPECTED_SIZE[i]); } if (!matchPref) { log_err("FAIL: Preferred values for locale \"%s\" does not match expected.\n", loc); break; } uenum_close(pref); all = ucurr_getKeywordValuesForLocale("currency", loc, FALSE, &status); size = uenum_count(all, &status); if (U_SUCCESS(status) && size == uenum_count(ALL, &status)) { matchAll = TRUE; ALLList = ulist_getListFromEnum(ALL); for (j = 0; j < size; j++) { if ((value = uenum_next(all, &valueLength, &status)) != NULL && U_SUCCESS(status)) { if (!ulist_containsString(ALLList, value, uprv_strlen(value))) { log_err("Locale %s have %s not in ALL\n", loc, value); matchAll = FALSE; break; } } else { matchAll = FALSE; log_err("ERROR getting \"all\" keyword value for locale \"%s\"\n", loc); break; } } if (!matchAll) { log_err("FAIL: All values for locale \"%s\" does not match expected.\n", loc); } } else { if(U_FAILURE(status)) { log_err("ERROR: %s\n", u_errorName(status)); } else if(size!=uenum_count(ALL, &status)) { log_err("ERROR: got size of %d, wanted %d\n", size, uenum_count(ALL, &status)); } } uenum_close(all); } uenum_close(ALL); } /** * Test proper handling of rounding modes. */ static void TestRounding5350(void) { UNumberFormat *nnf; UErrorCode status = U_ZERO_ERROR; /* this is supposed to open default date format, but later on it treats it like it is "en_US" - very bad if you try to run the tests on machine where default locale is NOT "en_US" */ /* nnf = unum_open(UNUM_DEFAULT, NULL, &status); */ nnf = unum_open(UNUM_DEFAULT, NULL,0,"en_US",NULL, &status); if(U_FAILURE(status)){ log_data_err("FAIL: failure in the construction of number format: %s (Are you missing data?)\n", myErrorName(status)); return; } unum_setAttribute(nnf, UNUM_MAX_FRACTION_DIGITS, 2); roundingTest2(nnf, -0.125, UNUM_ROUND_CEILING, "-0.12"); roundingTest2(nnf, -0.125, UNUM_ROUND_FLOOR, "-0.13"); roundingTest2(nnf, -0.125, UNUM_ROUND_DOWN, "-0.12"); roundingTest2(nnf, -0.125, UNUM_ROUND_UP, "-0.13"); roundingTest2(nnf, 0.125, UNUM_FOUND_HALFEVEN, "0.12"); roundingTest2(nnf, 0.135, UNUM_ROUND_HALFDOWN, "0.13"); roundingTest2(nnf, 0.125, UNUM_ROUND_HALFUP, "0.13"); roundingTest2(nnf, 0.135, UNUM_FOUND_HALFEVEN, "0.14"); /* The following are exactly represented, and shouldn't round */ roundingTest2(nnf, 1.00, UNUM_ROUND_UP, "1"); roundingTest2(nnf, 24.25, UNUM_ROUND_UP, "24.25"); roundingTest2(nnf, 24.25, UNUM_ROUND_CEILING, "24.25"); roundingTest2(nnf, -24.25, UNUM_ROUND_UP, "-24.25"); /* Differences pretty far out there */ roundingTest2(nnf, 1.0000001, UNUM_ROUND_CEILING, "1.01"); roundingTest2(nnf, 1.0000001, UNUM_ROUND_FLOOR, "1"); roundingTest2(nnf, 1.0000001, UNUM_ROUND_DOWN, "1"); roundingTest2(nnf, 1.0000001, UNUM_ROUND_UP, "1.01"); roundingTest2(nnf, 1.0000001, UNUM_FOUND_HALFEVEN, "1"); roundingTest2(nnf, 1.0000001, UNUM_ROUND_HALFDOWN, "1"); roundingTest2(nnf, 1.0000001, UNUM_ROUND_HALFUP, "1"); roundingTest2(nnf, -1.0000001, UNUM_ROUND_CEILING, "-1"); roundingTest2(nnf, -1.0000001, UNUM_ROUND_FLOOR, "-1.01"); roundingTest2(nnf, -1.0000001, UNUM_ROUND_DOWN, "-1"); roundingTest2(nnf, -1.0000001, UNUM_ROUND_UP, "-1.01"); roundingTest2(nnf, -1.0000001, UNUM_FOUND_HALFEVEN, "-1"); roundingTest2(nnf, -1.0000001, UNUM_ROUND_HALFDOWN, "-1"); roundingTest2(nnf, -1.0000001, UNUM_ROUND_HALFUP, "-1"); unum_close(nnf); } /*-------------------------------------*/ static void roundingTest2(UNumberFormat* nf, double x, int32_t roundingMode, const char* expected) { UChar *out = NULL; UChar *res; UFieldPosition pos; UErrorCode status; int32_t lneed; status=U_ZERO_ERROR; unum_setAttribute(nf, UNUM_ROUNDING_MODE, roundingMode); lneed=0; lneed=unum_formatDouble(nf, x, NULL, lneed, NULL, &status); if(status==U_BUFFER_OVERFLOW_ERROR){ status=U_ZERO_ERROR; out=(UChar*)malloc(sizeof(UChar) * (lneed+1) ); pos.field=0; unum_formatDouble(nf, x, out, lneed+1, &pos, &status); } if(U_FAILURE(status)) { log_err("Error in formatting using unum_formatDouble(.....): %s\n", myErrorName(status) ); } /*Need to use log_verbose here. Problem with the float*/ /*printf("%f format with %d fraction digits to %s\n", x, maxFractionDigits, austrdup(out) );*/ res=(UChar*)malloc(sizeof(UChar) * (strlen(expected)+1) ); u_uastrcpy(res, expected); if (u_strcmp(out, res) != 0) log_err("FAIL: Expected: \"%s\" Got: \"%s\"\n", expected, austrdup(out) ); free(res); if(out != NULL) { free(out); } } #endif /* #if !UCONFIG_NO_FORMATTING */