| /******************************************************************** |
| * COPYRIGHT: |
| * Copyright (c) 1997-2001, International Business Machines Corporation and |
| * others. All Rights Reserved. |
| ********************************************************************/ |
| /******************************************************************************** |
| * |
| * File CNUMTST.C |
| * |
| * Madhu Katragadda Creation |
| * |
| * Modification History: |
| * |
| * Date Name Description |
| * 06/24/99 helena Integrated Alan's NF enhancements and Java2 bug fixes |
| * 07/15/99 helena Ported to HPUX 10/11 CC. |
| ********************************************************************************* |
| */ |
| |
| /* C API TEST FOR NUMBER FORMAT */ |
| |
| #include "unicode/utypes.h" |
| |
| #if !UCONFIG_NO_FORMATTING |
| |
| #include "unicode/uloc.h" |
| #include "unicode/unum.h" |
| #include "unicode/ustring.h" |
| #include "cintltst.h" |
| #include "cnumtst.h" |
| #include "cmemory.h" |
| |
| #define LENGTH(arr) (sizeof(arr)/sizeof(arr[0])) |
| |
| void addNumForTest(TestNode** root); |
| |
| void addNumForTest(TestNode** root) |
| { |
| addTest(root, &TestNumberFormat, "tsformat/cnumtst/TestNumberFormat"); |
| addTest(root, &TestNumberFormatPadding, "tsformat/cnumtst/TestNumberFormatPadding"); |
| } |
| |
| /** copy src to dst with unicode-escapes for values < 0x20 and > 0x7e, null terminate if possible */ |
| static int32_t ustrToAstr(const UChar* src, int32_t srcLength, char* dst, int32_t dstLength) { |
| static const char hex[] = { '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'a', 'b', 'c', 'd', 'e', 'f' }; |
| |
| char *p = dst; |
| const char *e = p + dstLength; |
| if (srcLength < 0) { |
| const UChar* s = src; |
| while (*s) { |
| ++s; |
| } |
| srcLength = (int32_t)(s - src); |
| } |
| while (p < e && --srcLength >= 0) { |
| UChar c = *src++; |
| if (c == 0xd || c == 0xa || c == 0x9 || (c>= 0x20 && c <= 0x7e)) { |
| *p++ = (char) c & 0x7f; |
| } else if (e - p >= 6) { |
| *p++ = '\\'; |
| *p++ = 'u'; |
| *p++ = hex[(c >> 12) & 0xf]; |
| *p++ = hex[(c >> 8) & 0xf]; |
| *p++ = hex[(c >> 4) & 0xf]; |
| *p++ = hex[c & 0xf]; |
| } else { |
| break; |
| } |
| } |
| if (p < e) { |
| *p = 0; |
| } |
| return (int32_t)(p - dst); |
| } |
| |
| /* test Number Format API */ |
| static void TestNumberFormat() |
| { |
| UChar *result=NULL; |
| UChar temp1[512]; |
| UChar temp2[512]; |
| |
| UChar temp[5]; |
| |
| UChar prefix[5]; |
| UChar suffix[5]; |
| UChar symbol[20]; |
| int32_t resultlength; |
| int32_t resultlengthneeded; |
| int32_t parsepos; |
| double d1; |
| int32_t l1; |
| double d = -10456.37; |
| int32_t l = 100000000; |
| UFieldPosition pos1; |
| UFieldPosition pos2; |
| int32_t numlocales; |
| int32_t i; |
| |
| UNumberFormatAttribute attr; |
| UNumberFormatSymbol symType = UNUM_DECIMAL_SEPARATOR_SYMBOL; |
| int32_t newvalue; |
| UErrorCode status=U_ZERO_ERROR; |
| UNumberFormatStyle style= UNUM_DEFAULT; |
| UNumberFormat *pattern; |
| UNumberFormat *def, *fr, *cur_def, *cur_fr, *per_def, *per_fr, |
| *cur_frpattern, *myclone, *spellout_def; |
| |
| /* Testing unum_open() with various Numberformat styles and locales*/ |
| status = U_ZERO_ERROR; |
| log_verbose("Testing unum_open() with default style and locale\n"); |
| def=unum_open(style, NULL,0,NULL, NULL,&status); |
| if(U_FAILURE(status)) |
| log_err("Error in creating NumberFormat default using unum_open(): %s\n", myErrorName(status)); |
| |
| log_verbose("\nTesting unum_open() with french locale and default style(decimal)\n"); |
| fr=unum_open(style,NULL,0, "fr_FR",NULL, &status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not create NumberFormat (french): %s\n", myErrorName(status)); |
| |
| log_verbose("\nTesting unum_open(currency,NULL,status)\n"); |
| style=UNUM_CURRENCY; |
| /* Can't hardcode the result to assume the default locale is "en_US". */ |
| cur_def=unum_open(style, NULL,0,"en_US", NULL, &status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not create NumberFormat using \n unum_open(currency, NULL, &status) %s\n", |
| myErrorName(status) ); |
| |
| log_verbose("\nTesting unum_open(currency, frenchlocale, status)\n"); |
| cur_fr=unum_open(style,NULL,0, "fr_FR", NULL, &status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not create NumberFormat using unum_open(currency, french, &status): %s\n", |
| myErrorName(status)); |
| |
| log_verbose("\nTesting unum_open(percent, NULL, status)\n"); |
| style=UNUM_PERCENT; |
| per_def=unum_open(style,NULL,0, NULL,NULL, &status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not create NumberFormat using unum_open(percent, NULL, &status): %s\n", myErrorName(status)); |
| |
| log_verbose("\nTesting unum_open(percent,frenchlocale, status)\n"); |
| per_fr=unum_open(style, NULL,0,"fr_FR", NULL,&status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not create NumberFormat using unum_open(percent, french, &status): %s\n", myErrorName(status)); |
| |
| log_verbose("\nTesting unum_open(spellout, NULL, status)"); |
| style=UNUM_SPELLOUT; |
| spellout_def=unum_open(style, NULL, 0, "en_US", NULL, &status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not create NumberFormat using unum_open(spellout, NULL, &status): %s\n", myErrorName(status)); |
| |
| /* Testing unum_clone(..) */ |
| log_verbose("\nTesting unum_clone(fmt, status)"); |
| status = U_ZERO_ERROR; |
| myclone = unum_clone(def,&status); |
| if(U_FAILURE(status)) |
| log_err("Error: could not clone unum_clone(def, &status): %s\n", myErrorName(status)); |
| else |
| { |
| log_verbose("unum_clone() successful\n"); |
| } |
| |
| /*Testing unum_getAvailable() and unum_countAvailable()*/ |
| log_verbose("\nTesting getAvailableLocales and countAvailable()\n"); |
| numlocales=unum_countAvailable(); |
| if(numlocales < 0) |
| log_err("error in countAvailable"); |
| else{ |
| log_verbose("unum_countAvialable() successful\n"); |
| log_verbose("The no: of locales where number formattting is applicable is %d\n", numlocales); |
| } |
| for(i=0;i<numlocales;i++) |
| { |
| log_verbose("%s\n", unum_getAvailable(i)); |
| if (unum_getAvailable(i) == 0) |
| log_err("No locale for which number formatting patterns are applicable\n"); |
| else |
| log_verbose("A locale %s for which number formatting patterns are applicable\n",unum_getAvailable(i)); |
| } |
| |
| |
| /*Testing unum_format() and unum_formatdouble()*/ |
| u_uastrcpy(temp1, "$100,000,000.00"); |
| |
| log_verbose("\nTesting unum_format() \n"); |
| resultlength=0; |
| pos1.field = 0; /* Integer Section */ |
| resultlengthneeded=unum_format(cur_def, l, NULL, resultlength, &pos1, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| /* for (i = 0; i < 100000; i++) */ |
| { |
| unum_format(cur_def, l, result, resultlength, &pos1, &status); |
| } |
| } |
| |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status) ); |
| } |
| if(u_strcmp(result, temp1)==0) |
| log_verbose("Pass: Number formatting using unum_format() successful\n"); |
| else |
| log_err("Fail: Error in number Formatting using unum_format()\n"); |
| if(pos1.beginIndex == 1 && pos1.endIndex == 12) |
| log_verbose("Pass: Complete number formatting using unum_format() successful\n"); |
| else |
| log_err("Fail: Error in complete number Formatting using unum_format()\nGot: b=%d end=%d\nExpected: b=1 end=12\n", |
| pos1.beginIndex, pos1.endIndex); |
| |
| free(result); |
| result = 0; |
| |
| log_verbose("\nTesting unum_formatDouble()\n"); |
| u_uastrcpy(temp1, "($10,456.37)"); |
| resultlength=0; |
| pos2.field = 1; /* Fractional Section */ |
| resultlengthneeded=unum_formatDouble(cur_def, d, NULL, resultlength, &pos2, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| /* for (i = 0; i < 100000; i++) */ |
| { |
| unum_formatDouble(cur_def, d, result, resultlength, &pos2, &status); |
| } |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in formatting using unum_formatDouble(.....): %s\n", myErrorName(status)); |
| } |
| if(u_strcmp(result, temp1)==0) |
| log_verbose("Pass: Number Formatting using unum_formatDouble() Successful\n"); |
| else |
| log_err("FAIL: Error in number formatting using unum_formatDouble()\n"); |
| if(pos2.beginIndex == 9 && pos2.endIndex == 11) |
| log_verbose("Pass: Complete number formatting using unum_format() successful\n"); |
| else |
| log_err("Fail: Error in complete number Formatting using unum_formatDouble()\nGot: b=%d end=%d\nExpected: b=9 end=11", |
| pos1.beginIndex, pos1.endIndex); |
| |
| |
| /* Testing unum_parse() and unum_parseDouble() */ |
| log_verbose("\nTesting unum_parseDouble()\n"); |
| /* for (i = 0; i < 100000; i++)*/ |
| { |
| parsepos=0; |
| d1=unum_parseDouble(cur_def, result, u_strlen(result), &parsepos, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("parse failed. The error is : %s\n", myErrorName(status)); |
| } |
| |
| if(d1!=d) |
| log_err("Fail: Error in parsing\n"); |
| else |
| log_verbose("Pass: parsing successful\n"); |
| |
| /* performance testing */ |
| u_uastrcpy(temp1, "$462.12345"); |
| resultlength=u_strlen(temp1); |
| /* for (i = 0; i < 100000; i++) */ |
| { |
| parsepos=0; |
| d1=unum_parseDouble(cur_def, temp1, resultlength, &parsepos, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("parse failed. The error is : %s\n", myErrorName(status)); |
| } |
| |
| if(d1!=462.12345) |
| log_err("Fail: Error in parsing\n"); |
| else |
| log_verbose("Pass: parsing successful\n"); |
| |
| free(result); |
| |
| u_uastrcpy(temp1, "($10,456.3E1])"); |
| parsepos=0; |
| d1=unum_parseDouble(cur_def, temp1, u_strlen(temp1), &parsepos, &status); |
| if(U_SUCCESS(status)) |
| { |
| log_err("Error in unum_parseDouble(..., %s, ...): %s\n", temp1, myErrorName(status)); |
| } |
| |
| |
| log_verbose("\nTesting unum_format()\n"); |
| status=U_ZERO_ERROR; |
| resultlength=0; |
| parsepos=0; |
| resultlengthneeded=unum_format(per_fr, l, NULL, resultlength, &pos1, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| /* for (i = 0; i < 100000; i++)*/ |
| { |
| unum_format(per_fr, l, result, resultlength, &pos1, &status); |
| } |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status)); |
| } |
| |
| |
| log_verbose("\nTesting unum_parse()\n"); |
| /* for (i = 0; i < 100000; i++) */ |
| { |
| parsepos=0; |
| l1=unum_parse(per_fr, result, u_strlen(result), &parsepos, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("parse failed. The error is : %s\n", myErrorName(status)); |
| } |
| |
| if(l1!=l) |
| log_err("Fail: Error in parsing\n"); |
| else |
| log_verbose("Pass: parsing successful\n"); |
| |
| free(result); |
| |
| /* create a number format using unum_openPattern(....)*/ |
| log_verbose("\nTesting unum_openPattern()\n"); |
| u_uastrcpy(temp1, "#,##0.0#;(#,##0.0#)"); |
| pattern=unum_open(UNUM_IGNORE,temp1, u_strlen(temp1), NULL, NULL,&status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in unum_openPattern(): %s\n", myErrorName(status) );; |
| } |
| else |
| log_verbose("Pass: unum_openPattern() works fine\n"); |
| |
| /*test for unum_toPattern()*/ |
| log_verbose("\nTesting unum_toPattern()\n"); |
| resultlength=0; |
| resultlengthneeded=unum_toPattern(pattern, FALSE, NULL, resultlength, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| unum_toPattern(pattern, FALSE, result, resultlength, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("error in extracting the pattern from UNumberFormat: %s\n", myErrorName(status)); |
| } |
| else |
| { |
| if(u_strcmp(result, temp1)!=0) |
| log_err("FAIL: Error in extracting the pattern using unum_toPattern()\n"); |
| else |
| log_verbose("Pass: extracted the pattern correctly using unum_toPattern()\n"); |
| free(result); |
| } |
| |
| /*Testing unum_getSymbols() and unum_setSymbols()*/ |
| log_verbose("\nTesting unum_getSymbols and unum_setSymbols()\n"); |
| /*when we try to change the symbols of french to default we need to apply the pattern as well to fetch correct results */ |
| resultlength=0; |
| resultlengthneeded=unum_toPattern(cur_def, FALSE, NULL, resultlength, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| unum_toPattern(cur_def, FALSE, result, resultlength, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("error in extracting the pattern from UNumberFormat: %s\n", myErrorName(status)); |
| } |
| |
| status=U_ZERO_ERROR; |
| cur_frpattern=unum_open(UNUM_IGNORE,result, u_strlen(result), "fr_FR",NULL, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in unum_openPattern(): %s\n", myErrorName(status)); |
| } |
| |
| free(result); |
| |
| /*getting the symbols of cur_def */ |
| /*set the symbols of cur_frpattern to cur_def */ |
| for (symType = UNUM_DECIMAL_SEPARATOR_SYMBOL; symType < UNUM_FORMAT_SYMBOL_COUNT; symType++) { |
| status=U_ZERO_ERROR; |
| unum_getSymbol(cur_def, symType, temp1, sizeof(temp1), &status); |
| unum_setSymbol(cur_frpattern, symType, temp1, -1, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in get/set symbols: %s\n", myErrorName(status)); |
| } |
| } |
| |
| /*format to check the result */ |
| resultlength=0; |
| resultlengthneeded=unum_format(cur_def, l, NULL, resultlength, &pos1, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| unum_format(cur_def, l, result, resultlength, &pos1, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status)); |
| } |
| |
| if(U_FAILURE(status)){ |
| log_err("Fail: error in unum_setSymbols: %s\n", myErrorName(status)); |
| } |
| unum_applyPattern(cur_frpattern, FALSE, result, u_strlen(result),NULL,NULL); |
| |
| for (symType = UNUM_DECIMAL_SEPARATOR_SYMBOL; symType < UNUM_FORMAT_SYMBOL_COUNT; symType++) { |
| status=U_ZERO_ERROR; |
| unum_getSymbol(cur_def, symType, temp1, sizeof(temp1), &status); |
| unum_getSymbol(cur_frpattern, symType, temp2, sizeof(temp2), &status); |
| if(U_FAILURE(status) || u_strcmp(temp1, temp2) != 0) |
| { |
| log_err("Fail: error in getting symbols\n"); |
| } |
| else |
| log_verbose("Pass: get and set symbols successful\n"); |
| } |
| |
| /*format and check with the previous result */ |
| |
| resultlength=0; |
| resultlengthneeded=unum_format(cur_frpattern, l, NULL, resultlength, &pos1, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| unum_format(cur_frpattern, l, temp1, resultlength, &pos1, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in formatting using unum_format(.....): %s\n", myErrorName(status)); |
| } |
| /* TODO: |
| * This test fails because we have not called unum_applyPattern(). |
| * Currently, such an applyPattern() does not exist on the C API, and |
| * we have jitterbug 411 for it. |
| * Since it is close to the 1.5 release, I (markus) am disabling this test just |
| * for this release (I added the test itself only last week). |
| * For the next release, we need to fix this. |
| * Then, remove the uprv_strcmp("1.5", ...) and this comment, and the include "cstring.h" at the beginning of this file. |
| */ |
| if(u_strcmp(result, temp1) != 0) { |
| log_err("Formatting failed after setting symbols. result=%s temp1=%s\n", result, temp1); |
| } |
| |
| |
| /*----------- */ |
| |
| free(result); |
| |
| /* Testing unum_get/setSymbol() */ |
| for(i = 0; i < UNUM_FORMAT_SYMBOL_COUNT; ++i) { |
| symbol[0] = (UChar)(0x41 + i); |
| symbol[1] = (UChar)(0x61 + i); |
| unum_setSymbol(cur_frpattern, (UNumberFormatSymbol)i, symbol, 2, &status); |
| if(U_FAILURE(status)) { |
| log_err("Error from unum_setSymbol(%d): %s\n", i, myErrorName(status)); |
| return; |
| } |
| } |
| for(i = 0; i < UNUM_FORMAT_SYMBOL_COUNT; ++i) { |
| resultlength = unum_getSymbol(cur_frpattern, (UNumberFormatSymbol)i, symbol, sizeof(symbol)/U_SIZEOF_UCHAR, &status); |
| if(U_FAILURE(status)) { |
| log_err("Error from unum_getSymbol(%d): %s\n", i, myErrorName(status)); |
| return; |
| } |
| if(resultlength != 2 || symbol[0] != 0x41 + i || symbol[1] != 0x61 + i) { |
| log_err("Failure in unum_getSymbol(%d): got unexpected symbol\n", i); |
| } |
| } |
| /*try getting from a bogus symbol*/ |
| unum_getSymbol(cur_frpattern, (UNumberFormatSymbol)i, symbol, sizeof(symbol)/U_SIZEOF_UCHAR, &status); |
| if(U_SUCCESS(status)){ |
| log_err("Error : Expected U_ILLEGAL_ARGUMENT_ERROR for bogus symbol"); |
| } |
| if(U_FAILURE(status)){ |
| if(status != U_ILLEGAL_ARGUMENT_ERROR){ |
| log_err("Error: Expected U_ILLEGAL_ARGUMENT_ERROR for bogus symbol, Got %s\n", myErrorName(status)); |
| } |
| } |
| status=U_ZERO_ERROR; |
| |
| /* Testing unum_getTextAttribute() and unum_setTextAttribute()*/ |
| log_verbose("\nTesting getting and setting text attributes\n"); |
| resultlength=5; |
| unum_getTextAttribute(cur_fr, UNUM_NEGATIVE_SUFFIX, temp, resultlength, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("Failure in gettting the Text attributes of number format: %s\n", myErrorName(status)); |
| } |
| unum_setTextAttribute(cur_def, UNUM_NEGATIVE_SUFFIX, temp, u_strlen(temp), &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("Failure in gettting the Text attributes of number format: %s\n", myErrorName(status)); |
| } |
| unum_getTextAttribute(cur_def, UNUM_NEGATIVE_SUFFIX, suffix, resultlength, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("Failure in gettting the Text attributes of number format: %s\n", myErrorName(status)); |
| } |
| if(u_strcmp(suffix,temp)!=0) |
| log_err("Fail:Error in setTextAttribute or getTextAttribute in setting and getting suffix\n"); |
| else |
| log_verbose("Pass: setting and getting suffix works fine\n"); |
| /*set it back to normal */ |
| u_uastrcpy(temp,"$"); |
| unum_setTextAttribute(cur_def, UNUM_NEGATIVE_SUFFIX, temp, u_strlen(temp), &status); |
| |
| /*checking some more text setter conditions */ |
| u_uastrcpy(prefix, "+"); |
| unum_setTextAttribute(def, UNUM_POSITIVE_PREFIX, prefix, u_strlen(prefix) , &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in setting the text attributes : %s\n", myErrorName(status)); |
| } |
| unum_getTextAttribute(def, UNUM_POSITIVE_PREFIX, temp, resultlength, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in getting the text attributes : %s\n", myErrorName(status)); |
| } |
| |
| if(u_strcmp(prefix, temp)!=0) |
| log_err("ERROR: get and setTextAttributes with positive prefix failed\n"); |
| else |
| log_verbose("Pass: get and setTextAttributes with positive prefix works fine\n"); |
| |
| u_uastrcpy(prefix, "+"); |
| unum_setTextAttribute(def, UNUM_NEGATIVE_PREFIX, prefix, u_strlen(prefix), &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in setting the text attributes : %s\n", myErrorName(status)); |
| } |
| unum_getTextAttribute(def, UNUM_NEGATIVE_PREFIX, temp, resultlength, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in getting the text attributes : %s\n", myErrorName(status)); |
| } |
| if(u_strcmp(prefix, temp)!=0) |
| log_err("ERROR: get and setTextAttributes with negative prefix failed\n"); |
| else |
| log_verbose("Pass: get and setTextAttributes with negative prefix works fine\n"); |
| |
| u_uastrcpy(suffix, "+"); |
| unum_setTextAttribute(def, UNUM_NEGATIVE_SUFFIX, suffix, u_strlen(suffix) , &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in setting the text attributes: %s\n", myErrorName(status)); |
| } |
| |
| unum_getTextAttribute(def, UNUM_NEGATIVE_SUFFIX, temp, resultlength, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in getting the text attributes : %s\n", myErrorName(status)); |
| } |
| if(u_strcmp(suffix, temp)!=0) |
| log_err("ERROR: get and setTextAttributes with negative suffix failed\n"); |
| else |
| log_verbose("Pass: get and settextAttributes with negative suffix works fine\n"); |
| |
| u_uastrcpy(suffix, "++"); |
| unum_setTextAttribute(def, UNUM_POSITIVE_SUFFIX, suffix, u_strlen(suffix) , &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in setting the text attributes: %s\n", myErrorName(status)); |
| } |
| |
| unum_getTextAttribute(def, UNUM_POSITIVE_SUFFIX, temp, resultlength, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in getting the text attributes : %s\n", myErrorName(status)); |
| } |
| if(u_strcmp(suffix, temp)!=0) |
| log_err("ERROR: get and setTextAttributes with negative suffix failed\n"); |
| else |
| log_verbose("Pass: get and settextAttributes with negative suffix works fine\n"); |
| |
| |
| |
| |
| /*Testing unum_getAttribute and unum_setAttribute() */ |
| log_verbose("\nTesting get and set Attributes\n"); |
| attr=UNUM_GROUPING_SIZE; |
| newvalue=unum_getAttribute(def, attr); |
| newvalue=2; |
| unum_setAttribute(def, attr, newvalue); |
| if(unum_getAttribute(def,attr)!=2) |
| log_err("Fail: error in setting and getting attributes for UNUM_GROUPING_SIZE\n"); |
| else |
| log_verbose("Pass: setting and getting attributes for UNUM_GROUPING_SIZE works fine\n"); |
| |
| attr=UNUM_MULTIPLIER; |
| newvalue=unum_getAttribute(def, attr); |
| newvalue=8; |
| unum_setAttribute(def, attr, newvalue); |
| if(unum_getAttribute(def,attr) != 8) |
| log_err("error in setting and getting attributes for UNUM_MULTIPLIER\n"); |
| else |
| log_verbose("Pass:setting and getting attributes for UNUM_MULTIPLIER works fine\n"); |
| |
| /*testing set and get Attributes extensively */ |
| log_verbose("\nTesting get and set attributes extensively\n"); |
| for(attr=UNUM_PARSE_INT_ONLY; attr<= UNUM_PADDING_POSITION; attr=(UNumberFormatAttribute)((int32_t)attr + 1) ) |
| { |
| newvalue=unum_getAttribute(fr, attr); |
| unum_setAttribute(def, attr, newvalue); |
| if(unum_getAttribute(def,attr)!=unum_getAttribute(fr, attr)) |
| log_err("error in setting and getting attributes\n"); |
| else |
| log_verbose("Pass: attributes set and retrieved successfully\n"); |
| } |
| |
| /*testing spellout format to make sure we can use it successfully.*/ |
| log_verbose("\nTesting spellout format\n"); |
| if (spellout_def) |
| { |
| static const int32_t values[] = { 0, -5, 105, 1005, 105050 }; |
| for (i = 0; i < LENGTH(values); ++i) { |
| UChar buffer[128]; |
| int32_t len; |
| int32_t value = values[i]; |
| status = U_ZERO_ERROR; |
| len = unum_format(spellout_def, value, buffer, LENGTH(buffer), NULL, &status); |
| if(U_FAILURE(status)) { |
| log_err("Error in formatting using unum_format(spellout_fmt, ...): %s\n", myErrorName(status)); |
| } else { |
| int32_t pp = 0; |
| int32_t parseResult; |
| char logbuf[256]; |
| ustrToAstr(buffer, len, logbuf, LENGTH(logbuf)); |
| log_verbose("formatted %d as '%s', length: %d\n", value, logbuf, len); |
| |
| parseResult = unum_parse(spellout_def, buffer, len, &pp, &status); |
| if (U_FAILURE(status)) { |
| log_err("Error in parsing using unum_format(spellout_fmt, ...): %s\n", myErrorName(status)); |
| } else if (parseResult != value) { |
| log_err("unum_format result %d != value %d\n", parseResult, value); |
| } |
| } |
| } |
| } |
| else { |
| log_err("Spellout format is unavailable\n"); |
| } |
| |
| /*closing the NumberFormat() using unum_close(UNumberFormat*)")*/ |
| unum_close(def); |
| unum_close(fr); |
| unum_close(cur_def); |
| unum_close(cur_fr); |
| unum_close(per_def); |
| unum_close(per_fr); |
| unum_close(spellout_def); |
| unum_close(pattern); |
| unum_close(cur_frpattern); |
| unum_close(myclone); |
| |
| } |
| |
| static void TestNumberFormatPadding() |
| { |
| UChar *result=NULL; |
| UChar temp1[512]; |
| |
| UErrorCode status=U_ZERO_ERROR; |
| int32_t resultlength; |
| int32_t resultlengthneeded; |
| UNumberFormat *pattern; |
| double d1; |
| double d = -10456.37; |
| UFieldPosition pos1; |
| int32_t parsepos; |
| |
| /* create a number format using unum_openPattern(....)*/ |
| log_verbose("\nTesting unum_openPattern() with padding\n"); |
| u_uastrcpy(temp1, "*#,##0.0#*;(#,##0.0#)"); |
| status=U_ZERO_ERROR; |
| pattern=unum_open(UNUM_IGNORE,temp1, u_strlen(temp1), NULL, NULL,&status); |
| if(U_SUCCESS(status)) |
| { |
| log_err("error in unum_openPattern(%s): %s\n", temp1, myErrorName(status) );; |
| } |
| else |
| { |
| unum_close(pattern); |
| } |
| |
| /* u_uastrcpy(temp1, "*x#,###,###,##0.0#;(*x#,###,###,##0.0#)"); */ |
| u_uastrcpy(temp1, "*x#,###,###,##0.0#;*x(###,###,##0.0#)"); |
| status=U_ZERO_ERROR; |
| pattern=unum_open(UNUM_IGNORE,temp1, u_strlen(temp1), "en_US",NULL, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("error in padding unum_openPattern(%s): %s\n", temp1, myErrorName(status) );; |
| } |
| else { |
| log_verbose("Pass: padding unum_openPattern() works fine\n"); |
| |
| /*test for unum_toPattern()*/ |
| log_verbose("\nTesting padding unum_toPattern()\n"); |
| resultlength=0; |
| resultlengthneeded=unum_toPattern(pattern, FALSE, NULL, resultlength, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| unum_toPattern(pattern, FALSE, result, resultlength, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("error in extracting the padding pattern from UNumberFormat: %s\n", myErrorName(status)); |
| } |
| else |
| { |
| if(u_strcmp(result, temp1)!=0) |
| log_err("FAIL: Error in extracting the padding pattern using unum_toPattern()\n"); |
| else |
| log_verbose("Pass: extracted the padding pattern correctly using unum_toPattern()\n"); |
| free(result); |
| } |
| /* u_uastrcpy(temp1, "(xxxxxxx10,456.37)"); */ |
| u_uastrcpy(temp1, "xxxxx(10,456.37)"); |
| resultlength=0; |
| pos1.field = 1; /* Fraction field */ |
| resultlengthneeded=unum_formatDouble(pattern, d, NULL, resultlength, &pos1, &status); |
| if(status==U_BUFFER_OVERFLOW_ERROR) |
| { |
| status=U_ZERO_ERROR; |
| resultlength=resultlengthneeded+1; |
| result=(UChar*)malloc(sizeof(UChar) * resultlength); |
| unum_formatDouble(pattern, d, result, resultlength, NULL, &status); |
| } |
| if(U_FAILURE(status)) |
| { |
| log_err("Error in formatting using unum_formatDouble(.....) with padding : %s\n", myErrorName(status)); |
| } |
| else |
| { |
| if(u_strcmp(result, temp1)==0) |
| log_verbose("Pass: Number Formatting using unum_formatDouble() padding Successful\n"); |
| else |
| log_err("FAIL: Error in number formatting using unum_formatDouble() with padding\n"); |
| if(pos1.beginIndex == 13 && pos1.endIndex == 15) |
| log_verbose("Pass: Complete number formatting using unum_formatDouble() successful\n"); |
| else |
| log_err("Fail: Error in complete number Formatting using unum_formatDouble()\nGot: b=%d end=%d\nExpected: b=13 end=15\n", |
| pos1.beginIndex, pos1.endIndex); |
| |
| |
| /* Testing unum_parse() and unum_parseDouble() */ |
| log_verbose("\nTesting padding unum_parseDouble()\n"); |
| parsepos=0; |
| d1=unum_parseDouble(pattern, result, u_strlen(result), &parsepos, &status); |
| if(U_FAILURE(status)) |
| { |
| log_err("padding parse failed. The error is : %s\n", myErrorName(status)); |
| } |
| |
| if(d1!=d) |
| log_err("Fail: Error in padding parsing\n"); |
| else |
| log_verbose("Pass: padding parsing successful\n"); |
| free(result); |
| } |
| } |
| |
| unum_close(pattern); |
| } |
| |
| #endif /* #if !UCONFIG_NO_FORMATTING */ |