/***********************************************************************
* Copyright (c) 1997-2007, International Business Machines Corporation
* and others. All Rights Reserved.
***********************************************************************/
#include "unicode/utypes.h"
#if !UCONFIG_NO_FORMATTING
#include "numrgts.h"
#include <float.h> // DBL_MIN, DBL_MAX
#include <stdio.h>
#include "unicode/dcfmtsym.h"
#include "unicode/decimfmt.h"
#include "unicode/locid.h"
#include "unicode/resbund.h"
#include "unicode/calendar.h"
#include "unicode/datefmt.h"
#include "unicode/ucurr.h"
#include "putilimp.h"
class MyNumberFormatTest : public NumberFormat
{
public:
virtual UClassID getDynamicClassID(void) const;
virtual UnicodeString& format( double number,
UnicodeString& toAppendTo,
FieldPosition& pos,
UErrorCode& status) const
{
return NumberFormat::format(number, toAppendTo, pos, status);
}
/* Just keep this here to make some of the compilers happy */
virtual UnicodeString& format(const Formattable& obj,
UnicodeString& toAppendTo,
FieldPosition& pos,
UErrorCode& status) const
{
return NumberFormat::format(obj, toAppendTo, pos, status);
}
/* Just use one of the format functions */
virtual UnicodeString& format( double /* number */,
UnicodeString& toAppendTo,
FieldPosition& /* pos */) const
{
toAppendTo = "";
return toAppendTo;
}
/*
public Number parse(String text, ParsePosition parsePosition)
{ return new Integer(0); }
*/
/* Just use one of the parse functions */
virtual void parse( const UnicodeString& /* text */,
Formattable& result,
ParsePosition& /* parsePosition */) const
{
result.setLong((int32_t)0);
}
virtual void parse( const UnicodeString& text,
Formattable& result,
UErrorCode& status) const
{
NumberFormat::parse(text, result, status);
}
virtual Format* clone() const
{ return NULL; }
virtual UnicodeString& format(int32_t,
UnicodeString& foo,
FieldPosition&) const
{ return foo.remove(); }
virtual UnicodeString& format(int64_t,
UnicodeString& foo,
FieldPosition&) const
{ return foo.remove(); }
virtual void applyPattern(const UnicodeString&, UParseError&, UErrorCode&){
}
};
int32_t gMyNumberFormatTestClassID;
UClassID MyNumberFormatTest::getDynamicClassID() const
{
return (UClassID)&gMyNumberFormatTestClassID;
}
// *****************************************************************************
// class NumberFormatRegressionTest
// *****************************************************************************
#define CASE(id,test) case id: name = #test; if (exec) { logln(#test "---"); logln((UnicodeString)""); test(); } break
void
NumberFormatRegressionTest::runIndexedTest( int32_t index, UBool exec, const char* &name, char* /*par*/ )
{
// if (exec) logln((UnicodeString)"TestSuite NumberFormatRegressionTest");
switch (index) {
CASE(0,Test4075713);
CASE(1,Test4074620);
CASE(2,Test4088161);
CASE(3,Test4087245);
CASE(4,Test4087535);
CASE(5,Test4088503);
CASE(6,Test4066646);
CASE(7,Test4059870);
CASE(8,Test4083018);
CASE(9,Test4071492);
CASE(10,Test4086575);
CASE(11,Test4068693);
CASE(12,Test4069754);
CASE(13,Test4087251);
CASE(14,Test4090489);
CASE(15,Test4090504);
CASE(16,Test4095713);
CASE(17,Test4092561);
CASE(18,Test4092480);
CASE(19,Test4087244);
CASE(20,Test4070798);
CASE(21,Test4071005);
CASE(22,Test4071014);
CASE(23,Test4071859);
CASE(24,Test4093610);
CASE(25,Test4098741);
CASE(26,Test4074454);
CASE(27,Test4099404);
CASE(28,Test4101481);
CASE(29,Test4052223);
CASE(30,Test4061302);
CASE(31,Test4062486);
CASE(32,Test4108738);
CASE(33,Test4106658);
CASE(34,Test4106662);
CASE(35,Test4114639);
CASE(36,Test4106664);
CASE(37,Test4106667);
CASE(38,Test4110936);
CASE(39,Test4122840);
CASE(40,Test4125885);
CASE(41,Test4134034);
CASE(42,Test4134300);
CASE(43,Test4140009);
CASE(44,Test4141750);
CASE(45,Test4145457);
CASE(46,Test4147295);
CASE(47,Test4147706);
CASE(48,Test4162198);
CASE(49,Test4162852);
CASE(50,Test4167494);
CASE(51,Test4170798);
CASE(52,Test4176114);
CASE(53,Test4179818);
CASE(54,Test4212072);
CASE(55,Test4216742);
CASE(56,Test4217661);
CASE(57,Test4161100);
CASE(58,Test4243011);
CASE(59,Test4243108);
CASE(60,TestJ691);
default: name = ""; break;
}
}
UBool
NumberFormatRegressionTest::failure(UErrorCode status, const UnicodeString& msg, const Locale& l)
{
if(U_FAILURE(status)) {
errln(UnicodeString("FAIL: ", "") + msg
+ UnicodeString(" failed, error ", "") + UnicodeString(u_errorName(status), "") + UnicodeString(l.getName(),""));
return TRUE;
}
return FALSE;
}
UBool
NumberFormatRegressionTest::failure(UErrorCode status, const UnicodeString& msg, const char *l)
{
if(U_FAILURE(status)) {
errln(UnicodeString("FAIL: ", "") + msg
+ UnicodeString(" failed, error ", "") + UnicodeString(u_errorName(status), "") + UnicodeString(l, ""));
return TRUE;
}
return FALSE;
}
UBool
NumberFormatRegressionTest::failure(UErrorCode status, const UnicodeString& msg)
{
if(U_FAILURE(status)) {
errln(UnicodeString("FAIL: ", "") + msg
+ UnicodeString(" failed, error ", "") + UnicodeString(u_errorName(status), ""));
return TRUE;
}
return FALSE;
}
/**
* Convert Java-style strings with \u Unicode escapes into UnicodeString objects
*/
inline UnicodeString str(const char *input)
{
return CharsToUnicodeString(input);
}
/* @bug 4075713
* NumberFormat.equals comparing with null should always return false.
*/
// {sfb} kind of silly in C++, just checking for new success
void NumberFormatRegressionTest::Test4075713(void)
{
//try {
MyNumberFormatTest *tmp = new MyNumberFormatTest();
if(tmp != NULL)
logln("NumberFormat.equals passed");
/*} catch (NullPointerException e) {
errln("(new MyNumberFormatTest()).equals(null) throws unexpected exception");
}*/
delete tmp;
}
/* @bug 4074620
* NumberFormat.equals comparing two obj equal even the setGroupingUsed
* flag is different.
*/
void NumberFormatRegressionTest::Test4074620(void)
{
MyNumberFormatTest *nf1 = new MyNumberFormatTest();
MyNumberFormatTest *nf2 = new MyNumberFormatTest();
nf1->setGroupingUsed(FALSE);
nf2->setGroupingUsed(TRUE);
if(nf1 == nf2)
errln("Test for bug 4074620 failed");
else
logln("Test for bug 4074620 passed.");
delete nf1;
delete nf2;
}
/* @bug 4088161
* DecimalFormat.format() incorrectly uses maxFractionDigits setting.
*/
void NumberFormatRegressionTest::Test4088161 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
failure(status, "new DecimalFormat", "");
double d = 100;
df->setMinimumFractionDigits(0);
df->setMaximumFractionDigits(16);
UnicodeString sBuf1;
FieldPosition fp1(0);
logln(UnicodeString("d = ") + d);
logln("maxFractionDigits = " + df->getMaximumFractionDigits());
logln(" format(d) = '" + df->format(d, sBuf1, fp1) + "'");
df->setMaximumFractionDigits(17);
UnicodeString sBuf2;
FieldPosition fp2(0);
logln("maxFractionDigits = " + df->getMaximumFractionDigits());
sBuf2 = df->format(d, sBuf2, fp2);
if(sBuf2 != "100")
errln(" format(d) = '" + sBuf2 + "'");
delete df;
}
/* @bug 4087245
* DecimalFormatSymbols should be cloned in the ctor DecimalFormat.
* DecimalFormat(String, DecimalFormatSymbols).
*/
void NumberFormatRegressionTest::Test4087245 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols *symbols = new DecimalFormatSymbols(status);
failure(status, "new DecimalFormatSymbols", "");
// {sfb} One note about this test: if you pass in a pointer
// to the symbols, they are adopted and this test will fail,
// even though that is the correct behavior. To test the cloning
// of the symbols, it is necessary to pass in a reference to the symbols
DecimalFormat *df = new DecimalFormat("#,##0.0", *symbols, status);
failure(status, "new DecimalFormat with symbols", "");
int32_t n = 123;
UnicodeString buf1;
UnicodeString buf2;
FieldPosition pos(FieldPosition::DONT_CARE);
logln(UnicodeString("format(") + n + ") = " +
df->format(n, buf1, pos));
symbols->setSymbol(DecimalFormatSymbols::kDecimalSeparatorSymbol, UnicodeString((UChar)0x70)); // change value of field
logln(UnicodeString("format(") + n + ") = " +
df->format(n, buf2, pos));
if(buf1 != buf2)
errln("Test for bug 4087245 failed");
delete df;
delete symbols;
}
/* @bug 4087535
* DecimalFormat.format() incorrectly formats 0.0
*/
void NumberFormatRegressionTest::Test4087535 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
failure(status, "new DecimalFormat", "");
df->setMinimumIntegerDigits(0);
double n = 0;
UnicodeString buffer;
FieldPosition pos(FieldPosition::DONT_CARE);
buffer = df->format(n, buffer, pos);
if (buffer.length() == 0)
errln(/*n + */": '" + buffer + "'");
n = 0.1;
buffer = df->format(n, buffer, pos);
if (buffer.length() == 0)
errln(/*n + */": '" + buffer + "'");
delete df;
}
/* @bug 4088503
* DecimalFormat.format fails when groupingSize is set to 0.
*/
// {sfb} how do I tell if this worked? --> FieldPosition doesn't change ??
void NumberFormatRegressionTest::Test4088503 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
failure(status, "new DecimalFormat", "");
df->setGroupingSize(0);
UnicodeString sBuf;
FieldPosition fp(FieldPosition::DONT_CARE);
//try {
logln(df->format((int32_t)123, sBuf, fp));
//if(fp == FieldPosition(0))
// errln("Test for bug 4088503 failed.");
/*} catch (Exception foo) {
errln("Test for bug 4088503 failed.");
}*/
delete df;
}
/* @bug 4066646
* NumberFormat.getCurrencyInstance is wrong.
*/
void NumberFormatRegressionTest::Test4066646 (void)
{
assignFloatValue(2.04f);
assignFloatValue(2.03f);
assignFloatValue(2.02f);
assignFloatValue(0.0f);
}
float
NumberFormatRegressionTest::assignFloatValue(float returnfloat)
{
logln(UnicodeString(" VALUE ") + returnfloat);
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nfcommon = NumberFormat::createCurrencyInstance(Locale::getUS(), status);
if (failure(status, "NumberFormat::createCurrencyInstance", Locale::getUS())){
delete nfcommon;
return returnfloat;
}
nfcommon->setGroupingUsed(FALSE);
UnicodeString stringValue;
stringValue = nfcommon->format(returnfloat, stringValue);
logln(" DISPLAYVALUE " + stringValue);
Formattable result;
nfcommon->parse(stringValue, result, status);
failure(status, "nfcommon->parse", Locale::getUS());
float floatResult = (float) (result.getType() == Formattable::kDouble
? result.getDouble() : result.getLong());
if( uprv_fabs(floatResult - returnfloat) > 0.0001)
//String stringValue = nfcommon.format(returnfloat).substring(1);
//if (Float.valueOf(stringValue).floatValue() != returnfloat)
errln(UnicodeString("FAIL: expected ") + returnfloat + ", got " + floatResult + " (" + stringValue+")");
delete nfcommon;
return returnfloat;
} // End Of assignFloatValue()
/* @bug 4059870
* DecimalFormat throws exception when parsing "0"
*/
void NumberFormatRegressionTest::Test4059870(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *format = new DecimalFormat("00", status);
failure(status, "new Decimalformat", Locale::getUS());
//try {
Formattable result;
UnicodeString str;
format->parse(UnicodeString("0"), result, status);
failure(status, "format->parse", Locale::getUS());
/*}
catch (Exception e) {
errln("Test for bug 4059870 failed : " + e);
}*/
delete format;
}
/* @bug 4083018
* DecimalFormatSymbol.equals should always return false when
* comparing with null.
*/
// {sfb} this is silly in C++
void NumberFormatRegressionTest::Test4083018 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols *dfs = new DecimalFormatSymbols(status);
failure(status, "new DecimalFormatSymbols", Locale::getUS());
//try {
if (dfs != NULL)
logln("Test Passed!");
else
errln("Test for bug 4083018 failed");
/*} catch (Exception foo) {
errln("Test for bug 4083018 failed => Message : " + foo.getMessage());
}*/
delete dfs;
}
/* @bug 4071492
* DecimalFormat does not round up correctly.
*/
void NumberFormatRegressionTest::Test4071492 (void)
{
double x = 0.00159999;
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nf = NumberFormat::createInstance(status);
if (failure(status, "NumberFormat::createInstance", Locale::getUS())) {
delete nf;
return;
}
nf->setMaximumFractionDigits(4);
UnicodeString out;
FieldPosition pos(FieldPosition::DONT_CARE);
out = nf->format(x, out, pos);
logln("0.00159999 formats with 4 fractional digits to " + out);
UnicodeString expected("0.0016");
if (out != expected)
errln("FAIL: Expected " + expected);
delete nf;
}
/* @bug 4086575
* A space as a group separator for localized pattern causes
* wrong format. WorkAround : use non-breaking space.
*/
void NumberFormatRegressionTest::Test4086575(void)
{
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nf1 = NumberFormat::createInstance(Locale::getFrance(), status);
// TODO: There is not a good way to find out that the creation of this number format has
// failed. Major rewiring of format construction proposed.
if(U_FAILURE(status)) {
errln("Something is wrong with French number format - it should not fallback. Exitting");
delete nf1;
return;
}
failure(status, "NumberFormat::createInstance", Locale::getFrance());
// C++ workaround to make sure cast works
// Wouldn't dynamic_cast<DecimalFormat*> be great?
if(nf1->getDynamicClassID() != DecimalFormat::getStaticClassID()) {
errln("NumberFormat::createInstance returned incorrect type.");
return;
}
DecimalFormat *nf = (DecimalFormat*) nf1;
UnicodeString temp;
logln("nf toPattern1: " + nf->toPattern(temp));
logln("nf toLocPattern1: " + nf->toLocalizedPattern(temp));
// No group separator
logln("...applyLocalizedPattern ###,00;(###,00) ");
nf->applyLocalizedPattern(UnicodeString("###,00;(###,00)"), status);
failure(status, "nf->applyLocalizedPattern", Locale::getFrance());
logln("nf toPattern2: " + nf->toPattern(temp));
logln("nf toLocPattern2: " + nf->toLocalizedPattern(temp));
FieldPosition pos(FieldPosition::DONT_CARE);
logln("nf: " + nf->format((int32_t)1234, temp, pos)); // 1234,00
logln("nf: " + nf->format((int32_t)-1234, temp, pos)); // (1234,00)
// Space as group separator
logln("...applyLocalizedPattern # ###,00;(# ###,00) ");
// nbsp = \u00a0
//nf->applyLocalizedPattern("#\u00a0###,00;(#\u00a0###,00)");
UChar patChars[] = {
0x23, 0x00a0, 0x23, 0x23, 0x23, 0x2c, 0x30, 0x30, 0x3b,
0x28, 0x23, 0x00a0, 0x23, 0x23, 0x23, 0x2c, 0x30, 0x30, 0x29
};
UnicodeString pat(patChars, 19, 19);
nf->applyLocalizedPattern(pat, status);
failure(status, "nf->applyLocalizedPattern", Locale::getFrance());
logln("nf toPattern2: " + nf->toPattern(temp));
logln("nf toLocPattern2: " + nf->toLocalizedPattern(temp));
UnicodeString buffer;
buffer = nf->format((int32_t)1234, buffer, pos);
//if (buffer != UnicodeString("1\u00a0234,00"))
UChar c[] = {
0x31, 0x00a0, 0x32, 0x33, 0x34, 0x2c, 0x30, 0x30
};
UnicodeString cc(c, 8, 8);
if (buffer != cc)
errln("nf : " + buffer); // Expect 1 234,00
buffer.remove();
buffer = nf->format((int32_t)-1234, buffer, pos);
UChar c1[] = {
0x28, 0x31, 0x00a0, 0x32, 0x33, 0x34, 0x2c, 0x30, 0x30, 0x29
};
UnicodeString cc1(c1, 10, 10);
if (buffer != cc1)
errln("nf : " + buffer); // Expect (1 234,00)
// Erroneously prints:
// 1234,00 ,
// (1234,00 ,)
delete nf1;
}
/* @bug 4068693
* DecimalFormat.parse returns wrong value
*/
// {sfb} slightly converted into a round-trip test, since in C++
// there is no Double.toString()
void NumberFormatRegressionTest::Test4068693(void)
{
logln("----- Test Application -----");
ParsePosition pos(0);
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
Formattable d;
//Double d = (Double)df.parse("123.55456", pos=new ParsePosition(0));
df->parse(UnicodeString("123.55456"), d, pos);
//if (!d.toString().equals("123.55456")) {
UnicodeString dstr;
df->setMaximumFractionDigits(999);
df->setMaximumIntegerDigits(999);
FieldPosition fp(FieldPosition::DONT_CARE);
dstr = df->format(d.getDouble(), dstr, fp);
if (dstr != UnicodeString("123.55456")) {
errln(UnicodeString("Result -> ") + d.getDouble());
}
delete df;
}
/* @bug 4069754, 4067878
* null pointer thrown when accessing a deserialized DecimalFormat
* object.
*/
// {sfb} doesn't apply in C++
void NumberFormatRegressionTest::Test4069754(void)
{
/* try {
myformat it = new myformat();
logln(it.Now());
FileOutputStream ostream = new FileOutputStream("t.tmp");
ObjectOutputStream p = new ObjectOutputStream(ostream);
p.writeObject(it);
ostream.close();
logln("Saved ok.");
FileInputStream istream = new FileInputStream("t.tmp");
ObjectInputStream p2 = new ObjectInputStream(istream);
myformat it2 = (myformat)p2.readObject();
logln(it2.Now());
istream.close();
logln("Loaded ok.");
} catch (Exception foo) {
errln("Test for bug 4069754 or 4057878 failed => Exception: " + foo.getMessage());
}
*/}
/* @bug 4087251
* DecimalFormat.applyPattern(String) allows illegal patterns
*/
void NumberFormatRegressionTest::Test4087251 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
//try {
df->applyPattern(UnicodeString("#.#.#"), status);
if( ! U_FAILURE(status))
errln("df->applyPattern with illegal pattern didn't fail");
UnicodeString temp;
logln("toPattern() returns \"" + df->toPattern(temp) + "\"");
//errln("applyPattern(\"#.#.#\") doesn't throw IllegalArgumentException");
/*} catch (IllegalArgumentException e) {
logln("Caught Illegal Argument Error !");
}*/
// Second test; added 5/11/98 when reported to fail on 1.2b3
//try {
df->applyPattern("#0.0#0#0", status);
if( ! U_FAILURE(status))
errln("df->applyPattern with illegal pattern didn't fail");
logln("toPattern() returns \"" + df->toPattern(temp) + "\"");
//errln("applyPattern(\"#0.0#0#0\") doesn't throw IllegalArgumentException");
/*} catch (IllegalArgumentException e) {
logln("Ok - IllegalArgumentException for #0.0#0#0");
}*/
delete df;
}
/* @bug 4090489
* DecimalFormat.format() loses precision
*/
void NumberFormatRegressionTest::Test4090489 (void)
{
// {sfb} sprintf doesn't correctly handle the double, so there is nothing
// that NumberFormat can do. For some reason, it does not format the last 1.
/* UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
failure(status, "new DecimalFormat");
df->setMinimumFractionDigits(10);
df->setMaximumFractionDigits(999);
df->setGroupingUsed(FALSE);
double d = 1.000000000000001E7;
//BigDecimal bd = new BigDecimal(d);
UnicodeString sb;
FieldPosition fp(0);
logln(UnicodeString("d = ") + d);
//logln("BigDecimal.toString(): " + bd.toString());
df->format(d, sb, fp);
if (sb != "10000000.0000000100") {
errln("DecimalFormat.format(): " + sb);
}
*/
}
/* @bug 4090504
* DecimalFormat.format() loses precision
*/
void NumberFormatRegressionTest::Test4090504 (void)
{
double d = 1;
logln(UnicodeString("d = ") + d);
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
UnicodeString sb;
FieldPosition fp(FieldPosition::DONT_CARE);
//try {
for (int i = 17; i <= 20; i++) {
df->setMaximumFractionDigits(i);
//sb = new StringBuffer("");
fp.setField(0);
logln(" getMaximumFractionDigits() = " + i);
logln(" formated: " + df->format(d, sb, fp));
}
/*} catch (Exception foo) {
errln("Bug 4090504 regression test failed. Message : " + foo.getMessage());
}*/
delete df;
}
/* @bug 4095713
* DecimalFormat.parse(String str, ParsePosition pp) loses precision
*/
void NumberFormatRegressionTest::Test4095713 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
UnicodeString str("0.1234");
double d1 = 0.1234;
//Double d1 = new Double(str);
//Double d2 = (Double) df.parse(str, new ParsePosition(0));
Formattable d2;
ParsePosition pp(0);
df->parse(str, d2, pp);
logln(UnicodeString("") + d1);
if (d2.getDouble() != d1)
errln(UnicodeString("Bug 4095713 test failed, new double value : ") + d2.getDouble());
delete df;
}
/* @bug 4092561
* DecimalFormat.parse() fails when multiplier is not set to 1
*/
// {sfb} not sure what to do with this one
void NumberFormatRegressionTest::Test4092561 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
// {sfb} going to cheat here and use sprintf ??
/*UnicodeString str = Long.toString(Long.MIN_VALUE);
logln("Long.MIN_VALUE : " + df.parse(str, new ParsePosition(0)).toString());
df.setMultiplier(100);
Number num = df.parse(str, new ParsePosition(0));
if (num.doubleValue() != -9.223372036854776E16)
errln("Bug 4092561 test failed when multiplier is set to not 1.");
*/
delete df;
}
/* @bug 4092480
* DecimalFormat: Negative format ignored.
*/
void NumberFormatRegressionTest::Test4092480 (void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *dfFoo = new DecimalFormat(UnicodeString("000"), status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete dfFoo;
return;
}
failure(status, "new DecimalFormat");
//try {
dfFoo->applyPattern("0000;-000", status);
failure(status, "dfFoo->applyPattern");
UnicodeString temp;
if (dfFoo->toPattern(temp) != UnicodeString("#0000"))
errln("dfFoo.toPattern : " + dfFoo->toPattern(temp));
FieldPosition pos(FieldPosition::DONT_CARE);
logln(dfFoo->format((int32_t)42, temp, pos));
logln(dfFoo->format((int32_t)-42, temp, pos));
dfFoo->applyPattern("000;-000", status);
failure(status, "dfFoo->applyPattern");
if (dfFoo->toPattern(temp) != UnicodeString("#000"))
errln("dfFoo.toPattern : " + dfFoo->toPattern(temp));
logln(dfFoo->format((int32_t)42,temp, pos));
logln(dfFoo->format((int32_t)-42, temp, pos));
dfFoo->applyPattern("000;-0000", status);
failure(status, "dfFoo->applyPattern");
if (dfFoo->toPattern(temp) != UnicodeString("#000"))
errln("dfFoo.toPattern : " + dfFoo->toPattern(temp));
logln(dfFoo->format((int32_t)42, temp, pos));
logln(dfFoo->format((int32_t)-42, temp, pos));
dfFoo->applyPattern("0000;-000", status);
failure(status, "dfFoo->applyPattern");
if (dfFoo->toPattern(temp) != UnicodeString("#0000"))
errln("dfFoo.toPattern : " + dfFoo->toPattern(temp));
logln(dfFoo->format((int32_t)42, temp, pos));
logln(dfFoo->format((int32_t)-42, temp, pos));
/*} catch (Exception foo) {
errln("Message " + foo.getMessage());
}*/
delete dfFoo;
}
/* @bug 4087244
* NumberFormat.getCurrencyInstance() produces format that uses
* decimal separator instead of monetary decimal separator.
*
* Rewrote this test not to depend on the actual pattern. Pattern should
* never contain the monetary separator! Decimal separator in pattern is
* interpreted as monetary separator if currency symbol is seen!
*/
void NumberFormatRegressionTest::Test4087244 (void) {
UErrorCode status = U_ZERO_ERROR;
char loc[256] = {0};
uloc_canonicalize("pt_PT_PREEURO", loc, 256, &status);
Locale *de = new Locale(loc);
NumberFormat *nf = NumberFormat::createCurrencyInstance(*de, status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete nf;
return;
}
if (nf->getDynamicClassID() != DecimalFormat::getStaticClassID()) {
errln("expected DecimalFormat!");
return;
}
DecimalFormat *df = (DecimalFormat*) nf;
const DecimalFormatSymbols *sym = df->getDecimalFormatSymbols();
UnicodeString decSep = sym->getSymbol(DecimalFormatSymbols::kDecimalSeparatorSymbol);
UnicodeString monSep = sym->getSymbol(DecimalFormatSymbols::kMonetarySeparatorSymbol);
if (decSep == monSep) {
errln("ERROR in test: want decimal sep != monetary sep");
return;
}
df->setMinimumIntegerDigits(1);
df->setMinimumFractionDigits(2);
UnicodeString str;
FieldPosition pos;
df->format(1.23, str, pos);
UnicodeString monStr("1x23");
monStr.replace((int32_t)1, 1, monSep);
UnicodeString decStr("1x23");
decStr.replace((int32_t)1, 1, decSep);
if (str.indexOf(monStr) >= 0 && str.indexOf(decStr) < 0) {
logln(UnicodeString("OK: 1.23 -> \"") + str + "\" contains \"" +
monStr + "\" and not \"" + decStr + '"');
} else {
errln(UnicodeString("FAIL: 1.23 -> \"") + str + "\", should contain \"" +
monStr +
"\" and not \"" + decStr + '"');
}
delete de;
delete nf;
}
/* @bug 4070798
* Number format data rounding errors for locale FR
*/
void NumberFormatRegressionTest::Test4070798 (void)
{
NumberFormat *formatter;
UnicodeString tempString;
/* User error :
String expectedDefault = "-5\u00a0789,987";
String expectedCurrency = "5\u00a0789,98 F";
String expectedPercent = "-578\u00a0998%";
*/
UChar chars1 [] = {
0x2d, 0x35, 0x00a0, 0x37, 0x38, 0x39, 0x2c, 0x39, 0x38, 0x38
};
UChar chars2 [] = {
0x35, 0x00a0, 0x37, 0x38, 0x39, 0x2c, 0x39, 0x39, 0x20, 0x46
};
UChar chars3 [] = {
0x2d, 0x35, 0x37, 0x38, 0x00a0, 0x39, 0x39, 0x39, 0x00a0, 0x25
};
UnicodeString expectedDefault(chars1, 10, 10);
UnicodeString expectedCurrency(chars2, 10, 10);
UnicodeString expectedPercent(chars3, 10, 10);
UErrorCode status = U_ZERO_ERROR;
char loc[256]={0};
int len = uloc_canonicalize("fr_FR_PREEURO", loc, 256, &status);
formatter = NumberFormat::createInstance(Locale(loc), status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete formatter;
return;
}
failure(status, "NumberFormat::createNumberInstance", loc);
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedDefault) {
logln ("Bug 4070798 default test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedDefault +
" Received " + tempString );
}
delete formatter;
len = uloc_canonicalize("fr_FR_PREEURO", loc, 256, &status);
formatter = NumberFormat::createCurrencyInstance(loc, status);
failure(status, "NumberFormat::createCurrencyInstance", loc);
tempString.remove();
tempString = formatter->format( 5789.9876, tempString );
if (tempString == expectedCurrency) {
logln ("Bug 4070798 currency test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedCurrency +
" Received " + tempString );
}
delete formatter;
uloc_canonicalize("fr_FR_PREEURO", loc, 256, &status);
formatter = NumberFormat::createPercentInstance(Locale(loc), status);
failure(status, "NumberFormat::createPercentInstance", loc);
tempString.remove();
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedPercent) {
logln ("Bug 4070798 percentage test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedPercent +
" Received " + tempString );
}
delete formatter;
}
/* @bug 4071005
* Data rounding errors for French (Canada) locale
*/
void NumberFormatRegressionTest::Test4071005 (void)
{
NumberFormat *formatter;
UnicodeString tempString;
/* User error :
String expectedDefault = "-5\u00a0789,987";
String expectedCurrency = "5\u00a0789,98 $";
String expectedPercent = "-578\u00a0998%";
*/
UChar chars1 [] = {
0x2d, 0x35, 0x00a0, 0x37, 0x38, 0x39, 0x2c, 0x39, 0x38, 0x38
};
UChar chars2 [] = {
0x35, 0x00a0, 0x37, 0x38, 0x39, 0x2c, 0x39, 0x39, 0x20, 0x24
};
UChar chars3 [] = {
0x2d, 0x35, 0x37, 0x38, 0x00a0, 0x39, 0x39, 0x39, 0x00a0, 0x25
};
UnicodeString expectedDefault(chars1, 10, 10);
UnicodeString expectedCurrency(chars2, 10, 10);
UnicodeString expectedPercent(chars3, 10, 10);
UErrorCode status = U_ZERO_ERROR;
formatter = NumberFormat::createInstance(Locale::getCanadaFrench(), status);
if (failure(status, "NumberFormat::createNumberInstance", Locale::getCanadaFrench())){
delete formatter;
return;
};
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedDefault) {
logln ("Bug 4071005 default test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedDefault +
" Received " + tempString );
}
delete formatter;
formatter = NumberFormat::createCurrencyInstance(Locale::getCanadaFrench(), status);
failure(status, "NumberFormat::createCurrencyInstance", Locale::getCanadaFrench());
tempString.remove();
tempString = formatter->format( 5789.9876, tempString );
if (tempString == expectedCurrency) {
logln ("Bug 4071005 currency test assed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedCurrency +
" Received " + tempString );
}
delete formatter;
formatter = NumberFormat::createPercentInstance(Locale::getCanadaFrench(), status);
failure(status, "NumberFormat::createPercentInstance", Locale::getCanadaFrench());
tempString.remove();
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedPercent) {
logln ("Bug 4071005 percentage test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedPercent +
" Received " + tempString );
}
delete formatter;
}
/* @bug 4071014
* Data rounding errors for German (Germany) locale
*/
void NumberFormatRegressionTest::Test4071014 (void)
{
NumberFormat *formatter;
UnicodeString tempString;
/* user error :
String expectedDefault = "-5.789,987";
String expectedCurrency = "5.789,98 DM";
String expectedPercent = "-578.998%";
*/
UnicodeString expectedDefault("-5.789,988");
UnicodeString expectedCurrency("5.789,99 DM");
UnicodeString expectedPercent("-578.999 %");
UErrorCode status = U_ZERO_ERROR;
char loc[256]={0};
uloc_canonicalize("de_DE_PREEURO", loc, 256, &status);
formatter = NumberFormat::createInstance(Locale(loc), status);
if (failure(status, "NumberFormat::createNumberInstance", loc)){
delete formatter;
return;
}
tempString.remove();
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedDefault) {
logln ("Bug 4071014 default test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedDefault +
" Received " + tempString );
}
delete formatter;
uloc_canonicalize("de_DE_PREEURO", loc, 256, &status);
formatter = NumberFormat::createCurrencyInstance(Locale(loc), status);
failure(status, "NumberFormat::createCurrencyInstance", loc);
tempString.remove();
tempString = formatter->format( 5789.9876, tempString );
if (tempString == expectedCurrency) {
logln ("Bug 4071014 currency test assed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedCurrency +
" Received " + tempString );
}
delete formatter;
formatter = NumberFormat::createPercentInstance(Locale::getGermany(), status);
failure(status, "NumberFormat::createPercentInstance", Locale::getGermany());
tempString.remove();
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedPercent) {
logln ("Bug 4071014 percentage test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedPercent +
" Received " + tempString );
}
delete formatter;
}
/* @bug 4071859
* Data rounding errors for Italian locale number formats
*/
void NumberFormatRegressionTest::Test4071859 (void)
{
NumberFormat *formatter;
UnicodeString tempString;
/* user error :
String expectedDefault = "-5.789,987";
String expectedCurrency = "-L. 5.789,98";
String expectedPercent = "-578.998%";
*/
UnicodeString expectedDefault("-5.789,988");
UnicodeString expectedCurrency("-\\u20A4 5.790");
UnicodeString expectedPercent("-578.999%");
expectedCurrency = expectedCurrency.unescape();
UErrorCode status = U_ZERO_ERROR;
char loc[256]={0};
uloc_canonicalize("it_IT_PREEURO", loc, 256, &status);
formatter = NumberFormat::createInstance(Locale(loc), status);
if (failure(status, "NumberFormat::createNumberInstance")){
delete formatter;
return;
};
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedDefault) {
logln ("Bug 4071859 default test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedDefault +
" Received " + tempString );
}
delete formatter;
uloc_canonicalize("it_IT_PREEURO", loc, 256, &status);
formatter = NumberFormat::createCurrencyInstance(Locale(loc), status);
failure(status, "NumberFormat::createCurrencyInstance");
tempString.remove();
tempString = formatter->format( -5789.9876, tempString );
if (tempString == expectedCurrency) {
logln ("Bug 4071859 currency test assed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedCurrency +
" Received " + tempString );
}
delete formatter;
uloc_canonicalize("it_IT_PREEURO", loc, 256, &status);
formatter = NumberFormat::createPercentInstance(Locale(loc), status);
failure(status, "NumberFormat::createPercentInstance");
tempString.remove();
tempString = formatter->format (-5789.9876, tempString);
if (tempString == expectedPercent) {
logln ("Bug 4071859 percentage test passed.");
} else {
errln(UnicodeString("Failed:") +
" Expected " + expectedPercent +
" Received " + tempString );
}
delete formatter;
}
/* @bug 4071859
* Test rounding for nearest even.
*/
void NumberFormatRegressionTest::Test4093610(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat("#0.#", status);
failure(status, "new DecimalFormat");
UnicodeString s("12.4");
roundingTest(df, 12.35, s);
roundingTest(df, 12.45, s);
s = "12.5";
roundingTest(df, 12.452,s);
s = "12.6";
roundingTest(df, 12.55, s);
roundingTest(df, 12.65, s);
s = "12.7";
roundingTest(df, 12.652,s);
s = "12.8";
roundingTest(df, 12.75, s);
roundingTest(df, 12.752,s);
roundingTest(df, 12.85, s);
s = "12.9";
roundingTest(df, 12.852,s);
s = "13";
roundingTest(df, 12.95, s);
roundingTest(df, 12.952,s);
delete df;
}
void NumberFormatRegressionTest::roundingTest(DecimalFormat *df, double x, UnicodeString& expected)
{
UnicodeString out;
FieldPosition pos(FieldPosition::DONT_CARE);
out = df->format(x, out, pos);
logln(UnicodeString("") + x + " formats with 1 fractional digits to " + out);
if (out != expected)
errln("FAIL: Expected " + expected);
}
/* @bug 4098741
* Tests the setMaximumFractionDigits limit.
*/
void NumberFormatRegressionTest::Test4098741(void)
{
//try {
UErrorCode status = U_ZERO_ERROR;
NumberFormat *fmt = NumberFormat::createPercentInstance(status);
if (U_FAILURE(status)) {
dataerrln("Error calling NumberFormat::createPercentInstance");
delete fmt;
return;
}
fmt->setMaximumFractionDigits(20);
UnicodeString temp;
logln(fmt->format(.001, temp));
/*} catch (Exception foo) {
errln("Bug 4098471 failed with exception thrown : " + foo.getMessage());
}*/
delete fmt;
}
/* @bug 4074454
* Tests illegal pattern exception.
* Fix comment : HShih A31 Part1 will not be fixed and javadoc needs to be updated.
* Part2 has been fixed.
*/
void NumberFormatRegressionTest::Test4074454(void)
{
//try {
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *fmt = new DecimalFormat("#,#00.00;-#.#", status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete fmt;
return;
}
failure(status, "new DecimalFormat");
logln("Inconsistent negative pattern is fine.");
DecimalFormat *newFmt = new DecimalFormat("#,#00.00 p''ieces;-#,#00.00 p''ieces", status);
failure(status, "new DecimalFormat");
UnicodeString tempString;
FieldPosition pos(FieldPosition::DONT_CARE);
tempString = newFmt->format(3456.78, tempString, pos);
if (tempString != UnicodeString("3,456.78 p'ieces"))
errln("Failed! 3456.78 p'ieces expected, but got : " + tempString);
/*} catch (Exception foo) {
errln("An exception was thrown for any inconsistent negative pattern.");
}*/
delete fmt;
delete newFmt;
}
/* @bug 4099404
* Tests all different comments.
* Response to some comments :
* [1] DecimalFormat.parse API documentation is more than just one line.
* This is not a reproducable doc error in 116 source code.
* [2] See updated javadoc.
* [3] Fixed.
* [4] NumberFormat.parse(String, ParsePosition) : If parsing fails,
* a null object will be returned. The unchanged parse position also
* reflects an error.
* NumberFormat.parse(String) : If parsing fails, an ParseException
* will be thrown.
* See updated javadoc for more details.
* [5] See updated javadoc.
* [6] See updated javadoc.
* [7] This is a correct behavior if the DateFormat object is linient.
* Otherwise, an IllegalArgumentException will be thrown when formatting
* "January 35". See GregorianCalendar class javadoc for more details.
*/
void NumberFormatRegressionTest::Test4099404(void)
{
//try {
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *fmt = new DecimalFormat(UnicodeString("000.0#0"), status);
if(! U_FAILURE(status))
errln(UnicodeString("Bug 4099404 failed applying illegal pattern \"000.0#0\""));
/*} catch (Exception foo) {
logln("Bug 4099404 pattern \"000.0#0\" passed");
}*/
delete fmt;
fmt = 0;
//try {
fmt = new DecimalFormat(UnicodeString("0#0.000"), status);
if( !U_FAILURE(status))
errln("Bug 4099404 failed applying illegal pattern \"0#0.000\"");
/*} catch (Exception foo) {
logln("Bug 4099404 pattern \"0#0.000\" passed");
}*/
delete fmt;
}
/* @bug 4101481
* DecimalFormat.applyPattern doesn't set minimum integer digits
*/
void NumberFormatRegressionTest::Test4101481(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *sdf = new DecimalFormat(UnicodeString("#,##0"), status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete sdf;
return;
}
failure(status, "new DecimalFormat");
if (sdf->getMinimumIntegerDigits() != 1)
errln("Minimum integer digits : " + sdf->getMinimumIntegerDigits());
delete sdf;
}
/* @bug 4052223 (API addition request A27)
* Tests ParsePosition.setErrorPosition() and ParsePosition.getErrorPosition().
*/
void NumberFormatRegressionTest::Test4052223(void)
{
//try {
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *fmt = new DecimalFormat(UnicodeString("#,#00.00"), status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete fmt;
return;
}
failure(status, "new DecimalFormat");
Formattable num;
fmt->parse(UnicodeString("abc3"), num, status);
if(! U_FAILURE(status))
errln(UnicodeString("Bug 4052223 failed : can't parse string \"a\". Got ") /*+ num*/);
/*} catch (ParseException foo) {
logln("Caught expected ParseException : " + foo.getMessage() + " at index : " + foo.getErrorOffset());
}*/
delete fmt;
}
/* @bug 4061302
* API tests for API addition request A9.
*/
void NumberFormatRegressionTest::Test4061302(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols *fmt = new DecimalFormatSymbols(status);
failure(status, "new DecimalFormatSymbols");
UnicodeString currency(fmt->getSymbol(DecimalFormatSymbols::kCurrencySymbol));
UnicodeString intlCurrency(fmt->getSymbol(DecimalFormatSymbols::kIntlCurrencySymbol));
UnicodeString monDecSeparator(fmt->getSymbol(DecimalFormatSymbols::kMonetarySeparatorSymbol));
if (currency == UnicodeString("") ||
intlCurrency == UnicodeString("") ||
monDecSeparator == UnicodeString(""))
{
errln("getCurrencySymbols failed, got empty string.");
}
UnicodeString monDecSeparatorStr;
monDecSeparatorStr.append(monDecSeparator);
logln((UnicodeString)"Before set ==> Currency : " + currency +(UnicodeString)" Intl Currency : " + intlCurrency + (UnicodeString)" Monetary Decimal Separator : " + monDecSeparatorStr);
fmt->setSymbol(DecimalFormatSymbols::kCurrencySymbol, UnicodeString("XYZ"));
fmt->setSymbol(DecimalFormatSymbols::kIntlCurrencySymbol, UnicodeString("ABC"));
fmt->setSymbol(DecimalFormatSymbols::kMonetarySeparatorSymbol, UnicodeString((UChar)0x002A/*'*'*/));
currency = fmt->getSymbol(DecimalFormatSymbols::kCurrencySymbol);
intlCurrency = fmt->getSymbol(DecimalFormatSymbols::kIntlCurrencySymbol);
monDecSeparator = fmt->getSymbol(DecimalFormatSymbols::kMonetarySeparatorSymbol);
if (currency != UnicodeString("XYZ") ||
intlCurrency != UnicodeString("ABC") ||
monDecSeparator != UnicodeString((UChar)0x002A/*'*'*/)) {
errln("setCurrencySymbols failed.");
}
monDecSeparatorStr.remove();
monDecSeparatorStr.append(monDecSeparator);
logln("After set ==> Currency : " + currency + " Intl Currency : " + intlCurrency + " Monetary Decimal Separator : " + monDecSeparatorStr);
delete fmt;
}
/* @bug 4062486
* API tests for API addition request A23. FieldPosition.getBeginIndex and
* FieldPosition.getEndIndex.
*/
void NumberFormatRegressionTest::Test4062486(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *fmt = new DecimalFormat(UnicodeString("#,##0.00"), status);
failure(status, "new DecimalFormat");
UnicodeString formatted;
FieldPosition field(0);
double num = 1234.5;
fmt->format(num, formatted, field);
if (field.getBeginIndex() != 0 && field.getEndIndex() != 5)
errln(UnicodeString("Format 1234.5 failed. Begin index: ") /*+ field.getBeginIndex() + " End index: " + field.getEndIndex()*/);
field.setBeginIndex(7);
field.setEndIndex(4);
if (field.getBeginIndex() != 7 && field.getEndIndex() != 4)
errln("Set begin/end field indexes failed. Begin index: " /*+ field.getBeginIndex() + " End index: " + field.getEndIndex()*/);
delete fmt;
}
/* @bug 4108738
* DecimalFormat.parse incorrectly works with a group separator.
*/
void NumberFormatRegressionTest::Test4108738(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols *syms = new DecimalFormatSymbols(Locale::getUS(), status);
failure(status, "new DecimalFormatSymbols");
DecimalFormat *df = new DecimalFormat("#,##0.###", syms, status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
UnicodeString text("1.222,111");
Formattable num;
ParsePosition pp(0);
df->parse(text, num, pp);
// {sfb} how to do this (again) ?
// shouldn't just be another round-trip test, should it?
UnicodeString temp;
FieldPosition pos(FieldPosition::DONT_CARE);
temp = df->format(num.getDouble(), temp, pos);
//if (!num.toString().equals("1.222"))
if (temp != UnicodeString("1.222"))
//errln("\"" + text + "\" is parsed as " + num);
errln("\"" + text + "\" is parsed as " + temp);
text = UnicodeString("1.222x111");
pp = ParsePosition(0);
df->parse(text, num, pp);
temp.remove();
temp = df->format(num.getDouble(), temp, pos);
//if (!num.toString().equals("1.222"))
if (temp != UnicodeString("1.222"))
errln("\"" + text + "\" is parsed as " + temp);
delete df;
}
/* @bug 4106658
* DecimalFormat.format() incorrectly formats negative doubles.
*/
void NumberFormatRegressionTest::Test4106658(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status); // Corrected; see 4147706
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
volatile double d1 = 0.0; // volatile to prevent code optimization
double d2 = -0.0001;
UnicodeString buffer;
UnicodeString temp;
FieldPosition pos(FieldPosition::DONT_CARE);
#if defined(U_HPUX)
d1 = 0.0 * -1.0; // old HPUX compiler ignores volatile keyword
#else
d1 *= -1.0; // Some compilers have a problem with defining -0.0
#endif
logln("pattern: \"" + df->toPattern(temp) + "\"");
df->format(d1, buffer, pos);
if (buffer != UnicodeString("-0")) // Corrected; see 4147706
errln(UnicodeString("") + d1 + " is formatted as " + buffer);
buffer.remove();
df->format(d2, buffer, pos);
if (buffer != UnicodeString("-0")) // Corrected; see 4147706
errln(UnicodeString("") + d2 + " is formatted as " + buffer);
delete df;
}
/* @bug 4106662
* DecimalFormat.parse returns 0 if string parameter is incorrect.
*/
void NumberFormatRegressionTest::Test4106662(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
UnicodeString text("x");
ParsePosition pos1(0), pos2(0);
UnicodeString temp;
logln("pattern: \"" + df->toPattern(temp) + "\"");
Formattable num;
df->parse(text, num, pos1);
if (pos1 == ParsePosition(0)/*num != null*/) {
errln(UnicodeString("Test Failed: \"") + text + "\" is parsed as " /*+ num*/);
}
delete df;
df = new DecimalFormat(UnicodeString("$###.00"), status);
failure(status, "new DecimalFormat");
df->parse(UnicodeString("$"), num, pos2);
if (pos2 == ParsePosition(0) /*num != null*/){
errln(UnicodeString("Test Failed: \"$\" is parsed as ") /*+ num*/);
}
delete df;
}
/* @bug 4114639 (duplicate of 4106662)
* NumberFormat.parse doesn't return null
*/
void NumberFormatRegressionTest::Test4114639(void)
{
UErrorCode status = U_ZERO_ERROR;
NumberFormat *format = NumberFormat::createInstance(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete format;
return;
}
failure(status, "NumberFormat::createInstance");
UnicodeString text("time 10:x");
ParsePosition pos(8);
Formattable result;
format->parse(text, result, pos);
if (/*result != null*/pos.getErrorIndex() != 8)
errln(UnicodeString("Should return null but got : ") /*+ result*/); // Should be null; it isn't
delete format;
}
/* @bug 4106664
* TODO: this test does not work because we need to use a 64 bit number and a
* a double only MAY only have 52 bits of precision.
* DecimalFormat.format(long n) fails if n * multiplier > MAX_LONG.
*/
void NumberFormatRegressionTest::Test4106664(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
// {sfb} long in java is 64 bits
/*long*/double n = 1234567890123456.0;
/*int*/int32_t m = 12345678;
// {sfb} will this work?
//BigInteger bigN = BigInteger.valueOf(n);
//bigN = bigN.multiply(BigInteger.valueOf(m));
double bigN = n * m;
df->setMultiplier(m);
df->setGroupingUsed(FALSE);
UnicodeString temp;
FieldPosition pos(FieldPosition::DONT_CARE);
logln("formated: " +
df->format(n, temp, pos));
char buf [128];
sprintf(buf, "%g", bigN);
//logln("expected: " + bigN.toString());
logln(UnicodeString("expected: ") + buf);
delete df;
}
/* @bug 4106667 (duplicate of 4106658)
* DecimalFormat.format incorrectly formats -0.0.
*/
void NumberFormatRegressionTest::Test4106667(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete df;
return;
}
failure(status, "new DecimalFormat");
UChar foo [] = { 0x002B };
UnicodeString bar(foo, 1, 1);
volatile double d = 0.0; // volatile to prevent code optimization
UnicodeString temp;
UnicodeString buffer;
FieldPosition pos(FieldPosition::DONT_CARE);
logln("pattern: \"" + df->toPattern(temp) + "\"");
#if defined(U_HPUX)
d = 0.0 * -1.0; // old HPUX compiler ignores volatile keyword
#else
d *= -1.0; // Some compilers have a problem with defining -0.0
#endif
df->setPositivePrefix(/*"+"*/bar);
df->format(d, buffer, pos);
if (buffer != UnicodeString("-0")) // Corrected; see 4147706
errln(/*d + */UnicodeString(" is formatted as ") + buffer);
delete df;
}
/* @bug 4110936
* DecimalFormat.setMaximumIntegerDigits() works incorrectly.
*/
#ifdef OS390
# define MAX_INT_DIGITS 70
#else
# define MAX_INT_DIGITS 128
#endif
void NumberFormatRegressionTest::Test4110936(void)
{
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nf = NumberFormat::createInstance(status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete nf;
return;
}
failure(status, "NumberFormat::createInstance");
nf->setMaximumIntegerDigits(MAX_INT_DIGITS);
logln("setMaximumIntegerDigits(MAX_INT_DIGITS)");
if (nf->getMaximumIntegerDigits() != MAX_INT_DIGITS)
errln("getMaximumIntegerDigits() returns " +
nf->getMaximumIntegerDigits());
delete nf;
}
/* @bug 4122840
* Locale data should use generic currency symbol
*
* 1) Make sure that all currency formats use the generic currency symbol.
* 2) Make sure we get the same results using the generic symbol or a
* hard-coded one.
*/
void NumberFormatRegressionTest::Test4122840(void)
{
int32_t count = 0;
const Locale *locales = Locale::getAvailableLocales(count);
for (int i = 0; i < count; i++) {
UErrorCode status = U_ZERO_ERROR;
ResourceBundle *rb = new ResourceBundle(
NULL/*"java.text.resources.LocaleElements"*/,
locales[i], status);
failure(status, "new ResourceBundle");
ResourceBundle numPat = rb->get("NumberPatterns", status);
failure(status, "new ResourceBundle(NumberPatterns)");
//
// Get the currency pattern for this locale. We have to fish it
// out of the ResourceBundle directly, since DecimalFormat.toPattern
// will return the localized symbol, not \00a4
//
UnicodeString pattern = numPat.getStringEx(1, status);
failure(status, "rb->getStringArray");
UChar fo[] = { 0x00A4 };
UnicodeString foo(fo, 1, 1);
//if (pattern.indexOf("\u00A4") == -1 ) {
if (pattern.indexOf(foo) == -1 ) {
errln(UnicodeString("Currency format for ") + UnicodeString(locales[i].getName()) +
" does not contain generic currency symbol:" +
pattern );
}
// Create a DecimalFormat using the pattern we got and format a number
DecimalFormatSymbols *symbols = new DecimalFormatSymbols(locales[i], status);
failure(status, "new DecimalFormatSymbols");
DecimalFormat *fmt1 = new DecimalFormat(pattern, *symbols, status);
failure(status, "new DecimalFormat");
UnicodeString result1;
FieldPosition pos(FieldPosition::DONT_CARE);
result1 = fmt1->format(1.111, result1, pos);
//
// Now substitute in the locale's currency symbol and create another
// pattern. We have to skip locales where the currency symbol
// contains decimal separators, because that confuses things
//
UChar ba[] = { 0x002E/*'.'*/ };
UnicodeString bar(ba, 1, 1);
if (symbols->getSymbol(DecimalFormatSymbols::kCurrencySymbol).indexOf(bar) == -1) {
// {sfb} Also, switch the decimal separator to the monetary decimal
// separator to mimic the behavior of a currency format
symbols->setSymbol(DecimalFormatSymbols::kDecimalSeparatorSymbol,
symbols->getSymbol(DecimalFormatSymbols::kMonetarySeparatorSymbol));
UnicodeString buf(pattern);
for (int j = 0; j < buf.length(); j++) {
if (buf[j] == 0x00a4 ) {
if(buf[j + 1] == 0x00a4) {
// {sfb} added to support double currency marker (intl currency sign)
buf.replace(j, /*j+*/2, symbols->getSymbol(DecimalFormatSymbols::kIntlCurrencySymbol));
j += symbols->getSymbol(DecimalFormatSymbols::kIntlCurrencySymbol).length();
}
else {
buf.replace(j, /*j+*/1, symbols->getSymbol(DecimalFormatSymbols::kCurrencySymbol));
j += symbols->getSymbol(DecimalFormatSymbols::kCurrencySymbol).length() - 1;
}
}
}
DecimalFormat *fmt2 = new DecimalFormat(buf, *symbols, status);
failure(status, "new DecimalFormat");
// Get the currency (if there is one) so we can set the rounding and fraction
const UChar *currency = fmt1->getCurrency();
if (*currency != 0) {
double rounding = ucurr_getRoundingIncrement(currency, &status);
int32_t frac = ucurr_getDefaultFractionDigits(currency, &status);
if (U_SUCCESS(status)) {
fmt2->setRoundingIncrement(rounding);
fmt2->setMinimumFractionDigits(frac);
fmt2->setMaximumFractionDigits(frac);
}
else {
failure(status, "Fetching currency rounding/fractions");
}
}
UnicodeString result2;
fmt2->format(1.111, result2, pos);
if (result1 != result2) {
errln("Results for " + (UnicodeString)(locales[i].getName()) + " differ: " +
result1 + " vs " + result2);
}
delete fmt2;
}
delete rb;
delete fmt1;
delete symbols;
}
}
/* @bug 4125885
* DecimalFormat.format() delivers wrong string.
*/
void NumberFormatRegressionTest::Test4125885(void)
{
UErrorCode status = U_ZERO_ERROR;
double rate = 12.34;
DecimalFormat *formatDec = new DecimalFormat ("000.00", status);
if(U_FAILURE(status)) {
errln("Error creating DecimalFormat: %s", u_errorName(status));
delete formatDec;
return;
}
failure(status, "new DecimalFormat");
UnicodeString temp;
logln("toPattern: " + formatDec->toPattern(temp));
UnicodeString rateString;
FieldPosition pos(FieldPosition::DONT_CARE);
rateString = formatDec->format(rate, rateString, pos);
if (rateString != UnicodeString("012.34"))
errln("result : " + rateString + " expected : 012.34");
rate = 0.1234;
delete formatDec;// = null;
formatDec = new DecimalFormat ("+000.00%;-000.00%", status);
failure(status, "new DecimalFormat");
logln("toPattern: " + formatDec->toPattern(temp));
rateString.remove();
rateString = formatDec->format(rate, rateString, pos);
if (rateString != UnicodeString("+012.34%"))
errln("result : " + rateString + " expected : +012.34%");
delete formatDec;
}
/**
* @bug 4134034
* DecimalFormat produces extra zeros when formatting numbers.
*/
void NumberFormatRegressionTest::Test4134034(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *nf = new DecimalFormat("##,###,###.00", status);
failure(status, "new DecimalFormat");
UnicodeString f;
FieldPosition pos(FieldPosition::DONT_CARE);
f = nf->format(9.02, f, pos);
if (f == UnicodeString("9.02"))
logln(f + " ok");
else
errln("9.02 -> " + f + "; want 9.02");
f.remove();
f = nf->format((int32_t)0, f, pos);
if (f == UnicodeString(".00"))
logln(f + " ok");
else
errln("0 -> " + f + "; want .00");
delete nf;
}
/**
* @bug 4134300
* CANNOT REPRODUCE - This bug could not be reproduced. It may be
* a duplicate of 4134034.
*
* JDK 1.1.6 Bug, did NOT occur in 1.1.5
* Possibly related to bug 4125885.
*
* This class demonstrates a regression in version 1.1.6
* of DecimalFormat class.
*
* 1.1.6 Results
* Value 1.2 Format #.00 Result '01.20' !!!wrong
* Value 1.2 Format 0.00 Result '001.20' !!!wrong
* Value 1.2 Format 00.00 Result '0001.20' !!!wrong
* Value 1.2 Format #0.0# Result '1.2'
* Value 1.2 Format #0.00 Result '001.20' !!!wrong
*
* 1.1.5 Results
* Value 1.2 Format #.00 Result '1.20'
* Value 1.2 Format 0.00 Result '1.20'
* Value 1.2 Format 00.00 Result '01.20'
* Value 1.2 Format #0.0# Result '1.2'
* Value 1.2 Format #0.00 Result '1.20'
*/
void NumberFormatRegressionTest::Test4134300(void) {
UnicodeString DATA [] = {
// Pattern Expected string
UnicodeString("#.00"), UnicodeString("1.20"),
UnicodeString("0.00"), UnicodeString("1.20"),
UnicodeString("00.00"), UnicodeString("01.20"),
UnicodeString("#0.0#"), UnicodeString("1.2"),
UnicodeString("#0.00"), UnicodeString("1.20")
};
for (int i=0; i< 10; i+=2) {
UnicodeString result;
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat(DATA[i], status);
failure(status, "new DecimalFormat");
FieldPosition pos(FieldPosition::DONT_CARE);
result = df->format(1.2, result, pos);
if (result != DATA[i+1]) {
errln("Fail: 1.2 x " + DATA[i] + " = " + result +
"; want " + DATA[i+1]);
}
else {
logln("Ok: 1.2 x " + DATA[i] + " = " + result);
}
delete df;
}
}
/**
* @bug 4140009
* Empty pattern produces double negative prefix.
*/
void NumberFormatRegressionTest::Test4140009(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols *syms = new DecimalFormatSymbols(Locale::getEnglish(), status);
failure(status, "new DecimalFormatSymbols");
DecimalFormat *f = new DecimalFormat(UnicodeString(""), syms, status);
failure(status, "new DecimalFormat");
UnicodeString s;
FieldPosition pos(FieldPosition::DONT_CARE);
s = f->format(123.456, s, pos);
if (s != UnicodeString("123.456"))
errln("Fail: Format empty pattern x 123.456 => " + s);
s.remove();
s = f->format(-123.456, s, pos);
if (s != UnicodeString("-123.456"))
errln("Fail: Format empty pattern x -123.456 => " + s);
delete f;
}
/**
* @bug 4141750
* BigDecimal numbers get their fractions truncated by NumberFormat.
*/
// {sfb} not pertinent in C++ ??
void NumberFormatRegressionTest::Test4141750(void) {
/*try {
UnicodeString str("12345.67");
BigDecimal bd = new BigDecimal(str);
String sd = NumberFormat.getInstance(Locale.US).format(bd);
if (!sd.endsWith("67")) errln("Fail: " + str + " x format -> " + sd);
}
catch (Exception e) {
errln(e.toString());
e.printStackTrace();
}*/
}
/**
* @bug 4145457
* DecimalFormat toPattern() doesn't quote special characters or handle
* single quotes.
*/
void NumberFormatRegressionTest::Test4145457() {
//try {
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nff = NumberFormat::createInstance(status);
if (failure(status, "NumberFormat::createInstance")){
delete nff;
return;
};
if(nff->getDynamicClassID() != DecimalFormat::getStaticClassID()) {
errln("DecimalFormat needed to continue");
return;
}
DecimalFormat *nf = (DecimalFormat*)nff;
DecimalFormatSymbols *sym = (DecimalFormatSymbols*) nf->getDecimalFormatSymbols();
sym->setSymbol(DecimalFormatSymbols::kDecimalSeparatorSymbol, (UChar)/*'\''*/0x0027);
nf->setDecimalFormatSymbols(*sym);
double pi = 3.14159;
UnicodeString PATS [] = {
UnicodeString("#.00 'num''ber'"), UnicodeString("''#.00''")
};
for (int32_t i=0; i<2; ++i) {
nf->applyPattern(PATS[i], status);
failure(status, "nf->applyPattern");
UnicodeString out;
FieldPosition pos(FieldPosition::DONT_CARE);
out = nf->format(pi, out, pos);
UnicodeString pat;
pat = nf->toPattern(pat);
Formattable num;
ParsePosition pp(0);
nf->parse(out, num, pp);
double val = num.getDouble();
nf->applyPattern(pat, status);
failure(status, "nf->applyPattern");
UnicodeString out2;
out2 = nf->format(pi, out2, pos);
UnicodeString pat2;
pat2 = nf->toPattern(pat2);
nf->parse(out2, num, pp);
double val2 = num.getDouble();
if (pat != pat2)
errln("Fail with \"" + PATS[i] + "\": Patterns should concur, \"" +
pat + "\" vs. \"" + pat2 + "\"");
else
logln("Ok \"" + PATS[i] + "\" toPattern() -> \"" + pat + '"');
if (val == val2 && out == out2) {
logln(UnicodeString("Ok ") + pi + " x \"" + PATS[i] + "\" -> \"" +
out + "\" -> " + val + " -> \"" +
out2 + "\" -> " + val2);
}
else {
errln(UnicodeString("Fail ") + pi + " x \"" + PATS[i] + "\" -> \"" +
out + "\" -> " + val + " -> \"" +
out2 + "\" -> " + val2);
}
}
/*}
catch (ParseException e) {
errln("Fail: " + e);
e.printStackTrace();
}*/
delete nff;
}
/**
* @bug 4147295
* DecimalFormat.applyPattern() sets minimum integer digits incorrectly.
* CANNOT REPRODUCE
* This bug is a duplicate of 4139344, which is a duplicate of 4134300
*/
void NumberFormatRegressionTest::Test4147295(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *sdf = new DecimalFormat(status);
UnicodeString pattern("#,###");
logln("Applying pattern \"" + pattern + "\"");
sdf->applyPattern(pattern, status);
failure(status, "sdf->applyPattern");
int minIntDig = sdf->getMinimumIntegerDigits();
if (minIntDig != 0) {
errln("Test failed");
errln(" Minimum integer digits : " + minIntDig);
UnicodeString temp;
errln(" new pattern: " + sdf->toPattern(temp));
} else {
logln("Test passed");
logln(" Minimum integer digits : " + minIntDig);
}
delete sdf;
}
/**
* @bug 4147706
* DecimalFormat formats -0.0 as +0.0
* See also older related bug 4106658, 4106667
*/
void NumberFormatRegressionTest::Test4147706(void)
{
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *df = new DecimalFormat("#,##0.0##", status);
failure(status, "new DecimalFormat");
DecimalFormatSymbols *syms = new DecimalFormatSymbols(Locale::getEnglish(), status);
failure(status, "new DecimalFormatSymbols");
UnicodeString f1;
UnicodeString f2, temp;
FieldPosition pos(FieldPosition::DONT_CARE);
volatile double d1 = 0.0; // volatile to prevent code optimization
double d2 = -0.0001;
#if defined(U_HPUX)
d1 = 0.0 * -1.0; // old HPUX compiler ignores volatile keyword
#else
d1 *= -1.0; // Some compilers have a problem with defining -0.0
#endif
df->adoptDecimalFormatSymbols(syms);
f1 = df->format(d1, f1, pos);
f2 = df->format(d2, f2, pos);
if (f1 != UnicodeString("-0.0")) {
errln(UnicodeString("") + d1 + UnicodeString(" x \"") + df->toPattern(temp) + "\" is formatted as \"" + f1 + '"');
}
if (f2 != UnicodeString("-0.0")) {
errln(UnicodeString("") + d2 + UnicodeString(" x \"") + df->toPattern(temp) + "\" is formatted as \"" + f2 + '"');
}
delete df;
}
// Not applicable, since no serialization in C++
/*class myformat implements Serializable
{
DateFormat _dateFormat = DateFormat.getDateInstance();
public String Now()
{
GregorianCalendar calendar = new GregorianCalendar();
Date t = calendar.getTime();
String nowStr = _dateFormat.format(t);
return nowStr;
}
}*/
/**
* @bug 4162198
* NumberFormat cannot format Double.MAX_VALUE
*/
// TODO: make this test actually test something
void
NumberFormatRegressionTest::Test4162198(void)
{
// for some reason, DBL_MAX will not round trip. (bug in sprintf/atof)
double dbl = INT32_MAX * 1000.0;
UErrorCode status = U_ZERO_ERROR;
NumberFormat *f = NumberFormat::createInstance(status);
if(U_FAILURE(status)) {
errln("Couldn't create number format");
return;
}
f->setMaximumFractionDigits(INT32_MAX);
f->setMaximumIntegerDigits(INT32_MAX);
UnicodeString s;
f->format(dbl,s);
logln(UnicodeString("The number ") + dbl + " formatted to " + s);
Formattable n;
//try {
f->parse(s, n, status);
if(U_FAILURE(status))
errln("Couldn't parse!");
//} catch (java.text.ParseException e) {
// errln("Caught a ParseException:");
// e.printStackTrace();
//}
//logln("The string " + s + " parsed as " + n);
// {dlf} The old code assumes n is a double, but it isn't any more...
// Formattable apparently does not and never did interconvert... too bad.
//if(n.getDouble() != dbl) {
// errln("Round trip failure");
//}
if (n.getInt64() != dbl) {
errln("Round trip failure");
}
delete f;
}
/**
* @bug 4162852
* NumberFormat does not parse negative zero.
*/
void
NumberFormatRegressionTest::Test4162852(void)
{
UErrorCode status = U_ZERO_ERROR;
for(int32_t i=0; i < 2; ++i) {
NumberFormat *f = (i == 0) ? NumberFormat::createInstance(status)
: NumberFormat::createPercentInstance(status);
if(U_FAILURE(status)) {
errln("Couldn't create number format");
return;
}
double d = 0.0;
d *= -1.0;
UnicodeString s;
f->format(d, s);
Formattable n;
f->parse(s, n, status);
if(U_FAILURE(status))
errln("Couldn't parse!");
double e = n.getDouble();
logln(UnicodeString("") +
d + " -> " +
'"' + s + '"' + " -> " + e);
#if (defined(OS390) && !defined(IEEE_754)) || defined(OS400)
if (e != 0.0) {
#else
if (e != 0.0 || 1.0/e > 0.0) {
#endif
logln("Failed to parse negative zero");
}
delete f;
}
}
static double _u_abs(double a) { return a<0?-a:a; }
/**
* May 17 1999 sync up - liu
* @bug 4167494
* NumberFormat truncates data
*/
void NumberFormatRegressionTest::Test4167494(void) {
UErrorCode status = U_ZERO_ERROR;
NumberFormat *fmt = NumberFormat::createInstance(Locale::getUS(), status);
if (failure(status, "NumberFormat::createInstance")){
delete fmt;
return;
};
double a = DBL_MAX * 0.99; // DBL_MAX itself overflows to +Inf
UnicodeString s;
fmt->format(a, s);
Formattable num;
fmt->parse(s, num, status);
failure(status, "Parse");
if (num.getType() == Formattable::kDouble &&
_u_abs(num.getDouble() - a) / a < 0.01) { // RT within 1%
logln(UnicodeString("") + a + " -> \"" + s + "\" -> " +
toString(num) + " ok");
} else {
errln(UnicodeString("") + a + " -> \"" + s + "\" -> " +
toString(num) + " FAIL");
}
// We don't test Double.MIN_VALUE because the locale data for the US
// currently doesn't specify enough digits to display Double.MIN_VALUE.
// This is correct for now; however, we leave this here as a reminder
// in case we want to address this later.
delete fmt;
}
/**
* May 17 1999 sync up - liu
* @bug 4170798
* DecimalFormat.parse() fails when ParseIntegerOnly set to true
*/
void NumberFormatRegressionTest::Test4170798(void) {
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nf = NumberFormat::createInstance(Locale::getUS(), status);
if (failure(status, "NumberFormat::createInstance")){
delete nf;
return;
};
if(nf->getDynamicClassID() != DecimalFormat::getStaticClassID()) {
errln("DecimalFormat needed to continue");
return;
}
DecimalFormat *df = (DecimalFormat*) nf;
df->setParseIntegerOnly(TRUE);
Formattable n;
ParsePosition pos(0);
df->parse("-0.0", n, pos);
if (n.getType() != Formattable::kLong
|| n.getLong() != 0) {
errln(UnicodeString("FAIL: parse(\"-0.0\") returns ") + toString(n));
}
delete nf;
}
/**
* May 17 1999 sync up - liu
* toPattern only puts the first grouping separator in.
*/
void NumberFormatRegressionTest::Test4176114(void) {
const char* DATA[] = {
"00", "#00",
"000", "#000", // No grouping
"#000", "#000", // No grouping
"#,##0", "#,##0",
"#,000", "#,000",
"0,000", "#0,000",
"00,000", "#00,000",
"000,000", "#,000,000",
"0,000,000,000,000.0000", "#0,000,000,000,000.0000", // Reported
};
int DATA_length = (int)(sizeof(DATA) / sizeof(DATA[0]));
UErrorCode status = U_ZERO_ERROR;
UnicodeString s;
for (int i=0; i<DATA_length; i+=2) {
DecimalFormat df(DATA[i], status);
failure(status, "DecimalFormat constructor");
df.toPattern(s);
UnicodeString exp(DATA[i+1]);
if (s != exp) {
errln(UnicodeString("FAIL: ") + DATA[i] + " -> " +
s + ", want " + exp);
}
}
}
/**
* May 17 1999 sync up - liu
* @bug 4179818
* DecimalFormat is incorrectly rounding numbers like 1.2501 to 1.2
*/
void NumberFormatRegressionTest::Test4179818(void) {
const char* DATA[] = {
// Input Pattern Expected output
"1.2511", "#.#", "1.3",
"1.2501", "#.#", "1.3",
"0.9999", "#", "1",
};
int DATA_length = (int)(sizeof(DATA) / sizeof(DATA[0]));
double DOUBLE[] = {
1.2511,
1.2501,
0.9999,
};
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols sym(Locale::getUS(), status);
failure(status, "Construct DecimalFormatSymbols");
DecimalFormat fmt("#", sym, status);
failure(status, "Construct DecimalFormat");
for (int i=0; i<DATA_length; i+=3) {
double in = DOUBLE[i/3];
UnicodeString pat(DATA[i+1]);
UnicodeString exp(DATA[i+2]);
fmt.applyPattern(pat, status);
failure(status, "applyPattern");
UnicodeString out;
FieldPosition pos;
fmt.format(in, out, pos);
if (out == exp) {
logln(UnicodeString("Ok: ") + in + " x " + pat + " = " + out);
} else {
errln(UnicodeString("FAIL: ") + in + " x " + pat + " = " + out +
", expected " + exp);
}
}
}
/**
* May 17 1999 sync up - liu
* Some DecimalFormatSymbols changes are not picked up by DecimalFormat.
* This includes the minus sign, currency symbol, international currency
* symbol, percent, and permille. This is filed as bugs 4212072 and
* 4212073.
*/
void NumberFormatRegressionTest::Test4212072(void) {
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols sym(Locale::getUS(), status);
failure(status, "DecimalFormatSymbols ct", Locale::getUS());
DecimalFormat fmt(UnicodeString("#"), sym, status);
failure(status, "DecimalFormat ct", Locale::getUS());
UnicodeString s;
FieldPosition pos;
sym.setSymbol(DecimalFormatSymbols::kMinusSignSymbol, (UChar)0x5e);
fmt.setDecimalFormatSymbols(sym);
s.remove();
if (fmt.format((int32_t)-1, s, pos) != UNICODE_STRING("^1", 2)) {
errln(UnicodeString("FAIL: -1 x (minus=^) -> ") + s +
", exp ^1");
}
s.remove();
if (fmt.getNegativePrefix(s) != UnicodeString((UChar)0x5e)) {
errln(UnicodeString("FAIL: (minus=^).getNegativePrefix -> ") +
s + ", exp ^");
}
sym.setSymbol(DecimalFormatSymbols::kMinusSignSymbol, (UChar)0x2d);
fmt.applyPattern(UnicodeString("#%"), status);
failure(status, "applyPattern percent");
sym.setSymbol(DecimalFormatSymbols::kPercentSymbol, (UChar)0x5e);
fmt.setDecimalFormatSymbols(sym);
s.remove();
if (fmt.format(0.25, s, pos) != UNICODE_STRING("25^", 3)) {
errln(UnicodeString("FAIL: 0.25 x (percent=^) -> ") + s +
", exp 25^");
}
s.remove();
if (fmt.getPositiveSuffix(s) != UnicodeString((UChar)0x5e)) {
errln(UnicodeString("FAIL: (percent=^).getPositiveSuffix -> ") +
s + ", exp ^");
}
sym.setSymbol(DecimalFormatSymbols::kPercentSymbol, (UChar)0x25);
fmt.applyPattern(str("#\\u2030"), status);
failure(status, "applyPattern permill");
sym.setSymbol(DecimalFormatSymbols::kPerMillSymbol, (UChar)0x5e);
fmt.setDecimalFormatSymbols(sym);
s.remove();
if (fmt.format(0.25, s, pos) != UNICODE_STRING("250^", 4)) {
errln(UnicodeString("FAIL: 0.25 x (permill=^) -> ") + s +
", exp 250^");
}
s.remove();
if (fmt.getPositiveSuffix(s) != UnicodeString((UChar)0x5e)) {
errln(UnicodeString("FAIL: (permill=^).getPositiveSuffix -> ") +
s + ", exp ^");
}
sym.setSymbol(DecimalFormatSymbols::kPerMillSymbol, (UChar)0x2030);
fmt.applyPattern(str("\\u00A4#.00"), status);
failure(status, "applyPattern currency");
sym.setSymbol(DecimalFormatSymbols::kCurrencySymbol, "usd");
fmt.setDecimalFormatSymbols(sym);
s.remove();
if (fmt.format(12.5, s, pos) != UnicodeString("usd12.50")) {
errln(UnicodeString("FAIL: 12.5 x (currency=usd) -> ") + s +
", exp usd12.50");
}
s.remove();
if (fmt.getPositivePrefix(s) != UnicodeString("usd")) {
errln(UnicodeString("FAIL: (currency=usd).getPositivePrefix -> ") +
s + ", exp usd");
}
sym.setSymbol(DecimalFormatSymbols::kCurrencySymbol, "$");
fmt.applyPattern(str("\\u00A4\\u00A4#.00"), status);
failure(status, "applyPattern intl currency");
sym.setSymbol(DecimalFormatSymbols::kIntlCurrencySymbol, "DOL");
fmt.setDecimalFormatSymbols(sym);
s.remove();
if (fmt.format(12.5, s, pos) != UnicodeString("DOL12.50")) {
errln(UnicodeString("FAIL: 12.5 x (intlcurrency=DOL) -> ") + s +
", exp DOL12.50");
}
s.remove();
if (fmt.getPositivePrefix(s) != UnicodeString("DOL")) {
errln(UnicodeString("FAIL: (intlcurrency=DOL).getPositivePrefix -> ") +
s + ", exp DOL");
}
sym.setSymbol(DecimalFormatSymbols::kIntlCurrencySymbol, "USD");
// Since the pattern logic has changed, make sure that patterns round
// trip properly. Test stream in/out integrity too.
int32_t n;
const Locale* avail = NumberFormat::getAvailableLocales(n);
static const char* type[] = {
"",
"$ ",
"% ",
};
for (int i=0; i<n; ++i) {
for (int j=0; j<3; ++j) {
status = U_ZERO_ERROR;
NumberFormat *nf;
switch (j) {
case 0:
nf = NumberFormat::createInstance(avail[i], status);
failure(status, "createInstance", avail[i]);
break;
case 1:
nf = NumberFormat::createCurrencyInstance(avail[i], status);
failure(status, "createCurrencyInstance", avail[i]);
break;
default:
nf = NumberFormat::createPercentInstance(avail[i], status);
failure(status, "createPercentInstance", avail[i]);
break;
}
if (U_FAILURE(status)) {
continue;
}
DecimalFormat *df = (DecimalFormat*) nf;
// Test toPattern/applyPattern round trip
UnicodeString pat;
df->toPattern(pat);
DecimalFormatSymbols symb(avail[i], status);
failure(status, "Construct DecimalFormatSymbols", avail[i]);
DecimalFormat f2(pat, symb, status);
if (failure(status,
UnicodeString("Construct DecimalFormat(") + pat + ")")) {
continue;
}
if (*df != f2) {
UnicodeString l, p;
errln(UnicodeString("FAIL: ") + type[j] + avail[i].getDisplayName(l) +
" -> \"" + pat +
"\" -> \"" + f2.toPattern(p) + "\"");
}
// Test toLocalizedPattern/applyLocalizedPattern round trip
df->toLocalizedPattern(pat);
f2.applyLocalizedPattern(pat, status);
failure(status,
UnicodeString("applyLocalizedPattern(") + pat + ")", avail[i]);
if (U_FAILURE(status)) {
continue;
}
// Make sure we set the currency attributes appropriately
if (j == 1) { // Currency format
f2.setCurrency(f2.getCurrency(), status);
}
failure(status,
UnicodeString("setCurrency() for (") + pat + ")", avail[i]);
if (U_FAILURE(status)) {
continue;
}
if (*df != f2) {
UnicodeString l, p;
errln(UnicodeString("FAIL: ") + type[j] + avail[i].getDisplayName(l) +
" -> localized \"" + pat +
"\" -> \"" + f2.toPattern(p) + "\"");
}
delete nf;
// Test writeObject/readObject round trip
// NOT ON ICU -- Java only
}
}
}
/**
* May 17 1999 sync up - liu
* DecimalFormat.parse() fails for mulipliers 2^n.
*/
void NumberFormatRegressionTest::Test4216742(void) {
UErrorCode status = U_ZERO_ERROR;
DecimalFormat *fmt = (DecimalFormat*) NumberFormat::createInstance(Locale::getUS(), status);
if (failure(status, "createInstance", Locale::getUS())){
delete fmt;
return;
};
int32_t DATA[] = { INT32_MIN, INT32_MAX, -100000000, 100000000 };
int DATA_length = (int)(sizeof(DATA) / sizeof(DATA[0]));
for (int i=0; i<DATA_length; ++i) {
UnicodeString str((UnicodeString)"" + DATA[i]);
for (int m = 1; m <= 100; m++) {
fmt->setMultiplier(m);
Formattable num;
fmt->parse(str, num, status);
failure(status, "parse", Locale::getUS());
if (num.getType() != Formattable::kLong &&
num.getType() != Formattable::kDouble) {
errln(UnicodeString("FAIL: Wanted number, got ") +
toString(num));
} else {
double d = num.getType() == Formattable::kDouble ?
num.getDouble() : (double) num.getLong();
if (d > 0 != DATA[i] > 0) {
errln(UnicodeString("\"") + str + "\" parse(x " +
fmt->getMultiplier() +
") => " + toString(num));
}
}
}
}
delete fmt;
}
/**
* May 17 1999 sync up - liu
* DecimalFormat formats 1.001 to "1.00" instead of "1" with 2 fraction
* digits.
*/
void NumberFormatRegressionTest::Test4217661(void) {
const double D[] = { 0.001, 1.001, 0.006, 1.006 };
const char* S[] = { "0", "1", "0.01", "1.01" };
int D_length = (int)(sizeof(D) / sizeof(D[0]));
UErrorCode status = U_ZERO_ERROR;
NumberFormat *fmt = NumberFormat::createInstance(Locale::getUS(), status);
if (failure(status, "createInstance", Locale::getUS())){
delete fmt;
return;
};
fmt->setMaximumFractionDigits(2);
for (int i=0; i<D_length; i++) {
UnicodeString s;
fmt->format(D[i], s);
if (s != UnicodeString(S[i])) {
errln(UnicodeString("FAIL: Got ") + s + ", exp " + S[i]);
}
}
delete fmt;
}
/**
* alphaWorks upgrade
*/
void NumberFormatRegressionTest::Test4161100(void) {
UErrorCode status = U_ZERO_ERROR;
NumberFormat *nf = NumberFormat::createInstance(Locale::getUS(), status);
if (failure(status, "createInstance", Locale::getUS())){
delete nf;
return;
};
nf->setMinimumFractionDigits(1);
nf->setMaximumFractionDigits(1);
double a = -0.09;
UnicodeString s;
nf->format(a, s);
UnicodeString pat;
logln(UnicodeString() + a + " x " +
((DecimalFormat*) nf)->toPattern(pat) + " = " + s);
if (s != UnicodeString("-0.1")) {
errln("FAIL");
}
delete nf;
}
/**
* June 16 1999 sync up - liu
* Formatting .5 rounds to "1" instead of "0". (Regression in 1.2.2 RC1)
*/
void NumberFormatRegressionTest::Test4243011(void) {
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols sym(Locale::getUS(), status);
failure(status, "DecimalFormatSymbols ct", Locale::getUS());
DecimalFormat fmt(UnicodeString("0."), sym, status);
failure(status, "DecimalFormat ct", Locale::getUS());
const double NUM[] = { -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5, 4.5 };
const char* STR[] = { "-2.", "-2.", "-0.", "0.", "2.", "2.", "4.", "4." };
int32_t N = (int32_t)(sizeof(NUM) / sizeof(NUM[0]));
for (int32_t i=0; i<N; ++i) {
UnicodeString str;
UnicodeString exp(STR[i]);
FieldPosition pos;
fmt.format(NUM[i], str, pos);
if (str == exp) {
logln(UnicodeString("Ok ") + NUM[i] + " x 0. = " + str);
} else {
errln(UnicodeString("FAIL ") + NUM[i] + " x 0. = " + str +
", want " + exp);
}
}
}
/**
* June 16 1999 sync up - liu
* format(0.0) gives "0.1" if preceded by parse("99.99").
* (Regression in 1.2.2 RC1)
*/
void NumberFormatRegressionTest::Test4243108(void) {
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols sym(Locale::getUS(), status);
failure(status, "DecimalFormatSymbols ct", Locale::getUS());
DecimalFormat fmt(UnicodeString("#.#"), sym, status);
failure(status, "DecimalFormat ct", Locale::getUS());
UnicodeString str;
FieldPosition pos;
fmt.format(0.0, str, pos);
UnicodeString exp("0");
if (str == exp) {
logln(UnicodeString("Ok 0.0 x #.# = ") + str);
} else {
errln(UnicodeString("FAIL 0.0 x #.# = ") + str +
", want " + exp);
}
str = "99.99";
Formattable val;
fmt.parse(str, val, status);
failure(status, "DecimalFormat.parse(99.99)", Locale::getUS());
if (val.getType() == Formattable::kDouble &&
val.getDouble() == 99.99) {
logln(UnicodeString("Ok 99.99 / #.# = ") + toString(val));
} else {
errln(UnicodeString("FAIL 99.99 / #.# = ") + toString(val) +
", want " + 99.99);
}
str.remove();
fmt.format(0.0, str, pos);
if (str == exp) {
logln(UnicodeString("Ok 0.0 x #.# = ") + str);
} else {
errln(UnicodeString("FAIL 0.0 x #.# = ") + str +
", want " + exp);
}
}
/**
* DateFormat should call setIntegerParseOnly(TRUE) on adopted
* NumberFormat objects.
*/
void NumberFormatRegressionTest::TestJ691(void) {
UErrorCode status = U_ZERO_ERROR;
Locale loc("fr", "CH");
// set up the input date string & expected output
UnicodeString udt("11.10.2000", "");
UnicodeString exp("11.10.00", "");
// create a Calendar for this locale
Calendar *cal = Calendar::createInstance(loc, status);
if (U_FAILURE(status)) {
errln("FAIL: Calendar::createInstance() returned " + (UnicodeString)u_errorName(status));
return;
}
// create a NumberFormat for this locale
NumberFormat *nf = NumberFormat::createInstance(loc, status);
if (U_FAILURE(status)) {
errln("FAIL: NumberFormat::createInstance() returned " + (UnicodeString)u_errorName(status));
return;
}
// *** Here's the key: We don't want to have to do THIS:
// nf->setParseIntegerOnly(TRUE);
// create the DateFormat
DateFormat *df = DateFormat::createDateInstance(DateFormat::kShort, loc);
if (U_FAILURE(status)) {
errln("FAIL: DateFormat::createInstance() returned " + (UnicodeString)u_errorName(status));
return;
}
df->adoptCalendar(cal);
df->adoptNumberFormat(nf);
// set parsing to lenient & parse
df->setLenient(TRUE);
UDate ulocdat = df->parse(udt, status);
// format back to a string
UnicodeString outString;
df->format(ulocdat, outString);
if (outString != exp) {
errln("FAIL: " + udt + " => " + outString);
}
delete df;
}
#endif /* #if !UCONFIG_NO_FORMATTING */