/*
*******************************************************************************
* Copyright (C) 1996-2010, International Business Machines Corporation and *
* others. All Rights Reserved. *
*******************************************************************************
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_FORMATTING
#include "itrbnf.h"
#include "unicode/umachine.h"
#include "unicode/tblcoll.h"
#include "unicode/coleitr.h"
#include "unicode/ures.h"
#include "unicode/ustring.h"
#include "unicode/decimfmt.h"
#include "unicode/udata.h"
#include "testutil.h"
//#include "llong.h"
#include <string.h>
// import com.ibm.text.RuleBasedNumberFormat;
// import com.ibm.test.TestFmwk;
// import java.util.Locale;
// import java.text.NumberFormat;
// current macro not in icu1.8.1
#define TESTCASE(id,test) \
case id: \
name = #test; \
if (exec) { \
logln(#test "---"); \
logln(); \
test(); \
} \
break
void IntlTestRBNF::runIndexedTest(int32_t index, UBool exec, const char* &name, char* /*par*/)
{
if (exec) logln("TestSuite RuleBasedNumberFormat");
switch (index) {
#if U_HAVE_RBNF
TESTCASE(0, TestEnglishSpellout);
TESTCASE(1, TestOrdinalAbbreviations);
TESTCASE(2, TestDurations);
TESTCASE(3, TestSpanishSpellout);
TESTCASE(4, TestFrenchSpellout);
TESTCASE(5, TestSwissFrenchSpellout);
TESTCASE(6, TestItalianSpellout);
TESTCASE(7, TestGermanSpellout);
TESTCASE(8, TestThaiSpellout);
TESTCASE(9, TestAPI);
TESTCASE(10, TestFractionalRuleSet);
TESTCASE(11, TestSwedishSpellout);
TESTCASE(12, TestBelgianFrenchSpellout);
TESTCASE(13, TestSmallValues);
TESTCASE(14, TestLocalizations);
TESTCASE(15, TestAllLocales);
TESTCASE(16, TestHebrewFraction);
TESTCASE(17, TestPortugueseSpellout);
TESTCASE(18, TestMultiplierSubstitution);
#else
TESTCASE(0, TestRBNFDisabled);
#endif
default:
name = "";
break;
}
}
#if U_HAVE_RBNF
void IntlTestRBNF::TestHebrewFraction() {
// this is the expected output for 123.45, with no '<' in it.
UChar text1[] = {
0x05de, 0x05d0, 0x05d4, 0x0020,
0x05e2, 0x05e9, 0x05e8, 0x05d9, 0x05dd, 0x0020,
0x05d5, 0x05e9, 0x05dc, 0x05d5, 0x05e9, 0x0020,
0x05e0, 0x05e7, 0x05d5, 0x05d3, 0x05d4, 0x0020,
0x05d0, 0x05e8, 0x05d1, 0x05e2, 0x0020,
0x05d7, 0x05de, 0x05e9, 0x0000,
};
UChar text2[] = {
0x05DE, 0x05D0, 0x05D4, 0x0020,
0x05E2, 0x05E9, 0x05E8, 0x05D9, 0x05DD, 0x0020,
0x05D5, 0x05E9, 0x05DC, 0x05D5, 0x05E9, 0x0020,
0x05E0, 0x05E7, 0x05D5, 0x05D3, 0x05D4, 0x0020,
0x05D0, 0x05E4, 0x05E1, 0x0020,
0x05D0, 0x05E4, 0x05E1, 0x0020,
0x05D0, 0x05E8, 0x05D1, 0x05E2, 0x0020,
0x05D7, 0x05DE, 0x05E9, 0x0000,
};
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter = new RuleBasedNumberFormat(URBNF_SPELLOUT, "he_IL", status);
if (status == U_MISSING_RESOURCE_ERROR || status == U_FILE_ACCESS_ERROR) {
errcheckln(status, "Failed in constructing RuleBasedNumberFormat - %s", u_errorName(status));
delete formatter;
return;
}
UnicodeString result;
Formattable parseResult;
ParsePosition pp(0);
{
UnicodeString expected(text1);
formatter->format(123.45, result);
if (result != expected) {
errln((UnicodeString)"expected '" + TestUtility::hex(expected) + "'\nbut got: '" + TestUtility::hex(result) + "'");
} else {
// formatter->parse(result, parseResult, pp);
// if (parseResult.getDouble() != 123.45) {
// errln("expected 123.45 but got: %g", parseResult.getDouble());
// }
}
}
{
UnicodeString expected(text2);
result.remove();
formatter->format(123.0045, result);
if (result != expected) {
errln((UnicodeString)"expected '" + TestUtility::hex(expected) + "'\nbut got: '" + TestUtility::hex(result) + "'");
} else {
pp.setIndex(0);
// formatter->parse(result, parseResult, pp);
// if (parseResult.getDouble() != 123.0045) {
// errln("expected 123.0045 but got: %g", parseResult.getDouble());
// }
}
}
delete formatter;
}
void
IntlTestRBNF::TestAPI() {
// This test goes through the APIs that were not tested before.
// These tests are too small to have separate test classes/functions
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale::getUS(), status);
if (status == U_MISSING_RESOURCE_ERROR || status == U_FILE_ACCESS_ERROR) {
dataerrln("Unable to create formatter. - %s", u_errorName(status));
delete formatter;
return;
}
logln("RBNF API test starting");
// test clone
{
logln("Testing Clone");
RuleBasedNumberFormat* rbnfClone = (RuleBasedNumberFormat *)formatter->clone();
if(rbnfClone != NULL) {
if(!(*rbnfClone == *formatter)) {
errln("Clone should be semantically equivalent to the original!");
}
delete rbnfClone;
} else {
errln("Cloning failed!");
}
}
// test assignment
{
logln("Testing assignment operator");
RuleBasedNumberFormat assignResult(URBNF_SPELLOUT, Locale("es", "ES", ""), status);
assignResult = *formatter;
if(!(assignResult == *formatter)) {
errln("Assignment result should be semantically equivalent to the original!");
}
}
// test rule constructor
{
logln("Testing rule constructor");
LocalUResourceBundlePointer en(ures_open(U_ICUDATA_NAME U_TREE_SEPARATOR_STRING "rbnf", "en", &status));
if(U_FAILURE(status)) {
errln("Unable to access resource bundle with data!");
} else {
int32_t ruleLen = 0;
int32_t len = 0;
LocalUResourceBundlePointer rbnfRules(ures_getByKey(en.getAlias(), "RBNFRules", NULL, &status));
LocalUResourceBundlePointer ruleSets(ures_getByKey(rbnfRules.getAlias(), "SpelloutRules", NULL, &status));
UnicodeString desc;
while (ures_hasNext(ruleSets.getAlias())) {
const UChar* currentString = ures_getNextString(ruleSets.getAlias(), &len, NULL, &status);
ruleLen += len;
desc.append(currentString);
}
const UChar *spelloutRules = desc.getTerminatedBuffer();
if(U_FAILURE(status) || ruleLen == 0 || spelloutRules == NULL) {
errln("Unable to access the rules string!");
} else {
UParseError perror;
RuleBasedNumberFormat ruleCtorResult(spelloutRules, Locale::getUS(), perror, status);
if(!(ruleCtorResult == *formatter)) {
errln("Formatter constructed from the original rules should be semantically equivalent to the original!");
}
// Jitterbug 4452, for coverage
RuleBasedNumberFormat nf(spelloutRules, (UnicodeString)"", Locale::getUS(), perror, status);
if(!(nf == *formatter)) {
errln("Formatter constructed from the original rules should be semantically equivalent to the original!");
}
}
}
}
// test getRules
{
logln("Testing getRules function");
UnicodeString rules = formatter->getRules();
UParseError perror;
RuleBasedNumberFormat fromRulesResult(rules, Locale::getUS(), perror, status);
if(!(fromRulesResult == *formatter)) {
errln("Formatter constructed from rules obtained by getRules should be semantically equivalent to the original!");
}
}
{
logln("Testing copy constructor");
RuleBasedNumberFormat copyCtorResult(*formatter);
if(!(copyCtorResult == *formatter)) {
errln("Copy constructor result result should be semantically equivalent to the original!");
}
}
#if !UCONFIG_NO_COLLATION
// test ruleset names
{
logln("Testing getNumberOfRuleSetNames, getRuleSetName and format using rule set names");
int32_t noOfRuleSetNames = formatter->getNumberOfRuleSetNames();
if(noOfRuleSetNames == 0) {
errln("Number of rule set names should be more than zero");
}
UnicodeString ruleSetName;
int32_t i = 0;
int32_t intFormatNum = 34567;
double doubleFormatNum = 893411.234;
logln("number of rule set names is %i", noOfRuleSetNames);
for(i = 0; i < noOfRuleSetNames; i++) {
FieldPosition pos1, pos2;
UnicodeString intFormatResult, doubleFormatResult;
Formattable intParseResult, doubleParseResult;
ruleSetName = formatter->getRuleSetName(i);
log("Rule set name %i is ", i);
log(ruleSetName);
logln(". Format results are: ");
intFormatResult = formatter->format(intFormatNum, ruleSetName, intFormatResult, pos1, status);
doubleFormatResult = formatter->format(doubleFormatNum, ruleSetName, doubleFormatResult, pos2, status);
if(U_FAILURE(status)) {
errln("Format using a rule set failed");
break;
}
logln(intFormatResult);
logln(doubleFormatResult);
formatter->setLenient(TRUE);
formatter->parse(intFormatResult, intParseResult, status);
formatter->parse(doubleFormatResult, doubleParseResult, status);
logln("Parse results for lenient = TRUE, %i, %f", intParseResult.getLong(), doubleParseResult.getDouble());
formatter->setLenient(FALSE);
formatter->parse(intFormatResult, intParseResult, status);
formatter->parse(doubleFormatResult, doubleParseResult, status);
logln("Parse results for lenient = FALSE, %i, %f", intParseResult.getLong(), doubleParseResult.getDouble());
if(U_FAILURE(status)) {
errln("Error during parsing");
}
intFormatResult = formatter->format(intFormatNum, "BLABLA", intFormatResult, pos1, status);
if(U_SUCCESS(status)) {
errln("Using invalid rule set name should have failed");
break;
}
status = U_ZERO_ERROR;
doubleFormatResult = formatter->format(doubleFormatNum, "TRUC", doubleFormatResult, pos2, status);
if(U_SUCCESS(status)) {
errln("Using invalid rule set name should have failed");
break;
}
status = U_ZERO_ERROR;
}
status = U_ZERO_ERROR;
}
#endif
// test API
UnicodeString expected("four point five","");
logln("Testing format(double)");
UnicodeString result;
formatter->format(4.5,result);
if(result != expected) {
errln("Formatted 4.5, expected " + expected + " got " + result);
} else {
logln("Formatted 4.5, expected " + expected + " got " + result);
}
result.remove();
expected = "four";
formatter->format((int32_t)4,result);
if(result != expected) {
errln("Formatted 4, expected " + expected + " got " + result);
} else {
logln("Formatted 4, expected " + expected + " got " + result);
}
result.remove();
FieldPosition pos;
formatter->format((int64_t)4, result, pos, status = U_ZERO_ERROR);
if(result != expected) {
errln("Formatted 4 int64_t, expected " + expected + " got " + result);
} else {
logln("Formatted 4 int64_t, expected " + expected + " got " + result);
}
//Jitterbug 4452, for coverage
result.remove();
FieldPosition pos2;
formatter->format((int64_t)4, formatter->getRuleSetName(0), result, pos2, status = U_ZERO_ERROR);
if(result != expected) {
errln("Formatted 4 int64_t, expected " + expected + " got " + result);
} else {
logln("Formatted 4 int64_t, expected " + expected + " got " + result);
}
// clean up
logln("Cleaning up");
delete formatter;
}
void IntlTestRBNF::TestFractionalRuleSet()
{
UnicodeString fracRules(
"%main:\n"
// this rule formats the number if it's 1 or more. It formats
// the integral part using a DecimalFormat ("#,##0" puts
// thousands separators in the right places) and the fractional
// part using %%frac. If there is no fractional part, it
// just shows the integral part.
" x.0: <#,##0<[ >%%frac>];\n"
// this rule formats the number if it's between 0 and 1. It
// shows only the fractional part (0.5 shows up as "1/2," not
// "0 1/2")
" 0.x: >%%frac>;\n"
// the fraction rule set. This works the same way as the one in the
// preceding example: We multiply the fractional part of the number
// being formatted by each rule's base value and use the rule that
// produces the result closest to 0 (or the first rule that produces 0).
// Since we only provide rules for the numbers from 2 to 10, we know
// we'll get a fraction with a denominator between 2 and 10.
// "<0<" causes the numerator of the fraction to be formatted
// using numerals
"%%frac:\n"
" 2: 1/2;\n"
" 3: <0</3;\n"
" 4: <0</4;\n"
" 5: <0</5;\n"
" 6: <0</6;\n"
" 7: <0</7;\n"
" 8: <0</8;\n"
" 9: <0</9;\n"
" 10: <0</10;\n");
// mondo hack
int len = fracRules.length();
int change = 2;
for (int i = 0; i < len; ++i) {
UChar ch = fracRules.charAt(i);
if (ch == '\n') {
change = 2; // change ok
} else if (ch == ':') {
change = 1; // change, but once we hit a non-space char, don't change
} else if (ch == ' ') {
if (change != 0) {
fracRules.setCharAt(i, (UChar)0x200e);
}
} else {
if (change == 1) {
change = 0;
}
}
}
UErrorCode status = U_ZERO_ERROR;
UParseError perror;
RuleBasedNumberFormat formatter(fracRules, Locale::getEnglish(), perror, status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "0", "0" },
{ ".1", "1/10" },
{ ".11", "1/9" },
{ ".125", "1/8" },
{ ".1428", "1/7" },
{ ".1667", "1/6" },
{ ".2", "1/5" },
{ ".25", "1/4" },
{ ".333", "1/3" },
{ ".5", "1/2" },
{ "1.1", "1 1/10" },
{ "2.11", "2 1/9" },
{ "3.125", "3 1/8" },
{ "4.1428", "4 1/7" },
{ "5.1667", "5 1/6" },
{ "6.2", "6 1/5" },
{ "7.25", "7 1/4" },
{ "8.333", "8 1/3" },
{ "9.5", "9 1/2" },
{ ".2222", "2/9" },
{ ".4444", "4/9" },
{ ".5555", "5/9" },
{ "1.2856", "1 2/7" },
{ NULL, NULL }
};
doTest(&formatter, testData, FALSE); // exact values aren't parsable from fractions
}
}
#if 0
#define LLAssert(a) \
if (!(a)) errln("FAIL: " #a)
void IntlTestRBNF::TestLLongConstructors()
{
logln("Testing constructors");
// constant (shouldn't really be public)
LLAssert(llong(llong::kD32).asDouble() == llong::kD32);
// internal constructor (shouldn't really be public)
LLAssert(llong(0, 1).asDouble() == 1);
LLAssert(llong(1, 0).asDouble() == llong::kD32);
LLAssert(llong((uint32_t)-1, (uint32_t)-1).asDouble() == -1);
// public empty constructor
LLAssert(llong().asDouble() == 0);
// public int32_t constructor
LLAssert(llong((int32_t)0).asInt() == (int32_t)0);
LLAssert(llong((int32_t)1).asInt() == (int32_t)1);
LLAssert(llong((int32_t)-1).asInt() == (int32_t)-1);
LLAssert(llong((int32_t)0x7fffffff).asInt() == (int32_t)0x7fffffff);
LLAssert(llong((int32_t)0xffffffff).asInt() == (int32_t)-1);
LLAssert(llong((int32_t)0x80000000).asInt() == (int32_t)0x80000000);
// public int16_t constructor
LLAssert(llong((int16_t)0).asInt() == (int16_t)0);
LLAssert(llong((int16_t)1).asInt() == (int16_t)1);
LLAssert(llong((int16_t)-1).asInt() == (int16_t)-1);
LLAssert(llong((int16_t)0x7fff).asInt() == (int16_t)0x7fff);
LLAssert(llong((int16_t)0xffff).asInt() == (int16_t)0xffff);
LLAssert(llong((int16_t)0x8000).asInt() == (int16_t)0x8000);
// public int8_t constructor
LLAssert(llong((int8_t)0).asInt() == (int8_t)0);
LLAssert(llong((int8_t)1).asInt() == (int8_t)1);
LLAssert(llong((int8_t)-1).asInt() == (int8_t)-1);
LLAssert(llong((int8_t)0x7f).asInt() == (int8_t)0x7f);
LLAssert(llong((int8_t)0xff).asInt() == (int8_t)0xff);
LLAssert(llong((int8_t)0x80).asInt() == (int8_t)0x80);
// public uint16_t constructor
LLAssert(llong((uint16_t)0).asUInt() == (uint16_t)0);
LLAssert(llong((uint16_t)1).asUInt() == (uint16_t)1);
LLAssert(llong((uint16_t)-1).asUInt() == (uint16_t)-1);
LLAssert(llong((uint16_t)0x7fff).asUInt() == (uint16_t)0x7fff);
LLAssert(llong((uint16_t)0xffff).asUInt() == (uint16_t)0xffff);
LLAssert(llong((uint16_t)0x8000).asUInt() == (uint16_t)0x8000);
// public uint32_t constructor
LLAssert(llong((uint32_t)0).asUInt() == (uint32_t)0);
LLAssert(llong((uint32_t)1).asUInt() == (uint32_t)1);
LLAssert(llong((uint32_t)-1).asUInt() == (uint32_t)-1);
LLAssert(llong((uint32_t)0x7fffffff).asUInt() == (uint32_t)0x7fffffff);
LLAssert(llong((uint32_t)0xffffffff).asUInt() == (uint32_t)-1);
LLAssert(llong((uint32_t)0x80000000).asUInt() == (uint32_t)0x80000000);
// public double constructor
LLAssert(llong((double)0).asDouble() == (double)0);
LLAssert(llong((double)1).asDouble() == (double)1);
LLAssert(llong((double)0x7fffffff).asDouble() == (double)0x7fffffff);
LLAssert(llong((double)0x80000000).asDouble() == (double)0x80000000);
LLAssert(llong((double)0x80000001).asDouble() == (double)0x80000001);
// can't access uprv_maxmantissa, so fake it
double maxmantissa = (llong((int32_t)1) << 40).asDouble();
LLAssert(llong(maxmantissa).asDouble() == maxmantissa);
LLAssert(llong(-maxmantissa).asDouble() == -maxmantissa);
// copy constructor
LLAssert(llong(llong(0, 1)).asDouble() == 1);
LLAssert(llong(llong(1, 0)).asDouble() == llong::kD32);
LLAssert(llong(llong(-1, (uint32_t)-1)).asDouble() == -1);
// asInt - test unsigned to signed narrowing conversion
LLAssert(llong((uint32_t)-1).asInt() == (int32_t)0x7fffffff);
LLAssert(llong(-1, 0).asInt() == (int32_t)0x80000000);
// asUInt - test signed to unsigned narrowing conversion
LLAssert(llong((int32_t)-1).asUInt() == (uint32_t)-1);
LLAssert(llong((int32_t)0x80000000).asUInt() == (uint32_t)0x80000000);
// asDouble already tested
}
void IntlTestRBNF::TestLLongSimpleOperators()
{
logln("Testing simple operators");
// operator==
LLAssert(llong() == llong(0, 0));
LLAssert(llong(1,0) == llong(1, 0));
LLAssert(llong(0,1) == llong(0, 1));
// operator!=
LLAssert(llong(1,0) != llong(1,1));
LLAssert(llong(0,1) != llong(1,1));
LLAssert(llong(0xffffffff,0xffffffff) != llong(0x7fffffff, 0xffffffff));
// unsigned >
LLAssert(llong((int32_t)-1).ugt(llong(0x7fffffff, 0xffffffff)));
// unsigned <
LLAssert(llong(0x7fffffff, 0xffffffff).ult(llong((int32_t)-1)));
// unsigned >=
LLAssert(llong((int32_t)-1).uge(llong(0x7fffffff, 0xffffffff)));
LLAssert(llong((int32_t)-1).uge(llong((int32_t)-1)));
// unsigned <=
LLAssert(llong(0x7fffffff, 0xffffffff).ule(llong((int32_t)-1)));
LLAssert(llong((int32_t)-1).ule(llong((int32_t)-1)));
// operator>
LLAssert(llong(1, 1) > llong(1, 0));
LLAssert(llong(0, 0x80000000) > llong(0, 0x7fffffff));
LLAssert(llong(0x80000000, 1) > llong(0x80000000, 0));
LLAssert(llong(1, 0) > llong(0, 0x7fffffff));
LLAssert(llong(1, 0) > llong(0, 0xffffffff));
LLAssert(llong(0, 0) > llong(0x80000000, 1));
// operator<
LLAssert(llong(1, 0) < llong(1, 1));
LLAssert(llong(0, 0x7fffffff) < llong(0, 0x80000000));
LLAssert(llong(0x80000000, 0) < llong(0x80000000, 1));
LLAssert(llong(0, 0x7fffffff) < llong(1, 0));
LLAssert(llong(0, 0xffffffff) < llong(1, 0));
LLAssert(llong(0x80000000, 1) < llong(0, 0));
// operator>=
LLAssert(llong(1, 1) >= llong(1, 0));
LLAssert(llong(0, 0x80000000) >= llong(0, 0x7fffffff));
LLAssert(llong(0x80000000, 1) >= llong(0x80000000, 0));
LLAssert(llong(1, 0) >= llong(0, 0x7fffffff));
LLAssert(llong(1, 0) >= llong(0, 0xffffffff));
LLAssert(llong(0, 0) >= llong(0x80000000, 1));
LLAssert(llong() >= llong(0, 0));
LLAssert(llong(1,0) >= llong(1, 0));
LLAssert(llong(0,1) >= llong(0, 1));
// operator<=
LLAssert(llong(1, 0) <= llong(1, 1));
LLAssert(llong(0, 0x7fffffff) <= llong(0, 0x80000000));
LLAssert(llong(0x80000000, 0) <= llong(0x80000000, 1));
LLAssert(llong(0, 0x7fffffff) <= llong(1, 0));
LLAssert(llong(0, 0xffffffff) <= llong(1, 0));
LLAssert(llong(0x80000000, 1) <= llong(0, 0));
LLAssert(llong() <= llong(0, 0));
LLAssert(llong(1,0) <= llong(1, 0));
LLAssert(llong(0,1) <= llong(0, 1));
// operator==(int32)
LLAssert(llong() == (int32_t)0);
LLAssert(llong(0,1) == (int32_t)1);
// operator!=(int32)
LLAssert(llong(1,0) != (int32_t)0);
LLAssert(llong(0,1) != (int32_t)2);
LLAssert(llong(0,0xffffffff) != (int32_t)-1);
llong negOne(0xffffffff, 0xffffffff);
// operator>(int32)
LLAssert(llong(0, 0x80000000) > (int32_t)0x7fffffff);
LLAssert(negOne > (int32_t)-2);
LLAssert(llong(1, 0) > (int32_t)0x7fffffff);
LLAssert(llong(0, 0) > (int32_t)-1);
// operator<(int32)
LLAssert(llong(0, 0x7ffffffe) < (int32_t)0x7fffffff);
LLAssert(llong(0xffffffff, 0xfffffffe) < (int32_t)-1);
// operator>=(int32)
LLAssert(llong(0, 0x80000000) >= (int32_t)0x7fffffff);
LLAssert(negOne >= (int32_t)-2);
LLAssert(llong(1, 0) >= (int32_t)0x7fffffff);
LLAssert(llong(0, 0) >= (int32_t)-1);
LLAssert(llong() >= (int32_t)0);
LLAssert(llong(0,1) >= (int32_t)1);
// operator<=(int32)
LLAssert(llong(0, 0x7ffffffe) <= (int32_t)0x7fffffff);
LLAssert(llong(0xffffffff, 0xfffffffe) <= (int32_t)-1);
LLAssert(llong() <= (int32_t)0);
LLAssert(llong(0,1) <= (int32_t)1);
// operator=
LLAssert((llong(2,3) = llong((uint32_t)-1)).asUInt() == (uint32_t)-1);
// operator <<=
LLAssert((llong(1, 1) <<= 0) == llong(1, 1));
LLAssert((llong(1, 1) <<= 31) == llong(0x80000000, 0x80000000));
LLAssert((llong(1, 1) <<= 32) == llong(1, 0));
LLAssert((llong(1, 1) <<= 63) == llong(0x80000000, 0));
LLAssert((llong(1, 1) <<= 64) == llong(1, 1)); // only lower 6 bits are used
LLAssert((llong(1, 1) <<= -1) == llong(0x80000000, 0)); // only lower 6 bits are used
// operator <<
LLAssert((llong((int32_t)1) << 5).asUInt() == 32);
// operator >>= (sign extended)
LLAssert((llong(0x7fffa0a0, 0xbcbcdfdf) >>= 16) == llong(0x7fff,0xa0a0bcbc));
LLAssert((llong(0x8000789a, 0xbcde0000) >>= 16) == llong(0xffff8000,0x789abcde));
LLAssert((llong(0x80000000, 0) >>= 63) == llong(0xffffffff, 0xffffffff));
LLAssert((llong(0x80000000, 0) >>= 47) == llong(0xffffffff, 0xffff0000));
LLAssert((llong(0x80000000, 0x80000000) >> 64) == llong(0x80000000, 0x80000000)); // only lower 6 bits are used
LLAssert((llong(0x80000000, 0) >>= -1) == llong(0xffffffff, 0xffffffff)); // only lower 6 bits are used
// operator >> sign extended)
LLAssert((llong(0x8000789a, 0xbcde0000) >> 16) == llong(0xffff8000,0x789abcde));
// ushr (right shift without sign extension)
LLAssert(llong(0x7fffa0a0, 0xbcbcdfdf).ushr(16) == llong(0x7fff,0xa0a0bcbc));
LLAssert(llong(0x8000789a, 0xbcde0000).ushr(16) == llong(0x00008000,0x789abcde));
LLAssert(llong(0x80000000, 0).ushr(63) == llong(0, 1));
LLAssert(llong(0x80000000, 0).ushr(47) == llong(0, 0x10000));
LLAssert(llong(0x80000000, 0x80000000).ushr(64) == llong(0x80000000, 0x80000000)); // only lower 6 bits are used
LLAssert(llong(0x80000000, 0).ushr(-1) == llong(0, 1)); // only lower 6 bits are used
// operator&(llong)
LLAssert((llong(0x55555555, 0x55555555) & llong(0xaaaaffff, 0xffffaaaa)) == llong(0x00005555, 0x55550000));
// operator|(llong)
LLAssert((llong(0x55555555, 0x55555555) | llong(0xaaaaffff, 0xffffaaaa)) == llong(0xffffffff, 0xffffffff));
// operator^(llong)
LLAssert((llong(0x55555555, 0x55555555) ^ llong(0xaaaaffff, 0xffffaaaa)) == llong(0xffffaaaa, 0xaaaaffff));
// operator&(uint32)
LLAssert((llong(0x55555555, 0x55555555) & (uint32_t)0xffffaaaa) == llong(0, 0x55550000));
// operator|(uint32)
LLAssert((llong(0x55555555, 0x55555555) | (uint32_t)0xffffaaaa) == llong(0x55555555, 0xffffffff));
// operator^(uint32)
LLAssert((llong(0x55555555, 0x55555555) ^ (uint32_t)0xffffaaaa) == llong(0x55555555, 0xaaaaffff));
// operator~
LLAssert(~llong(0x55555555, 0x55555555) == llong(0xaaaaaaaa, 0xaaaaaaaa));
// operator&=(llong)
LLAssert((llong(0x55555555, 0x55555555) &= llong(0xaaaaffff, 0xffffaaaa)) == llong(0x00005555, 0x55550000));
// operator|=(llong)
LLAssert((llong(0x55555555, 0x55555555) |= llong(0xaaaaffff, 0xffffaaaa)) == llong(0xffffffff, 0xffffffff));
// operator^=(llong)
LLAssert((llong(0x55555555, 0x55555555) ^= llong(0xaaaaffff, 0xffffaaaa)) == llong(0xffffaaaa, 0xaaaaffff));
// operator&=(uint32)
LLAssert((llong(0x55555555, 0x55555555) &= (uint32_t)0xffffaaaa) == llong(0, 0x55550000));
// operator|=(uint32)
LLAssert((llong(0x55555555, 0x55555555) |= (uint32_t)0xffffaaaa) == llong(0x55555555, 0xffffffff));
// operator^=(uint32)
LLAssert((llong(0x55555555, 0x55555555) ^= (uint32_t)0xffffaaaa) == llong(0x55555555, 0xaaaaffff));
// prefix inc
LLAssert(llong(1, 0) == ++llong(0,0xffffffff));
// prefix dec
LLAssert(llong(0,0xffffffff) == --llong(1, 0));
// postfix inc
{
llong n(0, 0xffffffff);
LLAssert(llong(0, 0xffffffff) == n++);
LLAssert(llong(1, 0) == n);
}
// postfix dec
{
llong n(1, 0);
LLAssert(llong(1, 0) == n--);
LLAssert(llong(0, 0xffffffff) == n);
}
// unary minus
LLAssert(llong(0, 0) == -llong(0, 0));
LLAssert(llong(0xffffffff, 0xffffffff) == -llong(0, 1));
LLAssert(llong(0, 1) == -llong(0xffffffff, 0xffffffff));
LLAssert(llong(0x7fffffff, 0xffffffff) == -llong(0x80000000, 1));
LLAssert(llong(0x80000000, 0) == -llong(0x80000000, 0)); // !!! we don't handle overflow
// operator-=
{
llong n;
LLAssert((n -= llong(0, 1)) == llong(0xffffffff, 0xffffffff));
LLAssert(n == llong(0xffffffff, 0xffffffff));
n = llong(1, 0);
LLAssert((n -= llong(0, 1)) == llong(0, 0xffffffff));
LLAssert(n == llong(0, 0xffffffff));
}
// operator-
{
llong n;
LLAssert((n - llong(0, 1)) == llong(0xffffffff, 0xffffffff));
LLAssert(n == llong(0, 0));
n = llong(1, 0);
LLAssert((n - llong(0, 1)) == llong(0, 0xffffffff));
LLAssert(n == llong(1, 0));
}
// operator+=
{
llong n(0xffffffff, 0xffffffff);
LLAssert((n += llong(0, 1)) == llong(0, 0));
LLAssert(n == llong(0, 0));
n = llong(0, 0xffffffff);
LLAssert((n += llong(0, 1)) == llong(1, 0));
LLAssert(n == llong(1, 0));
}
// operator+
{
llong n(0xffffffff, 0xffffffff);
LLAssert((n + llong(0, 1)) == llong(0, 0));
LLAssert(n == llong(0xffffffff, 0xffffffff));
n = llong(0, 0xffffffff);
LLAssert((n + llong(0, 1)) == llong(1, 0));
LLAssert(n == llong(0, 0xffffffff));
}
}
void IntlTestRBNF::TestLLong()
{
logln("Starting TestLLong");
TestLLongConstructors();
TestLLongSimpleOperators();
logln("Testing operator*=, operator*");
// operator*=, operator*
// small and large values, positive, &NEGative, zero
// also test commutivity
{
const llong ZERO;
const llong ONE(0, 1);
const llong NEG_ONE((int32_t)-1);
const llong THREE(0, 3);
const llong NEG_THREE((int32_t)-3);
const llong TWO_TO_16(0, 0x10000);
const llong NEG_TWO_TO_16 = -TWO_TO_16;
const llong TWO_TO_32(1, 0);
const llong NEG_TWO_TO_32 = -TWO_TO_32;
const llong NINE(0, 9);
const llong NEG_NINE = -NINE;
const llong TWO_TO_16X3(0, 0x00030000);
const llong NEG_TWO_TO_16X3 = -TWO_TO_16X3;
const llong TWO_TO_32X3(3, 0);
const llong NEG_TWO_TO_32X3 = -TWO_TO_32X3;
const llong TWO_TO_48(0x10000, 0);
const llong NEG_TWO_TO_48 = -TWO_TO_48;
const int32_t VALUE_WIDTH = 9;
const llong* values[VALUE_WIDTH] = {
&ZERO, &ONE, &NEG_ONE, &THREE, &NEG_THREE, &TWO_TO_16, &NEG_TWO_TO_16, &TWO_TO_32, &NEG_TWO_TO_32
};
const llong* answers[VALUE_WIDTH*VALUE_WIDTH] = {
&ZERO, &ZERO, &ZERO, &ZERO, &ZERO, &ZERO, &ZERO, &ZERO, &ZERO,
&ZERO, &ONE, &NEG_ONE, &THREE, &NEG_THREE, &TWO_TO_16, &NEG_TWO_TO_16, &TWO_TO_32, &NEG_TWO_TO_32,
&ZERO, &NEG_ONE, &ONE, &NEG_THREE, &THREE, &NEG_TWO_TO_16, &TWO_TO_16, &NEG_TWO_TO_32, &TWO_TO_32,
&ZERO, &THREE, &NEG_THREE, &NINE, &NEG_NINE, &TWO_TO_16X3, &NEG_TWO_TO_16X3, &TWO_TO_32X3, &NEG_TWO_TO_32X3,
&ZERO, &NEG_THREE, &THREE, &NEG_NINE, &NINE, &NEG_TWO_TO_16X3, &TWO_TO_16X3, &NEG_TWO_TO_32X3, &TWO_TO_32X3,
&ZERO, &TWO_TO_16, &NEG_TWO_TO_16, &TWO_TO_16X3, &NEG_TWO_TO_16X3, &TWO_TO_32, &NEG_TWO_TO_32, &TWO_TO_48, &NEG_TWO_TO_48,
&ZERO, &NEG_TWO_TO_16, &TWO_TO_16, &NEG_TWO_TO_16X3, &TWO_TO_16X3, &NEG_TWO_TO_32, &TWO_TO_32, &NEG_TWO_TO_48, &TWO_TO_48,
&ZERO, &TWO_TO_32, &NEG_TWO_TO_32, &TWO_TO_32X3, &NEG_TWO_TO_32X3, &TWO_TO_48, &NEG_TWO_TO_48, &ZERO, &ZERO,
&ZERO, &NEG_TWO_TO_32, &TWO_TO_32, &NEG_TWO_TO_32X3, &TWO_TO_32X3, &NEG_TWO_TO_48, &TWO_TO_48, &ZERO, &ZERO
};
for (int i = 0; i < VALUE_WIDTH; ++i) {
for (int j = 0; j < VALUE_WIDTH; ++j) {
llong lhs = *values[i];
llong rhs = *values[j];
llong ans = *answers[i*VALUE_WIDTH + j];
llong n = lhs;
LLAssert((n *= rhs) == ans);
LLAssert(n == ans);
n = lhs;
LLAssert((n * rhs) == ans);
LLAssert(n == lhs);
}
}
}
logln("Testing operator/=, operator/");
// operator/=, operator/
// test num = 0, div = 0, pos/neg, > 2^32, div > num
{
const llong ZERO;
const llong ONE(0, 1);
const llong NEG_ONE = -ONE;
const llong MAX(0x7fffffff, 0xffffffff);
const llong MIN(0x80000000, 0);
const llong TWO(0, 2);
const llong NEG_TWO = -TWO;
const llong FIVE(0, 5);
const llong NEG_FIVE = -FIVE;
const llong TWO_TO_32(1, 0);
const llong NEG_TWO_TO_32 = -TWO_TO_32;
const llong TWO_TO_32d5 = llong(TWO_TO_32.asDouble()/5.0);
const llong NEG_TWO_TO_32d5 = -TWO_TO_32d5;
const llong TWO_TO_32X5 = TWO_TO_32 * FIVE;
const llong NEG_TWO_TO_32X5 = -TWO_TO_32X5;
const llong* tuples[] = { // lhs, rhs, ans
&ZERO, &ZERO, &ZERO,
&ONE, &ZERO,&MAX,
&NEG_ONE, &ZERO, &MIN,
&ONE, &ONE, &ONE,
&ONE, &NEG_ONE, &NEG_ONE,
&NEG_ONE, &ONE, &NEG_ONE,
&NEG_ONE, &NEG_ONE, &ONE,
&FIVE, &TWO, &TWO,
&FIVE, &NEG_TWO, &NEG_TWO,
&NEG_FIVE, &TWO, &NEG_TWO,
&NEG_FIVE, &NEG_TWO, &TWO,
&TWO, &FIVE, &ZERO,
&TWO, &NEG_FIVE, &ZERO,
&NEG_TWO, &FIVE, &ZERO,
&NEG_TWO, &NEG_FIVE, &ZERO,
&TWO_TO_32, &TWO_TO_32, &ONE,
&TWO_TO_32, &NEG_TWO_TO_32, &NEG_ONE,
&NEG_TWO_TO_32, &TWO_TO_32, &NEG_ONE,
&NEG_TWO_TO_32, &NEG_TWO_TO_32, &ONE,
&TWO_TO_32, &FIVE, &TWO_TO_32d5,
&TWO_TO_32, &NEG_FIVE, &NEG_TWO_TO_32d5,
&NEG_TWO_TO_32, &FIVE, &NEG_TWO_TO_32d5,
&NEG_TWO_TO_32, &NEG_FIVE, &TWO_TO_32d5,
&TWO_TO_32X5, &FIVE, &TWO_TO_32,
&TWO_TO_32X5, &NEG_FIVE, &NEG_TWO_TO_32,
&NEG_TWO_TO_32X5, &FIVE, &NEG_TWO_TO_32,
&NEG_TWO_TO_32X5, &NEG_FIVE, &TWO_TO_32,
&TWO_TO_32X5, &TWO_TO_32, &FIVE,
&TWO_TO_32X5, &NEG_TWO_TO_32, &NEG_FIVE,
&NEG_TWO_TO_32X5, &NEG_TWO_TO_32, &FIVE,
&NEG_TWO_TO_32X5, &TWO_TO_32, &NEG_FIVE
};
const int TUPLE_WIDTH = 3;
const int TUPLE_COUNT = (int)(sizeof(tuples)/sizeof(tuples[0]))/TUPLE_WIDTH;
for (int i = 0; i < TUPLE_COUNT; ++i) {
const llong lhs = *tuples[i*TUPLE_WIDTH+0];
const llong rhs = *tuples[i*TUPLE_WIDTH+1];
const llong ans = *tuples[i*TUPLE_WIDTH+2];
llong n = lhs;
if (!((n /= rhs) == ans)) {
errln("fail: (n /= rhs) == ans");
}
LLAssert(n == ans);
n = lhs;
LLAssert((n / rhs) == ans);
LLAssert(n == lhs);
}
}
logln("Testing operator%%=, operator%%");
//operator%=, operator%
{
const llong ZERO;
const llong ONE(0, 1);
const llong TWO(0, 2);
const llong THREE(0,3);
const llong FOUR(0, 4);
const llong FIVE(0, 5);
const llong SIX(0, 6);
const llong NEG_ONE = -ONE;
const llong NEG_TWO = -TWO;
const llong NEG_THREE = -THREE;
const llong NEG_FOUR = -FOUR;
const llong NEG_FIVE = -FIVE;
const llong NEG_SIX = -SIX;
const llong NINETY_NINE(0, 99);
const llong HUNDRED(0, 100);
const llong HUNDRED_ONE(0, 101);
const llong BIG(0x12345678, 0x9abcdef0);
const llong BIG_FIVE(BIG * FIVE);
const llong BIG_FIVEm1 = BIG_FIVE - ONE;
const llong BIG_FIVEp1 = BIG_FIVE + ONE;
const llong* tuples[] = {
&ZERO, &FIVE, &ZERO,
&ONE, &FIVE, &ONE,
&TWO, &FIVE, &TWO,
&THREE, &FIVE, &THREE,
&FOUR, &FIVE, &FOUR,
&FIVE, &FIVE, &ZERO,
&SIX, &FIVE, &ONE,
&ZERO, &NEG_FIVE, &ZERO,
&ONE, &NEG_FIVE, &ONE,
&TWO, &NEG_FIVE, &TWO,
&THREE, &NEG_FIVE, &THREE,
&FOUR, &NEG_FIVE, &FOUR,
&FIVE, &NEG_FIVE, &ZERO,
&SIX, &NEG_FIVE, &ONE,
&NEG_ONE, &FIVE, &NEG_ONE,
&NEG_TWO, &FIVE, &NEG_TWO,
&NEG_THREE, &FIVE, &NEG_THREE,
&NEG_FOUR, &FIVE, &NEG_FOUR,
&NEG_FIVE, &FIVE, &ZERO,
&NEG_SIX, &FIVE, &NEG_ONE,
&NEG_ONE, &NEG_FIVE, &NEG_ONE,
&NEG_TWO, &NEG_FIVE, &NEG_TWO,
&NEG_THREE, &NEG_FIVE, &NEG_THREE,
&NEG_FOUR, &NEG_FIVE, &NEG_FOUR,
&NEG_FIVE, &NEG_FIVE, &ZERO,
&NEG_SIX, &NEG_FIVE, &NEG_ONE,
&NINETY_NINE, &FIVE, &FOUR,
&HUNDRED, &FIVE, &ZERO,
&HUNDRED_ONE, &FIVE, &ONE,
&BIG_FIVEm1, &FIVE, &FOUR,
&BIG_FIVE, &FIVE, &ZERO,
&BIG_FIVEp1, &FIVE, &ONE
};
const int TUPLE_WIDTH = 3;
const int TUPLE_COUNT = (int)(sizeof(tuples)/sizeof(tuples[0]))/TUPLE_WIDTH;
for (int i = 0; i < TUPLE_COUNT; ++i) {
const llong lhs = *tuples[i*TUPLE_WIDTH+0];
const llong rhs = *tuples[i*TUPLE_WIDTH+1];
const llong ans = *tuples[i*TUPLE_WIDTH+2];
llong n = lhs;
if (!((n %= rhs) == ans)) {
errln("fail: (n %= rhs) == ans");
}
LLAssert(n == ans);
n = lhs;
LLAssert((n % rhs) == ans);
LLAssert(n == lhs);
}
}
logln("Testing pow");
// pow
LLAssert(llong(0, 0).pow(0) == llong(0, 0));
LLAssert(llong(0, 0).pow(2) == llong(0, 0));
LLAssert(llong(0, 2).pow(0) == llong(0, 1));
LLAssert(llong(0, 2).pow(2) == llong(0, 4));
LLAssert(llong(0, 2).pow(32) == llong(1, 0));
LLAssert(llong(0, 5).pow(10) == llong((double)5.0 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5));
// absolute value
{
const llong n(0xffffffff,0xffffffff);
LLAssert(n.abs() == llong(0, 1));
}
#ifdef RBNF_DEBUG
logln("Testing atoll");
// atoll
const char empty[] = "";
const char zero[] = "0";
const char neg_one[] = "-1";
const char neg_12345[] = "-12345";
const char big1[] = "123456789abcdef0";
const char big2[] = "fFfFfFfFfFfFfFfF";
LLAssert(llong::atoll(empty) == llong(0, 0));
LLAssert(llong::atoll(zero) == llong(0, 0));
LLAssert(llong::atoll(neg_one) == llong(0xffffffff, 0xffffffff));
LLAssert(llong::atoll(neg_12345) == -llong(0, 12345));
LLAssert(llong::atoll(big1, 16) == llong(0x12345678, 0x9abcdef0));
LLAssert(llong::atoll(big2, 16) == llong(0xffffffff, 0xffffffff));
#endif
// u_atoll
const UChar uempty[] = { 0 };
const UChar uzero[] = { 0x30, 0 };
const UChar uneg_one[] = { 0x2d, 0x31, 0 };
const UChar uneg_12345[] = { 0x2d, 0x31, 0x32, 0x33, 0x34, 0x35, 0 };
const UChar ubig1[] = { 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x30, 0 };
const UChar ubig2[] = { 0x66, 0x46, 0x66, 0x46, 0x66, 0x46, 0x66, 0x46, 0x66, 0x46, 0x66, 0x46, 0x66, 0x46, 0x66, 0x46, 0 };
LLAssert(llong::utoll(uempty) == llong(0, 0));
LLAssert(llong::utoll(uzero) == llong(0, 0));
LLAssert(llong::utoll(uneg_one) == llong(0xffffffff, 0xffffffff));
LLAssert(llong::utoll(uneg_12345) == -llong(0, 12345));
LLAssert(llong::utoll(ubig1, 16) == llong(0x12345678, 0x9abcdef0));
LLAssert(llong::utoll(ubig2, 16) == llong(0xffffffff, 0xffffffff));
#ifdef RBNF_DEBUG
logln("Testing lltoa");
// lltoa
{
char buf[64]; // ascii
LLAssert((llong(0, 0).lltoa(buf, (uint32_t)sizeof(buf)) == 1) && (strcmp(buf, zero) == 0));
LLAssert((llong(0xffffffff, 0xffffffff).lltoa(buf, (uint32_t)sizeof(buf)) == 2) && (strcmp(buf, neg_one) == 0));
LLAssert(((-llong(0, 12345)).lltoa(buf, (uint32_t)sizeof(buf)) == 6) && (strcmp(buf, neg_12345) == 0));
LLAssert((llong(0x12345678, 0x9abcdef0).lltoa(buf, (uint32_t)sizeof(buf), 16) == 16) && (strcmp(buf, big1) == 0));
}
#endif
logln("Testing u_lltoa");
// u_lltoa
{
UChar buf[64];
LLAssert((llong(0, 0).lltou(buf, (uint32_t)sizeof(buf)) == 1) && (u_strcmp(buf, uzero) == 0));
LLAssert((llong(0xffffffff, 0xffffffff).lltou(buf, (uint32_t)sizeof(buf)) == 2) && (u_strcmp(buf, uneg_one) == 0));
LLAssert(((-llong(0, 12345)).lltou(buf, (uint32_t)sizeof(buf)) == 6) && (u_strcmp(buf, uneg_12345) == 0));
LLAssert((llong(0x12345678, 0x9abcdef0).lltou(buf, (uint32_t)sizeof(buf), 16) == 16) && (u_strcmp(buf, ubig1) == 0));
}
}
/* if 0 */
#endif
void
IntlTestRBNF::TestEnglishSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale::getUS(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "one" },
{ "2", "two" },
{ "15", "fifteen" },
{ "20", "twenty" },
{ "23", "twenty-three" },
{ "73", "seventy-three" },
{ "88", "eighty-eight" },
{ "100", "one hundred" },
{ "106", "one hundred six" },
{ "127", "one hundred twenty-seven" },
{ "200", "two hundred" },
{ "579", "five hundred seventy-nine" },
{ "1,000", "one thousand" },
{ "2,000", "two thousand" },
{ "3,004", "three thousand four" },
{ "4,567", "four thousand five hundred sixty-seven" },
{ "15,943", "fifteen thousand nine hundred forty-three" },
{ "2,345,678", "two million three hundred forty-five thousand six hundred seventy-eight" },
{ "-36", "minus thirty-six" },
{ "234.567", "two hundred thirty-four point five six seven" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
#if !UCONFIG_NO_COLLATION
formatter->setLenient(TRUE);
static const char* lpTestData[][2] = {
{ "fifty-7", "57" },
{ " fifty-7", "57" },
{ " fifty-7", "57" },
{ "2 thousand six HUNDRED fifty-7", "2,657" },
{ "fifteen hundred and zero", "1,500" },
{ "FOurhundred thiRTY six", "436" },
{ NULL, NULL}
};
doLenientParseTest(formatter, lpTestData);
#endif
}
delete formatter;
}
void
IntlTestRBNF::TestOrdinalAbbreviations()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_ORDINAL, Locale::getUS(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "1\\u02e2\\u1d57" },
{ "2", "2\\u207f\\u1d48" },
{ "3", "3\\u02b3\\u1d48" },
{ "4", "4\\u1d57\\u02b0" },
{ "7", "7\\u1d57\\u02b0" },
{ "10", "10\\u1d57\\u02b0" },
{ "11", "11\\u1d57\\u02b0" },
{ "13", "13\\u1d57\\u02b0" },
{ "20", "20\\u1d57\\u02b0" },
{ "21", "21\\u02e2\\u1d57" },
{ "22", "22\\u207f\\u1d48" },
{ "23", "23\\u02b3\\u1d48" },
{ "24", "24\\u1d57\\u02b0" },
{ "33", "33\\u02b3\\u1d48" },
{ "102", "102\\u207f\\u1d48" },
{ "312", "312\\u1d57\\u02b0" },
{ "12,345", "12,345\\u1d57\\u02b0" },
{ NULL, NULL}
};
doTest(formatter, testData, FALSE);
}
delete formatter;
}
void
IntlTestRBNF::TestDurations()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_DURATION, Locale::getUS(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "3,600", "1:00:00" }, //move me and I fail
{ "0", "0 sec." },
{ "1", "1 sec." },
{ "24", "24 sec." },
{ "60", "1:00" },
{ "73", "1:13" },
{ "145", "2:25" },
{ "666", "11:06" },
// { "3,600", "1:00:00" },
{ "3,740", "1:02:20" },
{ "10,293", "2:51:33" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
#if !UCONFIG_NO_COLLATION
formatter->setLenient(TRUE);
static const char* lpTestData[][2] = {
{ "2-51-33", "10,293" },
{ NULL, NULL}
};
doLenientParseTest(formatter, lpTestData);
#endif
}
delete formatter;
}
void
IntlTestRBNF::TestSpanishSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("es", "ES", ""), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "uno" },
{ "6", "seis" },
{ "16", "diecis\\u00e9is" },
{ "20", "veinte" },
{ "24", "veinticuatro" },
{ "26", "veintis\\u00e9is" },
{ "73", "setenta y tres" },
{ "88", "ochenta y ocho" },
{ "100", "cien" },
{ "106", "ciento seis" },
{ "127", "ciento veintisiete" },
{ "200", "doscientos" },
{ "579", "quinientos setenta y nueve" },
{ "1,000", "mil" },
{ "2,000", "dos mil" },
{ "3,004", "tres mil cuatro" },
{ "4,567", "cuatro mil quinientos sesenta y siete" },
{ "15,943", "quince mil novecientos cuarenta y tres" },
{ "2,345,678", "dos millones trescientos cuarenta y cinco mil seiscientos setenta y ocho"},
{ "-36", "menos treinta y seis" },
{ "234.567", "doscientos treinta y cuatro coma cinco seis siete" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
}
delete formatter;
}
void
IntlTestRBNF::TestFrenchSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale::getFrance(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "un" },
{ "15", "quinze" },
{ "20", "vingt" },
{ "21", "vingt-et-un" },
{ "23", "vingt-trois" },
{ "62", "soixante-deux" },
{ "70", "soixante-dix" },
{ "71", "soixante-et-onze" },
{ "73", "soixante-treize" },
{ "80", "quatre-vingts" },
{ "88", "quatre-vingt-huit" },
{ "100", "cent" },
{ "106", "cent-six" },
{ "127", "cent-vingt-sept" },
{ "200", "deux-cents" },
{ "579", "cinq-cent-soixante-dix-neuf" },
{ "1,000", "mille" },
{ "1,123", "mille-cent-vingt-trois" },
{ "1,594", "mille-cinq-cent-quatre-vingt-quatorze" },
{ "2,000", "deux-mille" },
{ "3,004", "trois-mille-quatre" },
{ "4,567", "quatre-mille-cinq-cent-soixante-sept" },
{ "15,943", "quinze-mille-neuf-cent-quarante-trois" },
{ "2,345,678", "deux millions trois-cent-quarante-cinq-mille-six-cent-soixante-dix-huit" },
{ "-36", "moins trente-six" },
{ "234.567", "deux-cent-trente-quatre virgule cinq six sept" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
#if !UCONFIG_NO_COLLATION
formatter->setLenient(TRUE);
static const char* lpTestData[][2] = {
{ "trente-et-un", "31" },
{ "un cent quatre vingt dix huit", "198" },
{ NULL, NULL}
};
doLenientParseTest(formatter, lpTestData);
#endif
}
delete formatter;
}
static const char* const swissFrenchTestData[][2] = {
{ "1", "un" },
{ "15", "quinze" },
{ "20", "vingt" },
{ "21", "vingt-et-un" },
{ "23", "vingt-trois" },
{ "62", "soixante-deux" },
{ "70", "septante" },
{ "71", "septante-et-un" },
{ "73", "septante-trois" },
{ "80", "huitante" },
{ "88", "huitante-huit" },
{ "100", "cent" },
{ "106", "cent-six" },
{ "127", "cent-vingt-sept" },
{ "200", "deux-cents" },
{ "579", "cinq-cent-septante-neuf" },
{ "1,000", "mille" },
{ "1,123", "mille-cent-vingt-trois" },
{ "1,594", "mille-cinq-cent-nonante-quatre" },
{ "2,000", "deux-mille" },
{ "3,004", "trois-mille-quatre" },
{ "4,567", "quatre-mille-cinq-cent-soixante-sept" },
{ "15,943", "quinze-mille-neuf-cent-quarante-trois" },
{ "2,345,678", "deux millions trois-cent-quarante-cinq-mille-six-cent-septante-huit" },
{ "-36", "moins trente-six" },
{ "234.567", "deux-cent-trente-quatre virgule cinq six sept" },
{ NULL, NULL}
};
void
IntlTestRBNF::TestSwissFrenchSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("fr", "CH", ""), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
doTest(formatter, swissFrenchTestData, TRUE);
}
delete formatter;
}
static const char* const belgianFrenchTestData[][2] = {
{ "1", "un" },
{ "15", "quinze" },
{ "20", "vingt" },
{ "21", "vingt-et-un" },
{ "23", "vingt-trois" },
{ "62", "soixante-deux" },
{ "70", "septante" },
{ "71", "septante-et-un" },
{ "73", "septante-trois" },
{ "80", "quatre-vingts" },
{ "88", "quatre-vingt-huit" },
{ "90", "nonante" },
{ "91", "nonante-et-un" },
{ "95", "nonante-cinq" },
{ "100", "cent" },
{ "106", "cent-six" },
{ "127", "cent-vingt-sept" },
{ "200", "deux-cents" },
{ "579", "cinq-cent-septante-neuf" },
{ "1,000", "mille" },
{ "1,123", "mille-cent-vingt-trois" },
{ "1,594", "mille-cinq-cent-nonante-quatre" },
{ "2,000", "deux-mille" },
{ "3,004", "trois-mille-quatre" },
{ "4,567", "quatre-mille-cinq-cent-soixante-sept" },
{ "15,943", "quinze-mille-neuf-cent-quarante-trois" },
{ "2,345,678", "deux millions trois-cent-quarante-cinq-mille-six-cent-septante-huit" },
{ "-36", "moins trente-six" },
{ "234.567", "deux-cent-trente-quatre virgule cinq six sept" },
{ NULL, NULL}
};
void
IntlTestRBNF::TestBelgianFrenchSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("fr", "BE", ""), status);
if (U_FAILURE(status)) {
errcheckln(status, "rbnf status: 0x%x (%s)\n", status, u_errorName(status));
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
// Belgian french should match Swiss french.
doTest(formatter, belgianFrenchTestData, TRUE);
}
delete formatter;
}
void
IntlTestRBNF::TestItalianSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale::getItalian(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "uno" },
{ "15", "quindici" },
{ "20", "venti" },
{ "23", "venti\\u00ADtr\\u00E9" },
{ "73", "settanta\\u00ADtr\\u00E9" },
{ "88", "ottant\\u00ADotto" },
{ "100", "cento" },
{ "101", "cent\\u00ADuno" },
{ "103", "cento\\u00ADtr\\u00E9" },
{ "106", "cento\\u00ADsei" },
{ "108", "cent\\u00ADotto" },
{ "127", "cento\\u00ADventi\\u00ADsette" },
{ "181", "cent\\u00ADottant\\u00ADuno" },
{ "200", "due\\u00ADcento" },
{ "579", "cinque\\u00ADcento\\u00ADsettanta\\u00ADnove" },
{ "1,000", "mille" },
{ "2,000", "due\\u00ADmila" },
{ "3,004", "tre\\u00ADmila\\u00ADquattro" },
{ "4,567", "quattro\\u00ADmila\\u00ADcinque\\u00ADcento\\u00ADsessanta\\u00ADsette" },
{ "15,943", "quindici\\u00ADmila\\u00ADnove\\u00ADcento\\u00ADquaranta\\u00ADtr\\u00E9" },
{ "-36", "meno trenta\\u00ADsei" },
{ "234.567", "due\\u00ADcento\\u00ADtrenta\\u00ADquattro virgola cinque sei sette" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
}
delete formatter;
}
void
IntlTestRBNF::TestPortugueseSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("pt","BR",""), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "um" },
{ "15", "quinze" },
{ "20", "vinte" },
{ "23", "vinte e tr\\u00EAs" },
{ "73", "setenta e tr\\u00EAs" },
{ "88", "oitenta e oito" },
{ "100", "cem" },
{ "106", "cento e seis" },
{ "108", "cento e oito" },
{ "127", "cento e vinte e sete" },
{ "181", "cento e oitenta e um" },
{ "200", "duzcentos" },
{ "579", "quinhentos e setenta e nove" },
{ "1,000", "mil" },
{ "2,000", "dois mil" },
{ "3,004", "tr\\u00EAs mil e quatro" },
{ "4,567", "quatro mil e quinhentos e sessenta e sete" },
{ "15,943", "quinze mil e novecentos e quarenta e tr\\u00EAs" },
{ "-36", "menos trinta e seis" },
{ "234.567", "duzcentos e trinta e quatro v\\u00EDrgula cinco seis sete" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
}
delete formatter;
}
void
IntlTestRBNF::TestGermanSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale::getGermany(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "1", "eins" },
{ "15", "f\\u00fcnfzehn" },
{ "20", "zwanzig" },
{ "23", "drei\\u00ADund\\u00ADzwanzig" },
{ "73", "drei\\u00ADund\\u00ADsiebzig" },
{ "88", "acht\\u00ADund\\u00ADachtzig" },
{ "100", "ein\\u00ADhundert" },
{ "106", "ein\\u00ADhundert\\u00ADsechs" },
{ "127", "ein\\u00ADhundert\\u00ADsieben\\u00ADund\\u00ADzwanzig" },
{ "200", "zwei\\u00ADhundert" },
{ "579", "f\\u00fcnf\\u00ADhundert\\u00ADneun\\u00ADund\\u00ADsiebzig" },
{ "1,000", "ein\\u00ADtausend" },
{ "2,000", "zwei\\u00ADtausend" },
{ "3,004", "drei\\u00ADtausend\\u00ADvier" },
{ "4,567", "vier\\u00ADtausend\\u00ADf\\u00fcnf\\u00ADhundert\\u00ADsieben\\u00ADund\\u00ADsechzig" },
{ "15,943", "f\\u00fcnfzehn\\u00ADtausend\\u00ADneun\\u00ADhundert\\u00ADdrei\\u00ADund\\u00ADvierzig" },
{ "2,345,678", "zwei Millionen drei\\u00ADhundert\\u00ADf\\u00fcnf\\u00ADund\\u00ADvierzig\\u00ADtausend\\u00ADsechs\\u00ADhundert\\u00ADacht\\u00ADund\\u00ADsiebzig" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
#if !UCONFIG_NO_COLLATION
formatter->setLenient(TRUE);
static const char* lpTestData[][2] = {
{ "ein Tausend sechs Hundert fuenfunddreissig", "1,635" },
{ NULL, NULL}
};
doLenientParseTest(formatter, lpTestData);
#endif
}
delete formatter;
}
void
IntlTestRBNF::TestThaiSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("th"), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testData[][2] = {
{ "0", "\\u0e28\\u0e39\\u0e19\\u0e22\\u0e4c" },
{ "1", "\\u0e2b\\u0e19\\u0e36\\u0e48\\u0e07" },
{ "10", "\\u0e2a\\u0e34\\u0e1a" },
{ "11", "\\u0e2a\\u0e34\\u0e1a\\u200b\\u0e40\\u0e2d\\u0e47\\u0e14" },
{ "21", "\\u0e22\\u0e35\\u0e48\\u200b\\u0e2a\\u0e34\\u0e1a\\u200b\\u0e40\\u0e2d\\u0e47\\u0e14" },
{ "101", "\\u0e2b\\u0e19\\u0e36\\u0e48\\u0e07\\u200b\\u0e23\\u0e49\\u0e2d\\u0e22\\u200b\\u0e2b\\u0e19\\u0e36\\u0e48\\u0e07" },
{ "1.234", "\\u0e2b\\u0e19\\u0e36\\u0e48\\u0e07\\u200b\\u0e08\\u0e38\\u0e14\\u200b\\u0e2a\\u0e2d\\u0e07\\u0e2a\\u0e32\\u0e21\\u0e2a\\u0e35\\u0e48" },
{ NULL, NULL}
};
doTest(formatter, testData, TRUE);
}
delete formatter;
}
void
IntlTestRBNF::TestSwedishSpellout()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("sv"), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* testDataDefault[][2] = {
{ "101", "ett\\u00adhundra\\u00adett" },
{ "123", "ett\\u00adhundra\\u00adtjugo\\u00adtre" },
{ "1,001", "et\\u00adtusen ett" },
{ "1,100", "et\\u00adtusen ett\\u00adhundra" },
{ "1,101", "et\\u00adtusen ett\\u00adhundra\\u00adett" },
{ "1,234", "et\\u00adtusen tv\\u00e5\\u00adhundra\\u00adtrettio\\u00adfyra" },
{ "10,001", "tio\\u00adtusen ett" },
{ "11,000", "elva\\u00adtusen" },
{ "12,000", "tolv\\u00adtusen" },
{ "20,000", "tjugo\\u00adtusen" },
{ "21,000", "tjugo\\u00adet\\u00adtusen" },
{ "21,001", "tjugo\\u00adet\\u00adtusen ett" },
{ "200,000", "tv\\u00e5\\u00adhundra\\u00adtusen" },
{ "201,000", "tv\\u00e5\\u00adhundra\\u00adet\\u00adtusen" },
{ "200,200", "tv\\u00e5\\u00adhundra\\u00adtusen tv\\u00e5\\u00adhundra" },
{ "2,002,000", "tv\\u00e5 miljoner tv\\u00e5\\u00adtusen" },
{ "12,345,678", "tolv miljoner tre\\u00adhundra\\u00adfyrtio\\u00adfem\\u00adtusen sex\\u00adhundra\\u00adsjuttio\\u00ad\\u00e5tta" },
{ "123,456.789", "ett\\u00adhundra\\u00adtjugo\\u00adtre\\u00adtusen fyra\\u00adhundra\\u00adfemtio\\u00adsex komma sju \\u00e5tta nio" },
{ "-12,345.678", "minus tolv\\u00adtusen tre\\u00adhundra\\u00adfyrtio\\u00adfem komma sex sju \\u00e5tta" },
{ NULL, NULL }
};
doTest(formatter, testDataDefault, TRUE);
static const char* testDataNeutrum[][2] = {
{ "101", "ett\\u00adhundra\\u00aden" },
{ "1,001", "ettusen en" },
{ "1,101", "ettusen ett\\u00adhundra\\u00aden" },
{ "10,001", "tio\\u00adtusen en" },
{ "21,001", "tjugo\\u00aden\\u00adtusen en" },
{ NULL, NULL }
};
formatter->setDefaultRuleSet("%spellout-cardinal-neutre", status);
if (U_SUCCESS(status)) {
logln(" testing spellout-cardinal-neutre rules");
doTest(formatter, testDataNeutrum, TRUE);
}
else {
errln("Can't test spellout-cardinal-neutre rules");
}
static const char* testDataYear[][2] = {
{ "101", "ett\\u00adhundra\\u00adett" },
{ "900", "nio\\u00adhundra" },
{ "1,001", "et\\u00adtusen ett" },
{ "1,100", "elva\\u00adhundra" },
{ "1,101", "elva\\u00adhundra\\u00adett" },
{ "1,234", "tolv\\u00adhundra\\u00adtrettio\\u00adfyra" },
{ "2,001", "tjugo\\u00adhundra\\u00adett" },
{ "10,001", "tio\\u00adtusen ett" },
{ NULL, NULL }
};
formatter->setDefaultRuleSet("%spellout-numbering-year", status);
if (U_SUCCESS(status)) {
logln("testing year rules");
doTest(formatter, testDataYear, TRUE);
}
else {
errln("Can't test year rules");
}
}
delete formatter;
}
void
IntlTestRBNF::TestSmallValues()
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* formatter
= new RuleBasedNumberFormat(URBNF_SPELLOUT, Locale("en_US"), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
static const char* const testDataDefault[][2] = {
{ "0.001", "zero point zero zero one" },
{ "0.0001", "zero point zero zero zero one" },
{ "0.00001", "zero point zero zero zero zero one" },
{ "0.000001", "zero point zero zero zero zero zero one" },
{ "0.0000001", "zero point zero zero zero zero zero zero one" },
{ "0.00000001", "zero point zero zero zero zero zero zero zero one" },
{ "0.000000001", "zero point zero zero zero zero zero zero zero zero one" },
{ "0.0000000001", "zero point zero zero zero zero zero zero zero zero zero one" },
{ "0.00000000001", "zero point zero zero zero zero zero zero zero zero zero zero one" },
{ "0.000000000001", "zero point zero zero zero zero zero zero zero zero zero zero zero one" },
{ "0.0000000000001", "zero point zero zero zero zero zero zero zero zero zero zero zero zero one" },
{ "0.00000000000001", "zero point zero zero zero zero zero zero zero zero zero zero zero zero zero one" },
{ "0.000000000000001", "zero point zero zero zero zero zero zero zero zero zero zero zero zero zero zero one" },
{ "10,000,000.001", "ten million point zero zero one" },
{ "10,000,000.0001", "ten million point zero zero zero one" },
{ "10,000,000.00001", "ten million point zero zero zero zero one" },
{ "10,000,000.000001", "ten million point zero zero zero zero zero one" },
{ "10,000,000.0000001", "ten million point zero zero zero zero zero zero one" },
// { "10,000,000.00000001", "ten million point zero zero zero zero zero zero zero one" },
// { "10,000,000.000000002", "ten million point zero zero zero zero zero zero zero zero two" },
{ "10,000,000", "ten million" },
// { "1,234,567,890.0987654", "one billion, two hundred and thirty-four million, five hundred and sixty-seven thousand, eight hundred and ninety point zero nine eight seven six five four" },
// { "123,456,789.9876543", "one hundred and twenty-three million, four hundred and fifty-six thousand, seven hundred and eighty-nine point nine eight seven six five four three" },
// { "12,345,678.87654321", "twelve million, three hundred and forty-five thousand, six hundred and seventy-eight point eight seven six five four three two one" },
{ "1,234,567.7654321", "one million two hundred thirty-four thousand five hundred sixty-seven point seven six five four three two one" },
{ "123,456.654321", "one hundred twenty-three thousand four hundred fifty-six point six five four three two one" },
{ "12,345.54321", "twelve thousand three hundred forty-five point five four three two one" },
{ "1,234.4321", "one thousand two hundred thirty-four point four three two one" },
{ "123.321", "one hundred twenty-three point three two one" },
{ "0.0000000011754944", "zero point zero zero zero zero zero zero zero zero one one seven five four nine four four" },
{ "0.000001175494351", "zero point zero zero zero zero zero one one seven five four nine four three five one" },
{ NULL, NULL }
};
doTest(formatter, testDataDefault, TRUE);
delete formatter;
}
}
void
IntlTestRBNF::TestLocalizations(void)
{
int i;
UnicodeString rules("%main:0:no;1:some;100:a lot;1000:tons;\n"
"%other:0:nada;1:yah, some;100:plenty;1000:more'n you'll ever need");
UErrorCode status = U_ZERO_ERROR;
UParseError perror;
RuleBasedNumberFormat formatter(rules, perror, status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not construct formatter - %s", u_errorName(status));
} else {
{
static const char* const testData[][2] = {
{ "0", "nada" },
{ "5", "yah, some" },
{ "423", "plenty" },
{ "12345", "more'n you'll ever need" },
{ NULL, NULL }
};
doTest(&formatter, testData, FALSE);
}
{
UnicodeString loc("<<%main, %other>,<en, Main, Other>,<fr, leMain, leOther>,<de, 'das Main', 'etwas anderes'>>");
static const char* const testData[][2] = {
{ "0", "no" },
{ "5", "some" },
{ "423", "a lot" },
{ "12345", "tons" },
{ NULL, NULL }
};
RuleBasedNumberFormat formatter0(rules, loc, perror, status);
if (U_FAILURE(status)) {
errln("failed to build second formatter");
} else {
doTest(&formatter0, testData, FALSE);
{
// exercise localization info
Locale locale0("en__VALLEY@turkey=gobblegobble");
Locale locale1("de_DE_FOO");
Locale locale2("ja_JP");
UnicodeString name = formatter0.getRuleSetName(0);
if ( formatter0.getRuleSetDisplayName(0, locale0) == "Main"
&& formatter0.getRuleSetDisplayName(0, locale1) == "das Main"
&& formatter0.getRuleSetDisplayName(0, locale2) == "%main"
&& formatter0.getRuleSetDisplayName(name, locale0) == "Main"
&& formatter0.getRuleSetDisplayName(name, locale1) == "das Main"
&& formatter0.getRuleSetDisplayName(name, locale2) == "%main"){
logln("getRuleSetDisplayName tested");
}else {
errln("failed to getRuleSetDisplayName");
}
}
for (i = 0; i < formatter0.getNumberOfRuleSetDisplayNameLocales(); ++i) {
Locale locale = formatter0.getRuleSetDisplayNameLocale(i, status);
if (U_SUCCESS(status)) {
for (int j = 0; j < formatter0.getNumberOfRuleSetNames(); ++j) {
UnicodeString name = formatter0.getRuleSetName(j);
UnicodeString lname = formatter0.getRuleSetDisplayName(j, locale);
UnicodeString msg = locale.getName();
msg.append(": ");
msg.append(name);
msg.append(" = ");
msg.append(lname);
logln(msg);
}
}
}
}
}
{
static const char* goodLocs[] = {
"", // zero-length ok, same as providing no localization data
"<<>>", // no public rule sets ok
"<<%main>>", // no localizations ok
"<<%main,>,<en, Main,>>", // comma before close angle ok
"<<%main>,<en, ',<>\" '>>", // quotes everything until next quote
"<<%main>,<'en', \"it's ok\">>", // double quotes work too
" \n <\n <\n %main\n >\n , \t <\t en\t , \tfoo \t\t > \n\n > \n ", // rule whitespace ok
};
int32_t goodLocsLen = sizeof(goodLocs)/sizeof(goodLocs[0]);
static const char* badLocs[] = {
" ", // non-zero length
"<>", // empty array
"<", // unclosed outer array
"<<", // unclosed inner array
"<<,>>", // unexpected comma
"<<''>>", // empty string
" x<<%main>>", // first non space char not open angle bracket
"<%main>", // missing inner array
"<<%main %other>>", // elements missing separating commma (spaces must be quoted)
"<<%main><en, Main>>", // arrays missing separating comma
"<<%main>,<en, main, foo>>", // too many elements in locale data
"<<%main>,<en>>", // too few elements in locale data
"<<<%main>>>", // unexpected open angle
"<<%main<>>>", // unexpected open angle
"<<%main, %other>,<en,,>>", // implicit empty strings
"<<%main>,<en,''>>", // empty string
"<<%main>, < en, '>>", // unterminated quote
"<<%main>, < en, \"<>>", // unterminated quote
"<<%main\">>", // quote in string
"<<%main'>>", // quote in string
"<<%main<>>", // open angle in string
"<<%main>> x", // extra non-space text at end
};
int32_t badLocsLen = sizeof(badLocs)/sizeof(badLocs[0]);
for (i = 0; i < goodLocsLen; ++i) {
logln("[%d] '%s'", i, goodLocs[i]);
UErrorCode status = U_ZERO_ERROR;
UnicodeString loc(goodLocs[i]);
RuleBasedNumberFormat fmt(rules, loc, perror, status);
if (U_FAILURE(status)) {
errln("Failed parse of good localization string: '%s'", goodLocs[i]);
}
}
for (i = 0; i < badLocsLen; ++i) {
logln("[%d] '%s'", i, badLocs[i]);
UErrorCode status = U_ZERO_ERROR;
UnicodeString loc(badLocs[i]);
RuleBasedNumberFormat fmt(rules, loc, perror, status);
if (U_SUCCESS(status)) {
errln("Successful parse of bad localization string: '%s'", badLocs[i]);
}
}
}
}
}
void
IntlTestRBNF::TestAllLocales()
{
const char* names[] = {
" (spellout) ",
" (ordinal) ",
" (duration) "
};
double numbers[] = {45.678, 1, 2, 10, 11, 100, 110, 200, 1000, 1111, -1111};
// RBNF parse is extremely slow when lenient option is enabled.
// For non-exhaustive mode, we only test a few locales.
const char* parseLocales[] = {"en_US", "nl_NL", "be", NULL};
int32_t count = 0;
const Locale* locales = Locale::getAvailableLocales(count);
for (int i = 0; i < count; ++i) {
const Locale* loc = &locales[i];
UBool testParse = TRUE;
if (quick) {
testParse = FALSE;
for (int k = 0; parseLocales[k] != NULL; k++) {
if (strcmp(loc->getLanguage(), parseLocales[k]) == 0) {
testParse = TRUE;
break;
}
}
}
for (int j = 0; j < 3; ++j) {
UErrorCode status = U_ZERO_ERROR;
RuleBasedNumberFormat* f = new RuleBasedNumberFormat((URBNFRuleSetTag)j, *loc, status);
if (U_FAILURE(status)) {
errln(UnicodeString(loc->getName()) + names[j]
+ "ERROR could not instantiate -> " + u_errorName(status));
continue;
}
#if !UCONFIG_NO_COLLATION
for (unsigned int numidx = 0; numidx < sizeof(numbers)/sizeof(double); numidx++) {
double n = numbers[numidx];
UnicodeString str;
f->format(n, str);
logln(UnicodeString(loc->getName()) + names[j]
+ "success: " + n + " -> " + str);
if (testParse) {
// We do not validate the result in this test case,
// because there are cases which do not round trip by design.
Formattable num;
// regular parse
status = U_ZERO_ERROR;
f->setLenient(FALSE);
f->parse(str, num, status);
if (U_FAILURE(status)) {
//TODO: We need to fix parse problems - see #6895 / #6896
if (status == U_INVALID_FORMAT_ERROR) {
logln(UnicodeString(loc->getName()) + names[j]
+ "WARNING could not parse '" + str + "' -> " + u_errorName(status));
} else {
errln(UnicodeString(loc->getName()) + names[j]
+ "ERROR could not parse '" + str + "' -> " + u_errorName(status));
}
}
// lenient parse
status = U_ZERO_ERROR;
f->setLenient(TRUE);
f->parse(str, num, status);
if (U_FAILURE(status)) {
//TODO: We need to fix parse problems - see #6895 / #6896
if (status == U_INVALID_FORMAT_ERROR) {
logln(UnicodeString(loc->getName()) + names[j]
+ "WARNING could not parse(lenient) '" + str + "' -> " + u_errorName(status));
} else {
errln(UnicodeString(loc->getName()) + names[j]
+ "ERROR could not parse(lenient) '" + str + "' -> " + u_errorName(status));
}
}
}
}
#endif
delete f;
}
}
}
void
IntlTestRBNF::TestMultiplierSubstitution(void) {
UnicodeString rules("=#,##0=;1,000,000: <##0.###< million;");
UErrorCode status = U_ZERO_ERROR;
UParseError parse_error;
RuleBasedNumberFormat *rbnf =
new RuleBasedNumberFormat(rules, Locale::getUS(), parse_error, status);
if (U_SUCCESS(status)) {
UnicodeString res;
FieldPosition pos;
double n = 1234000.0;
rbnf->format(n, res, pos);
delete rbnf;
UnicodeString expected = UNICODE_STRING_SIMPLE("1.234 million");
if (expected != res) {
UnicodeString msg = "Expected: ";
msg.append(expected);
msg.append(" but got ");
msg.append(res);
errln(msg);
}
}
}
void
IntlTestRBNF::doTest(RuleBasedNumberFormat* formatter, const char* const testData[][2], UBool testParsing)
{
// man, error reporting would be easier with printf-style syntax for unicode string and formattable
UErrorCode status = U_ZERO_ERROR;
DecimalFormatSymbols dfs("en", status);
// NumberFormat* decFmt = NumberFormat::createInstance(Locale::getUS(), status);
DecimalFormat decFmt("#,###.################", dfs, status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not create NumberFormat - %s", u_errorName(status));
} else {
for (int i = 0; testData[i][0]; ++i) {
const char* numString = testData[i][0];
const char* expectedWords = testData[i][1];
log("[%i] %s = ", i, numString);
Formattable expectedNumber;
decFmt.parse(numString, expectedNumber, status);
if (U_FAILURE(status)) {
errln("FAIL: decFmt could not parse %s", numString);
break;
} else {
UnicodeString actualString;
FieldPosition pos;
formatter->format(expectedNumber, actualString/* , pos*/, status);
if (U_FAILURE(status)) {
UnicodeString msg = "Fail: formatter could not format ";
decFmt.format(expectedNumber, msg, status);
errln(msg);
break;
} else {
UnicodeString expectedString = UnicodeString(expectedWords, -1, US_INV).unescape();
if (actualString != expectedString) {
UnicodeString msg = "FAIL: check failed for ";
decFmt.format(expectedNumber, msg, status);
msg.append(", expected ");
msg.append(expectedString);
msg.append(" but got ");
msg.append(actualString);
errln(msg);
break;
} else {
logln(actualString);
if (testParsing) {
Formattable parsedNumber;
formatter->parse(actualString, parsedNumber, status);
if (U_FAILURE(status)) {
UnicodeString msg = "FAIL: formatter could not parse ";
msg.append(actualString);
msg.append(" status code: " );
msg.append(u_errorName(status));
errln(msg);
break;
} else {
if (parsedNumber != expectedNumber) {
UnicodeString msg = "FAIL: parse failed for ";
msg.append(actualString);
msg.append(", expected ");
decFmt.format(expectedNumber, msg, status);
msg.append(", but got ");
decFmt.format(parsedNumber, msg, status);
errln(msg);
break;
}
}
}
}
}
}
}
}
}
void
IntlTestRBNF::doLenientParseTest(RuleBasedNumberFormat* formatter, const char* testData[][2])
{
UErrorCode status = U_ZERO_ERROR;
NumberFormat* decFmt = NumberFormat::createInstance(Locale::getUS(), status);
if (U_FAILURE(status)) {
errcheckln(status, "FAIL: could not create NumberFormat - %s", u_errorName(status));
} else {
for (int i = 0; testData[i][0]; ++i) {
const char* spelledNumber = testData[i][0]; // spelled-out number
const char* asciiUSNumber = testData[i][1]; // number as ascii digits formatted for US locale
UnicodeString spelledNumberString = UnicodeString(spelledNumber).unescape();
Formattable actualNumber;
formatter->parse(spelledNumberString, actualNumber, status);
if (U_FAILURE(status)) {
UnicodeString msg = "FAIL: formatter could not parse ";
msg.append(spelledNumberString);
errln(msg);
break;
} else {
// I changed the logic of this test somewhat from Java-- instead of comparing the
// strings, I compare the Formattables. Hmmm, but the Formattables don't compare,
// so change it back.
UnicodeString asciiUSNumberString = asciiUSNumber;
Formattable expectedNumber;
decFmt->parse(asciiUSNumberString, expectedNumber, status);
if (U_FAILURE(status)) {
UnicodeString msg = "FAIL: decFmt could not parse ";
msg.append(asciiUSNumberString);
errln(msg);
break;
} else {
UnicodeString actualNumberString;
UnicodeString expectedNumberString;
decFmt->format(actualNumber, actualNumberString, status);
decFmt->format(expectedNumber, expectedNumberString, status);
if (actualNumberString != expectedNumberString) {
UnicodeString msg = "FAIL: parsing";
msg.append(asciiUSNumberString);
msg.append("\n");
msg.append(" lenient parse failed for ");
msg.append(spelledNumberString);
msg.append(", expected ");
msg.append(expectedNumberString);
msg.append(", but got ");
msg.append(actualNumberString);
errln(msg);
break;
}
}
}
}
delete decFmt;
}
}
/* U_HAVE_RBNF */
#else
void
IntlTestRBNF::TestRBNFDisabled() {
errln("*** RBNF currently disabled on this platform ***\n");
}
/* U_HAVE_RBNF */
#endif
#endif /* #if !UCONFIG_NO_FORMATTING */