/*
*******************************************************************************
* Copyright (C) 2007-2010, International Business Machines Corporation and
* others. All Rights Reserved.
*******************************************************************************
*
* File DTPTNGEN.CPP
*
*******************************************************************************
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_FORMATTING
#include "unicode/datefmt.h"
#include "unicode/decimfmt.h"
#include "unicode/dtfmtsym.h"
#include "unicode/dtptngen.h"
#include "unicode/msgfmt.h"
#include "unicode/smpdtfmt.h"
#include "unicode/udat.h"
#include "unicode/udatpg.h"
#include "unicode/uniset.h"
#include "unicode/uloc.h"
#include "unicode/ures.h"
#include "unicode/ustring.h"
#include "unicode/rep.h"
#include "cpputils.h"
#include "ucln_in.h"
#include "mutex.h"
#include "cmemory.h"
#include "cstring.h"
#include "locbased.h"
#include "gregoimp.h"
#include "hash.h"
#include "uresimp.h"
#include "dtptngen_impl.h"
#define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
#if U_CHARSET_FAMILY==U_EBCDIC_FAMILY
/**
* If we are on EBCDIC, use an iterator which will
* traverse the bundles in ASCII order.
*/
#define U_USE_ASCII_BUNDLE_ITERATOR
#define U_SORT_ASCII_BUNDLE_ITERATOR
#endif
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
#include "unicode/ustring.h"
#include "uarrsort.h"
struct UResAEntry {
UChar *key;
UResourceBundle *item;
};
struct UResourceBundleAIterator {
UResourceBundle *bund;
UResAEntry *entries;
int32_t num;
int32_t cursor;
};
/* Must be C linkage to pass function pointer to the sort function */
#if !defined (OS390) && !defined (OS400)
extern "C"
#endif
static int32_t U_CALLCONV
ures_a_codepointSort(const void *context, const void *left, const void *right) {
//CompareContext *cmp=(CompareContext *)context;
return u_strcmp(((const UResAEntry *)left)->key,
((const UResAEntry *)right)->key);
}
static void ures_a_open(UResourceBundleAIterator *aiter, UResourceBundle *bund, UErrorCode *status) {
if(U_FAILURE(*status)) {
return;
}
aiter->bund = bund;
aiter->num = ures_getSize(aiter->bund);
aiter->cursor = 0;
#if !defined(U_SORT_ASCII_BUNDLE_ITERATOR)
aiter->entries = NULL;
#else
aiter->entries = (UResAEntry*)uprv_malloc(sizeof(UResAEntry)*aiter->num);
for(int i=0;i<aiter->num;i++) {
aiter->entries[i].item = ures_getByIndex(aiter->bund, i, NULL, status);
const char *akey = ures_getKey(aiter->entries[i].item);
int32_t len = uprv_strlen(akey)+1;
aiter->entries[i].key = (UChar*)uprv_malloc(len*sizeof(UChar));
u_charsToUChars(akey, aiter->entries[i].key, len);
}
uprv_sortArray(aiter->entries, aiter->num, sizeof(UResAEntry), ures_a_codepointSort, NULL, TRUE, status);
#endif
}
static void ures_a_close(UResourceBundleAIterator *aiter) {
#if defined(U_SORT_ASCII_BUNDLE_ITERATOR)
for(int i=0;i<aiter->num;i++) {
uprv_free(aiter->entries[i].key);
ures_close(aiter->entries[i].item);
}
#endif
}
static const UChar *ures_a_getNextString(UResourceBundleAIterator *aiter, int32_t *len, const char **key, UErrorCode *err) {
#if !defined(U_SORT_ASCII_BUNDLE_ITERATOR)
return ures_getNextString(aiter->bund, len, key, err);
#else
if(U_FAILURE(*err)) return NULL;
UResourceBundle *item = aiter->entries[aiter->cursor].item;
const UChar* ret = ures_getString(item, len, err);
*key = ures_getKey(item);
aiter->cursor++;
return ret;
#endif
}
#endif
U_NAMESPACE_BEGIN
// *****************************************************************************
// class DateTimePatternGenerator
// *****************************************************************************
static const UChar Canonical_Items[] = {
// GyQMwWedDFHmsSv
CAP_G, LOW_Y, CAP_Q, CAP_M, LOW_W, CAP_W, LOW_E, LOW_D, CAP_D, CAP_F,
CAP_H, LOW_M, LOW_S, CAP_S, LOW_V, 0
};
static const dtTypeElem dtTypes[] = {
// patternChar, field, type, minLen, weight
{CAP_G, UDATPG_ERA_FIELD, DT_SHORT, 1, 3,},
{CAP_G, UDATPG_ERA_FIELD, DT_LONG, 4, 0},
{LOW_Y, UDATPG_YEAR_FIELD, DT_NUMERIC, 1, 20},
{CAP_Y, UDATPG_YEAR_FIELD, DT_NUMERIC + DT_DELTA, 1, 20},
{LOW_U, UDATPG_YEAR_FIELD, DT_NUMERIC + 2*DT_DELTA, 1, 20},
{CAP_Q, UDATPG_QUARTER_FIELD, DT_NUMERIC, 1, 2},
{CAP_Q, UDATPG_QUARTER_FIELD, DT_SHORT, 3, 0},
{CAP_Q, UDATPG_QUARTER_FIELD, DT_LONG, 4, 0},
{LOW_Q, UDATPG_QUARTER_FIELD, DT_NUMERIC + DT_DELTA, 1, 2},
{LOW_Q, UDATPG_QUARTER_FIELD, DT_SHORT + DT_DELTA, 3, 0},
{LOW_Q, UDATPG_QUARTER_FIELD, DT_LONG + DT_DELTA, 4, 0},
{CAP_M, UDATPG_MONTH_FIELD, DT_NUMERIC, 1, 2},
{CAP_M, UDATPG_MONTH_FIELD, DT_SHORT, 3, 0},
{CAP_M, UDATPG_MONTH_FIELD, DT_LONG, 4, 0},
{CAP_M, UDATPG_MONTH_FIELD, DT_NARROW, 5, 0},
{CAP_L, UDATPG_MONTH_FIELD, DT_NUMERIC + DT_DELTA, 1, 2},
{CAP_L, UDATPG_MONTH_FIELD, DT_SHORT - DT_DELTA, 3, 0},
{CAP_L, UDATPG_MONTH_FIELD, DT_LONG - DT_DELTA, 4, 0},
{CAP_L, UDATPG_MONTH_FIELD, DT_NARROW - DT_DELTA, 5, 0},
{LOW_W, UDATPG_WEEK_OF_YEAR_FIELD, DT_NUMERIC, 1, 2},
{CAP_W, UDATPG_WEEK_OF_MONTH_FIELD, DT_NUMERIC + DT_DELTA, 1, 0},
{LOW_E, UDATPG_WEEKDAY_FIELD, DT_NUMERIC + DT_DELTA, 1, 2},
{LOW_E, UDATPG_WEEKDAY_FIELD, DT_SHORT - DT_DELTA, 3, 0},
{LOW_E, UDATPG_WEEKDAY_FIELD, DT_LONG - DT_DELTA, 4, 0},
{LOW_E, UDATPG_WEEKDAY_FIELD, DT_NARROW - DT_DELTA, 5, 0},
{CAP_E, UDATPG_WEEKDAY_FIELD, DT_SHORT, 1, 3},
{CAP_E, UDATPG_WEEKDAY_FIELD, DT_LONG, 4, 0},
{CAP_E, UDATPG_WEEKDAY_FIELD, DT_NARROW, 5, 0},
{LOW_C, UDATPG_WEEKDAY_FIELD, DT_NUMERIC + 2*DT_DELTA, 1, 2},
{LOW_C, UDATPG_WEEKDAY_FIELD, DT_SHORT - 2*DT_DELTA, 3, 0},
{LOW_C, UDATPG_WEEKDAY_FIELD, DT_LONG - 2*DT_DELTA, 4, 0},
{LOW_C, UDATPG_WEEKDAY_FIELD, DT_NARROW - 2*DT_DELTA, 5, 0},
{LOW_D, UDATPG_DAY_FIELD, DT_NUMERIC, 1, 2},
{CAP_D, UDATPG_DAY_OF_YEAR_FIELD, DT_NUMERIC + DT_DELTA, 1, 3},
{CAP_F, UDATPG_DAY_OF_WEEK_IN_MONTH_FIELD, DT_NUMERIC + 2*DT_DELTA, 1, 0},
{LOW_G, UDATPG_DAY_FIELD, DT_NUMERIC + 3*DT_DELTA, 1, 20}, // really internal use, so we don't care
{LOW_A, UDATPG_DAYPERIOD_FIELD, DT_SHORT, 1, 0},
{CAP_H, UDATPG_HOUR_FIELD, DT_NUMERIC + 10*DT_DELTA, 1, 2}, // 24 hour
{LOW_K, UDATPG_HOUR_FIELD, DT_NUMERIC + 11*DT_DELTA, 1, 2},
{LOW_H, UDATPG_HOUR_FIELD, DT_NUMERIC, 1, 2}, // 12 hour
{LOW_K, UDATPG_HOUR_FIELD, DT_NUMERIC + DT_DELTA, 1, 2},
{LOW_M, UDATPG_MINUTE_FIELD, DT_NUMERIC, 1, 2},
{LOW_S, UDATPG_SECOND_FIELD, DT_NUMERIC, 1, 2},
{CAP_S, UDATPG_FRACTIONAL_SECOND_FIELD, DT_NUMERIC + DT_DELTA, 1, 1000},
{CAP_A, UDATPG_SECOND_FIELD, DT_NUMERIC + 2*DT_DELTA, 1, 1000},
{LOW_V, UDATPG_ZONE_FIELD, DT_SHORT - 2*DT_DELTA, 1, 0},
{LOW_V, UDATPG_ZONE_FIELD, DT_LONG - 2*DT_DELTA, 4, 0},
{LOW_Z, UDATPG_ZONE_FIELD, DT_SHORT, 1, 3},
{LOW_Z, UDATPG_ZONE_FIELD, DT_LONG, 4, 0},
{CAP_Z, UDATPG_ZONE_FIELD, DT_SHORT - DT_DELTA, 1, 3},
{CAP_Z, UDATPG_ZONE_FIELD, DT_LONG - DT_DELTA, 4, 0},
{CAP_V, UDATPG_ZONE_FIELD, DT_SHORT - DT_DELTA, 1, 3},
{CAP_V, UDATPG_ZONE_FIELD, DT_LONG - DT_DELTA, 4, 0},
{0, UDATPG_FIELD_COUNT, 0, 0, 0} , // last row of dtTypes[]
};
static const char* const CLDR_FIELD_APPEND[] = {
"Era", "Year", "Quarter", "Month", "Week", "*", "Day-Of-Week", "Day", "*", "*", "*",
"Hour", "Minute", "Second", "*", "Timezone"
};
static const char* const CLDR_FIELD_NAME[] = {
"era", "year", "quarter", "month", "week", "*", "weekday", "day", "*", "*", "dayperiod",
"hour", "minute", "second", "*", "zone"
};
static const char* const Resource_Fields[] = {
"day", "dayperiod", "era", "hour", "minute", "month", "second", "week",
"weekday", "year", "zone", "quarter" };
// For appendItems
static const UChar UDATPG_ItemFormat[]= {0x7B, 0x30, 0x7D, 0x20, 0x251C, 0x7B, 0x32, 0x7D, 0x3A,
0x20, 0x7B, 0x31, 0x7D, 0x2524, 0}; // {0} \u251C{2}: {1}\u2524
static const UChar repeatedPatterns[6]={CAP_G, CAP_E, LOW_Z, LOW_V, CAP_Q, 0}; // "GEzvQ"
static const char DT_DateTimePatternsTag[]="DateTimePatterns";
static const char DT_DateTimeCalendarTag[]="calendar";
static const char DT_DateTimeGregorianTag[]="gregorian";
static const char DT_DateTimeAppendItemsTag[]="appendItems";
static const char DT_DateTimeFieldsTag[]="fields";
static const char DT_DateTimeAvailableFormatsTag[]="availableFormats";
//static const UnicodeString repeatedPattern=UnicodeString(repeatedPatterns);
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(DateTimePatternGenerator)
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(DTSkeletonEnumeration)
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(DTRedundantEnumeration)
DateTimePatternGenerator* U_EXPORT2
DateTimePatternGenerator::createInstance(UErrorCode& status) {
return createInstance(Locale::getDefault(), status);
}
DateTimePatternGenerator* U_EXPORT2
DateTimePatternGenerator::createInstance(const Locale& locale, UErrorCode& status) {
DateTimePatternGenerator *result = new DateTimePatternGenerator(locale, status);
if (result == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
if (U_FAILURE(status)) {
delete result;
result = NULL;
}
return result;
}
DateTimePatternGenerator* U_EXPORT2
DateTimePatternGenerator::createEmptyInstance(UErrorCode& status) {
DateTimePatternGenerator *result = new DateTimePatternGenerator(status);
if (result == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
if (U_FAILURE(status)) {
delete result;
result = NULL;
}
return result;
}
DateTimePatternGenerator::DateTimePatternGenerator(UErrorCode &status) :
skipMatcher(NULL),
fAvailableFormatKeyHash(NULL)
{
fp = new FormatParser();
dtMatcher = new DateTimeMatcher();
distanceInfo = new DistanceInfo();
patternMap = new PatternMap();
if (fp == NULL || dtMatcher == NULL || distanceInfo == NULL || patternMap == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
}
DateTimePatternGenerator::DateTimePatternGenerator(const Locale& locale, UErrorCode &status) :
skipMatcher(NULL),
fAvailableFormatKeyHash(NULL)
{
fp = new FormatParser();
dtMatcher = new DateTimeMatcher();
distanceInfo = new DistanceInfo();
patternMap = new PatternMap();
if (fp == NULL || dtMatcher == NULL || distanceInfo == NULL || patternMap == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
else {
initData(locale, status);
}
}
DateTimePatternGenerator::DateTimePatternGenerator(const DateTimePatternGenerator& other) :
UObject(),
skipMatcher(NULL),
fAvailableFormatKeyHash(NULL)
{
fp = new FormatParser();
dtMatcher = new DateTimeMatcher();
distanceInfo = new DistanceInfo();
patternMap = new PatternMap();
*this=other;
}
DateTimePatternGenerator&
DateTimePatternGenerator::operator=(const DateTimePatternGenerator& other) {
pLocale = other.pLocale;
fDefaultHourFormatChar = other.fDefaultHourFormatChar;
*fp = *(other.fp);
dtMatcher->copyFrom(other.dtMatcher->skeleton);
*distanceInfo = *(other.distanceInfo);
dateTimeFormat = other.dateTimeFormat;
decimal = other.decimal;
// NUL-terminate for the C API.
dateTimeFormat.getTerminatedBuffer();
decimal.getTerminatedBuffer();
delete skipMatcher;
if ( other.skipMatcher == NULL ) {
skipMatcher = NULL;
}
else {
skipMatcher = new DateTimeMatcher(*other.skipMatcher);
}
for (int32_t i=0; i< UDATPG_FIELD_COUNT; ++i ) {
appendItemFormats[i] = other.appendItemFormats[i];
appendItemNames[i] = other.appendItemNames[i];
// NUL-terminate for the C API.
appendItemFormats[i].getTerminatedBuffer();
appendItemNames[i].getTerminatedBuffer();
}
UErrorCode status = U_ZERO_ERROR;
patternMap->copyFrom(*other.patternMap, status);
copyHashtable(other.fAvailableFormatKeyHash, status);
return *this;
}
UBool
DateTimePatternGenerator::operator==(const DateTimePatternGenerator& other) const {
if (this == &other) {
return TRUE;
}
if ((pLocale==other.pLocale) && (patternMap->equals(*other.patternMap)) &&
(dateTimeFormat==other.dateTimeFormat) && (decimal==other.decimal)) {
for ( int32_t i=0 ; i<UDATPG_FIELD_COUNT; ++i ) {
if ((appendItemFormats[i] != other.appendItemFormats[i]) ||
(appendItemNames[i] != other.appendItemNames[i]) ) {
return FALSE;
}
}
return TRUE;
}
else {
return FALSE;
}
}
UBool
DateTimePatternGenerator::operator!=(const DateTimePatternGenerator& other) const {
return !operator==(other);
}
DateTimePatternGenerator::~DateTimePatternGenerator() {
if (fAvailableFormatKeyHash!=NULL) {
delete fAvailableFormatKeyHash;
}
if (fp != NULL) delete fp;
if (dtMatcher != NULL) delete dtMatcher;
if (distanceInfo != NULL) delete distanceInfo;
if (patternMap != NULL) delete patternMap;
if (skipMatcher != NULL) delete skipMatcher;
}
void
DateTimePatternGenerator::initData(const Locale& locale, UErrorCode &status) {
//const char *baseLangName = locale.getBaseName(); // unused
skipMatcher = NULL;
fAvailableFormatKeyHash=NULL;
addCanonicalItems();
addICUPatterns(locale, status);
if (U_FAILURE(status)) {
return;
}
addCLDRData(locale, status);
setDateTimeFromCalendar(locale, status);
setDecimalSymbols(locale, status);
} // DateTimePatternGenerator::initData
UnicodeString
DateTimePatternGenerator::getSkeleton(const UnicodeString& pattern, UErrorCode&
/*status*/) {
dtMatcher->set(pattern, fp);
return dtMatcher->getSkeletonPtr()->getSkeleton();
}
UnicodeString
DateTimePatternGenerator::getBaseSkeleton(const UnicodeString& pattern, UErrorCode& /*status*/) {
dtMatcher->set(pattern, fp);
return dtMatcher->getSkeletonPtr()->getBaseSkeleton();
}
void
DateTimePatternGenerator::addICUPatterns(const Locale& locale, UErrorCode& status) {
UnicodeString dfPattern;
UnicodeString conflictingString;
UDateTimePatternConflict conflictingStatus;
DateFormat* df;
if (U_FAILURE(status)) {
return;
}
// Load with ICU patterns
for (int32_t i=DateFormat::kFull; i<=DateFormat::kShort; i++) {
DateFormat::EStyle style = (DateFormat::EStyle)i;
df = DateFormat::createDateInstance(style, locale);
if (df != NULL && df->getDynamicClassID() == SimpleDateFormat::getStaticClassID()) {
SimpleDateFormat* sdf = (SimpleDateFormat*)df;
conflictingStatus = addPattern(sdf->toPattern(dfPattern), FALSE, conflictingString, status);
}
// TODO Maybe we should return an error when the date format isn't simple.
delete df;
if (U_FAILURE(status)) {
return;
}
df = DateFormat::createTimeInstance(style, locale);
if (df != NULL && df->getDynamicClassID() == SimpleDateFormat::getStaticClassID()) {
SimpleDateFormat* sdf = (SimpleDateFormat*)df;
conflictingStatus = addPattern(sdf->toPattern(dfPattern), FALSE, conflictingString, status);
// HACK for hh:ss
if ( i==DateFormat::kMedium ) {
hackPattern = dfPattern;
}
}
// TODO Maybe we should return an error when the date format isn't simple.
delete df;
if (U_FAILURE(status)) {
return;
}
}
}
void
DateTimePatternGenerator::hackTimes(const UnicodeString& hackPattern, UErrorCode& status) {
UDateTimePatternConflict conflictingStatus;
UnicodeString conflictingString;
fp->set(hackPattern);
UnicodeString mmss;
UBool gotMm=FALSE;
for (int32_t i=0; i<fp->itemNumber; ++i) {
UnicodeString field = fp->items[i];
if ( fp->isQuoteLiteral(field) ) {
if ( gotMm ) {
UnicodeString quoteLiteral;
fp->getQuoteLiteral(quoteLiteral, &i);
mmss += quoteLiteral;
}
}
else {
if (fp->isPatternSeparator(field) && gotMm) {
mmss+=field;
}
else {
UChar ch=field.charAt(0);
if (ch==LOW_M) {
gotMm=TRUE;
mmss+=field;
}
else {
if (ch==LOW_S) {
if (!gotMm) {
break;
}
mmss+= field;
conflictingStatus = addPattern(mmss, FALSE, conflictingString, status);
break;
}
else {
if (gotMm || ch==LOW_Z || ch==CAP_Z || ch==LOW_V || ch==CAP_V) {
break;
}
}
}
}
}
}
}
#define ULOC_LOCALE_IDENTIFIER_CAPACITY (ULOC_FULLNAME_CAPACITY + 1 + ULOC_KEYWORD_AND_VALUES_CAPACITY)
static const UChar hourFormatChars[] = { CAP_H, LOW_H, CAP_K, LOW_K, 0 }; // HhKk, the hour format characters
void
DateTimePatternGenerator::addCLDRData(const Locale& locale, UErrorCode& err) {
UResourceBundle *rb, *calTypeBundle, *calBundle;
UResourceBundle *patBundle, *fieldBundle, *fBundle;
UnicodeString rbPattern, value, field;
UnicodeString conflictingPattern;
UDateTimePatternConflict conflictingStatus;
const char *key=NULL;
int32_t i;
UnicodeString defaultItemFormat(TRUE, UDATPG_ItemFormat, LENGTHOF(UDATPG_ItemFormat)-1); // Read-only alias.
err = U_ZERO_ERROR;
fDefaultHourFormatChar = 0;
for (i=0; i<UDATPG_FIELD_COUNT; ++i ) {
appendItemNames[i]=CAP_F;
if (i<10) {
appendItemNames[i]+=(UChar)(i+0x30);
}
else {
appendItemNames[i]+=(UChar)0x31;
appendItemNames[i]+=(UChar)(i-10 + 0x30);
}
// NUL-terminate for the C API.
appendItemNames[i].getTerminatedBuffer();
}
rb = ures_open(NULL, locale.getName(), &err);
if (rb == NULL || U_FAILURE(err)) {
return;
}
const char *curLocaleName=ures_getLocaleByType(rb, ULOC_ACTUAL_LOCALE, &err);
const char * calendarTypeToUse = DT_DateTimeGregorianTag; // initial default
char calendarType[ULOC_KEYWORDS_CAPACITY]; // to be filled in with the type to use, if all goes well
if ( U_SUCCESS(err) ) {
char localeWithCalendarKey[ULOC_LOCALE_IDENTIFIER_CAPACITY];
// obtain a locale that always has the calendar key value that should be used
(void)ures_getFunctionalEquivalent(localeWithCalendarKey, ULOC_LOCALE_IDENTIFIER_CAPACITY, NULL,
"calendar", "calendar", locale.getName(), NULL, FALSE, &err);
localeWithCalendarKey[ULOC_LOCALE_IDENTIFIER_CAPACITY-1] = 0; // ensure null termination
// now get the calendar key value from that locale
int32_t calendarTypeLen = uloc_getKeywordValue(localeWithCalendarKey, "calendar", calendarType, ULOC_KEYWORDS_CAPACITY, &err);
if (U_SUCCESS(err) && calendarTypeLen < ULOC_KEYWORDS_CAPACITY) {
calendarTypeToUse = calendarType;
}
err = U_ZERO_ERROR;
}
calBundle = ures_getByKeyWithFallback(rb, DT_DateTimeCalendarTag, NULL, &err);
calTypeBundle = ures_getByKeyWithFallback(calBundle, calendarTypeToUse, NULL, &err);
key=NULL;
int32_t dtCount=0;
patBundle = ures_getByKeyWithFallback(calTypeBundle, DT_DateTimePatternsTag, NULL, &err);
while (U_SUCCESS(err)) {
rbPattern = ures_getNextUnicodeString(patBundle, &key, &err);
dtCount++;
if (rbPattern.length()==0 ) {
break; // no more pattern
}
else {
if (dtCount==9) {
setDateTimeFormat(rbPattern);
} else if (dtCount==4) { // short time format
// set fDefaultHourFormatChar to the hour format character from this pattern
int32_t tfIdx, tfLen = rbPattern.length();
UBool ignoreChars = FALSE;
for (tfIdx = 0; tfIdx < tfLen; tfIdx++) {
UChar tfChar = rbPattern.charAt(tfIdx);
if ( tfChar == SINGLE_QUOTE ) {
ignoreChars = !ignoreChars; // toggle (handle quoted literals & '' for single quote)
} else if ( !ignoreChars && u_strchr(hourFormatChars, tfChar) != NULL ) {
fDefaultHourFormatChar = tfChar;
break;
}
}
}
}
}
ures_close(patBundle);
err = U_ZERO_ERROR;
patBundle = ures_getByKeyWithFallback(calTypeBundle, DT_DateTimeAppendItemsTag, NULL, &err);
key=NULL;
UnicodeString itemKey;
while (U_SUCCESS(err)) {
rbPattern = ures_getNextUnicodeString(patBundle, &key, &err);
if (rbPattern.length()==0 ) {
break; // no more pattern
}
else {
setAppendItemFormat(getAppendFormatNumber(key), rbPattern);
}
}
ures_close(patBundle);
key=NULL;
err = U_ZERO_ERROR;
fBundle = ures_getByKeyWithFallback(calTypeBundle, DT_DateTimeFieldsTag, NULL, &err);
for (i=0; i<MAX_RESOURCE_FIELD; ++i) {
err = U_ZERO_ERROR;
patBundle = ures_getByKeyWithFallback(fBundle, Resource_Fields[i], NULL, &err);
fieldBundle = ures_getByKeyWithFallback(patBundle, "dn", NULL, &err);
rbPattern = ures_getNextUnicodeString(fieldBundle, &key, &err);
ures_close(fieldBundle);
ures_close(patBundle);
if (rbPattern.length()==0 ) {
continue;
}
else {
setAppendItemName(getAppendNameNumber(Resource_Fields[i]), rbPattern);
}
}
ures_close(fBundle);
// add available formats
err = U_ZERO_ERROR;
initHashtable(err);
patBundle = ures_getByKeyWithFallback(calTypeBundle, DT_DateTimeAvailableFormatsTag, NULL, &err);
if (U_SUCCESS(err)) {
int32_t numberKeys = ures_getSize(patBundle);
int32_t len;
const UChar *retPattern;
key=NULL;
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
UResourceBundleAIterator aiter;
ures_a_open(&aiter, patBundle, &err);
#endif
for(i=0; i<numberKeys; ++i) {
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
retPattern=ures_a_getNextString(&aiter, &len, &key, &err);
#else
retPattern=ures_getNextString(patBundle, &len, &key, &err);
#endif
UnicodeString format=UnicodeString(retPattern);
UnicodeString retKey=UnicodeString(key, -1, US_INV);
setAvailableFormat(retKey, err);
// Add pattern with its associated skeleton. Override any duplicate derived from std patterns,
// but not a previous availableFormats entry:
conflictingStatus = addPatternWithSkeleton(format, &retKey, TRUE, conflictingPattern, err);
}
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
ures_a_close(&aiter);
#endif
}
ures_close(patBundle);
ures_close(calTypeBundle);
ures_close(calBundle);
ures_close(rb);
err = U_ZERO_ERROR;
char parentLocale[50];
int32_t localeNameLen=0;
uprv_strcpy(parentLocale, curLocaleName);
while((localeNameLen=uloc_getParent(parentLocale, parentLocale, 50, &err))>=0 ) {
rb = ures_open(NULL, parentLocale, &err);
curLocaleName=ures_getLocaleByType(rb, ULOC_ACTUAL_LOCALE, &err);
uprv_strcpy(parentLocale, curLocaleName);
calBundle = ures_getByKey(rb, DT_DateTimeCalendarTag, NULL, &err);
calTypeBundle = ures_getByKey(calBundle, calendarTypeToUse, NULL, &err);
patBundle = ures_getByKeyWithFallback(calTypeBundle, DT_DateTimeAvailableFormatsTag, NULL, &err);
if (U_SUCCESS(err)) {
int32_t numberKeys = ures_getSize(patBundle);
int32_t len;
const UChar *retPattern;
key=NULL;
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
UResourceBundleAIterator aiter;
ures_a_open(&aiter, patBundle, &err);
#endif
for(i=0; i<numberKeys; ++i) {
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
retPattern=ures_a_getNextString(&aiter, &len, &key, &err);
#else
retPattern=ures_getNextString(patBundle, &len, &key, &err);
#endif
UnicodeString format=UnicodeString(retPattern);
UnicodeString retKey=UnicodeString(key, -1, US_INV);
if ( !isAvailableFormatSet(retKey) ) {
setAvailableFormat(retKey, err);
// Add pattern with its associated skeleton. Override any duplicate derived from std patterns,
// but not a previous availableFormats entry:
conflictingStatus = addPatternWithSkeleton(format, &retKey, TRUE, conflictingPattern, err);
}
}
#if defined(U_USE_ASCII_BUNDLE_ITERATOR)
ures_a_close(&aiter);
#endif
}
err = U_ZERO_ERROR; // reset; if this locale lacks the necessary data, need to keep checking up to root.
ures_close(patBundle);
ures_close(calTypeBundle);
ures_close(calBundle);
ures_close(rb);
if (localeNameLen==0) {
break;
}
}
if (hackPattern.length()>0) {
hackTimes(hackPattern, err);
}
}
void
DateTimePatternGenerator::initHashtable(UErrorCode& err) {
if (fAvailableFormatKeyHash!=NULL) {
return;
}
if ((fAvailableFormatKeyHash = new Hashtable(FALSE, err))==NULL) {
err=U_MEMORY_ALLOCATION_ERROR;
return;
}
}
void
DateTimePatternGenerator::setAppendItemFormat(UDateTimePatternField field, const UnicodeString& value) {
appendItemFormats[field] = value;
// NUL-terminate for the C API.
appendItemFormats[field].getTerminatedBuffer();
}
const UnicodeString&
DateTimePatternGenerator::getAppendItemFormat(UDateTimePatternField field) const {
return appendItemFormats[field];
}
void
DateTimePatternGenerator::setAppendItemName(UDateTimePatternField field, const UnicodeString& value) {
appendItemNames[field] = value;
// NUL-terminate for the C API.
appendItemNames[field].getTerminatedBuffer();
}
const UnicodeString&
DateTimePatternGenerator:: getAppendItemName(UDateTimePatternField field) const {
return appendItemNames[field];
}
void
DateTimePatternGenerator::getAppendName(UDateTimePatternField field, UnicodeString& value) {
value = SINGLE_QUOTE;
value += appendItemNames[field];
value += SINGLE_QUOTE;
}
UnicodeString
DateTimePatternGenerator::getBestPattern(const UnicodeString& patternForm, UErrorCode& status) {
return getBestPattern(patternForm, UDATPG_MATCH_NO_OPTIONS, status);
}
UnicodeString
DateTimePatternGenerator::getBestPattern(const UnicodeString& patternForm, UDateTimePatternMatchOptions options, UErrorCode& status) {
const UnicodeString *bestPattern=NULL;
UnicodeString dtFormat;
UnicodeString resultPattern;
int32_t dateMask=(1<<UDATPG_DAYPERIOD_FIELD) - 1;
int32_t timeMask=(1<<UDATPG_FIELD_COUNT) - 1 - dateMask;
UnicodeString patternFormCopy = UnicodeString(patternForm);
patternFormCopy.findAndReplace(UnicodeString(LOW_J), UnicodeString(fDefaultHourFormatChar));
resultPattern.remove();
dtMatcher->set(patternFormCopy, fp);
const PtnSkeleton* specifiedSkeleton=NULL;
bestPattern=getBestRaw(*dtMatcher, -1, distanceInfo, &specifiedSkeleton);
if ( distanceInfo->missingFieldMask==0 && distanceInfo->extraFieldMask==0 ) {
resultPattern = adjustFieldTypes(*bestPattern, specifiedSkeleton, FALSE, options);
return resultPattern;
}
int32_t neededFields = dtMatcher->getFieldMask();
UnicodeString datePattern=getBestAppending(neededFields & dateMask, options);
UnicodeString timePattern=getBestAppending(neededFields & timeMask, options);
if (datePattern.length()==0) {
if (timePattern.length()==0) {
resultPattern.remove();
}
else {
return timePattern;
}
}
if (timePattern.length()==0) {
return datePattern;
}
resultPattern.remove();
status = U_ZERO_ERROR;
dtFormat=getDateTimeFormat();
Formattable dateTimeObject[] = { timePattern, datePattern };
resultPattern = MessageFormat::format(dtFormat, dateTimeObject, 2, resultPattern, status );
return resultPattern;
}
UnicodeString
DateTimePatternGenerator::replaceFieldTypes(const UnicodeString& pattern,
const UnicodeString& skeleton,
UErrorCode& status) {
return replaceFieldTypes(pattern, skeleton, UDATPG_MATCH_NO_OPTIONS, status);
}
UnicodeString
DateTimePatternGenerator::replaceFieldTypes(const UnicodeString& pattern,
const UnicodeString& skeleton,
UDateTimePatternMatchOptions options,
UErrorCode& /*status*/) {
dtMatcher->set(skeleton, fp);
UnicodeString result = adjustFieldTypes(pattern, NULL, FALSE, options);
return result;
}
void
DateTimePatternGenerator::setDecimal(const UnicodeString& newDecimal) {
this->decimal = newDecimal;
// NUL-terminate for the C API.
this->decimal.getTerminatedBuffer();
}
const UnicodeString&
DateTimePatternGenerator::getDecimal() const {
return decimal;
}
void
DateTimePatternGenerator::addCanonicalItems() {
UnicodeString conflictingPattern;
UDateTimePatternConflict conflictingStatus;
UErrorCode status = U_ZERO_ERROR;
for (int32_t i=0; i<UDATPG_FIELD_COUNT; i++) {
conflictingStatus = addPattern(UnicodeString(Canonical_Items[i]), FALSE, conflictingPattern, status);
}
}
void
DateTimePatternGenerator::setDateTimeFormat(const UnicodeString& dtFormat) {
dateTimeFormat = dtFormat;
// NUL-terminate for the C API.
dateTimeFormat.getTerminatedBuffer();
}
const UnicodeString&
DateTimePatternGenerator::getDateTimeFormat() const {
return dateTimeFormat;
}
void
DateTimePatternGenerator::setDateTimeFromCalendar(const Locale& locale, UErrorCode& status) {
const UChar *resStr;
int32_t resStrLen = 0;
Calendar* fCalendar = Calendar::createInstance(locale, status);
CalendarData calData(locale, fCalendar?fCalendar->getType():NULL, status);
UResourceBundle *dateTimePatterns = calData.getByKey(DT_DateTimePatternsTag, status);
if (U_FAILURE(status)) return;
if (ures_getSize(dateTimePatterns) <= DateFormat::kDateTime)
{
status = U_INVALID_FORMAT_ERROR;
return;
}
resStr = ures_getStringByIndex(dateTimePatterns, (int32_t)DateFormat::kDateTime, &resStrLen, &status);
setDateTimeFormat(UnicodeString(TRUE, resStr, resStrLen));
delete fCalendar;
}
void
DateTimePatternGenerator::setDecimalSymbols(const Locale& locale, UErrorCode& status) {
DecimalFormatSymbols dfs = DecimalFormatSymbols(locale, status);
if(U_SUCCESS(status)) {
decimal = dfs.getSymbol(DecimalFormatSymbols::kDecimalSeparatorSymbol);
// NUL-terminate for the C API.
decimal.getTerminatedBuffer();
}
}
UDateTimePatternConflict
DateTimePatternGenerator::addPattern(
const UnicodeString& pattern,
UBool override,
UnicodeString &conflictingPattern,
UErrorCode& status)
{
return addPatternWithSkeleton(pattern, NULL, override, conflictingPattern, status);
}
// For DateTimePatternGenerator::addPatternWithSkeleton -
// If skeletonToUse is specified, then an availableFormats entry is being added. In this case:
// 1. We pass that skeleton to matcher.set instead of having it derive a skeleton from the pattern.
// 2. If the new entry's skeleton or basePattern does match an existing entry but that entry also had a skeleton specified
// (i.e. it was also from availableFormats), then the new entry does not override it regardless of the value of the override
// parameter. This prevents later availableFormats entries from a parent locale overriding earlier ones from the actual
// specified locale. However, availableFormats entries *should* override entries with matching skeleton whose skeleton was
// derived (i.e. entries derived from the standard date/time patters for the specified locale).
// 3. When adding the pattern (patternMap->add), we set a new boolean to indicate that the added entry had a
// specified skeleton (which sets a new field in the PtnElem in the PatternMap).
UDateTimePatternConflict
DateTimePatternGenerator::addPatternWithSkeleton(
const UnicodeString& pattern,
const UnicodeString* skeletonToUse,
UBool override,
UnicodeString& conflictingPattern,
UErrorCode& status)
{
UnicodeString basePattern;
PtnSkeleton skeleton;
UDateTimePatternConflict conflictingStatus = UDATPG_NO_CONFLICT;
DateTimeMatcher matcher;
if ( skeletonToUse == NULL ) {
matcher.set(pattern, fp, skeleton);
matcher.getBasePattern(basePattern);
} else {
matcher.set(*skeletonToUse, fp, skeleton); // this still trims skeleton fields to max len 3, may need to change it.
matcher.getBasePattern(basePattern); // or perhaps instead: basePattern = *skeletonToUse;
}
UBool entryHadSpecifiedSkeleton;
const UnicodeString *duplicatePattern = patternMap->getPatternFromBasePattern(basePattern, entryHadSpecifiedSkeleton);
if (duplicatePattern != NULL ) {
conflictingStatus = UDATPG_BASE_CONFLICT;
conflictingPattern = *duplicatePattern;
if (!override || (skeletonToUse != NULL && entryHadSpecifiedSkeleton)) {
return conflictingStatus;
}
}
const PtnSkeleton* entrySpecifiedSkeleton = NULL;
duplicatePattern = patternMap->getPatternFromSkeleton(skeleton, &entrySpecifiedSkeleton);
if (duplicatePattern != NULL ) {
conflictingStatus = UDATPG_CONFLICT;
conflictingPattern = *duplicatePattern;
if (!override || (skeletonToUse != NULL && entrySpecifiedSkeleton != NULL)) {
return conflictingStatus;
}
}
patternMap->add(basePattern, skeleton, pattern, skeletonToUse != NULL, status);
if(U_FAILURE(status)) {
return conflictingStatus;
}
return UDATPG_NO_CONFLICT;
}
UDateTimePatternField
DateTimePatternGenerator::getAppendFormatNumber(const char* field) const {
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i ) {
if (uprv_strcmp(CLDR_FIELD_APPEND[i], field)==0) {
return (UDateTimePatternField)i;
}
}
return UDATPG_FIELD_COUNT;
}
UDateTimePatternField
DateTimePatternGenerator::getAppendNameNumber(const char* field) const {
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i ) {
if (uprv_strcmp(CLDR_FIELD_NAME[i],field)==0) {
return (UDateTimePatternField)i;
}
}
return UDATPG_FIELD_COUNT;
}
const UnicodeString*
DateTimePatternGenerator::getBestRaw(DateTimeMatcher& source,
int32_t includeMask,
DistanceInfo* missingFields,
const PtnSkeleton** specifiedSkeletonPtr) {
int32_t bestDistance = 0x7fffffff;
DistanceInfo tempInfo;
const UnicodeString *bestPattern=NULL;
const PtnSkeleton* specifiedSkeleton=NULL;
PatternMapIterator it;
for (it.set(*patternMap); it.hasNext(); ) {
DateTimeMatcher trial = it.next();
if (trial.equals(skipMatcher)) {
continue;
}
int32_t distance=source.getDistance(trial, includeMask, tempInfo);
if (distance<bestDistance) {
bestDistance=distance;
bestPattern=patternMap->getPatternFromSkeleton(*trial.getSkeletonPtr(), &specifiedSkeleton);
missingFields->setTo(tempInfo);
if (distance==0) {
break;
}
}
}
// If the best raw match had a specified skeleton and that skeleton was requested by the caller,
// then return it too. This generally happens when the caller needs to pass that skeleton
// through to adjustFieldTypes so the latter can do a better job.
if (bestPattern && specifiedSkeletonPtr) {
*specifiedSkeletonPtr = specifiedSkeleton;
}
return bestPattern;
}
UnicodeString
DateTimePatternGenerator::adjustFieldTypes(const UnicodeString& pattern,
const PtnSkeleton* specifiedSkeleton,
UBool fixFractionalSeconds,
UDateTimePatternMatchOptions options) {
UnicodeString newPattern;
fp->set(pattern);
for (int32_t i=0; i < fp->itemNumber; i++) {
UnicodeString field = fp->items[i];
if ( fp->isQuoteLiteral(field) ) {
UnicodeString quoteLiteral;
fp->getQuoteLiteral(quoteLiteral, &i);
newPattern += quoteLiteral;
}
else {
if (fp->isPatternSeparator(field)) {
newPattern+=field;
continue;
}
int32_t canonicalIndex = fp->getCanonicalIndex(field);
if (canonicalIndex < 0) {
newPattern+=field;
continue; // don't adjust
}
const dtTypeElem *row = &dtTypes[canonicalIndex];
int32_t typeValue = row->field;
if (fixFractionalSeconds && typeValue == UDATPG_SECOND_FIELD) {
UnicodeString newField=dtMatcher->skeleton.original[UDATPG_FRACTIONAL_SECOND_FIELD];
field = field + decimal + newField;
}
else {
if (dtMatcher->skeleton.type[typeValue]!=0) {
// Here:
// - "reqField" is the field from the originally requested skeleton, with length
// "reqFieldLen".
// - "field" is the field from the found pattern.
//
// The adjusted field should consist of characters from the originally requested
// skeleton, except in the case of UDATPG_HOUR_FIELD or UDATPG_MONTH_FIELD, in
// which case it should consist of characters from the found pattern.
//
// The length of the adjusted field (adjFieldLen) should match that in the originally
// requested skeleton, except that in the following cases the length of the adjusted field
// should match that in the found pattern (i.e. the length of this pattern field should
// not be adjusted):
// 1. typeValue is UDATPG_HOUR_FIELD/MINUTE/SECOND and the corresponding bit in options is
// not set (ticket #7180). Note, we may want to implement a similar change for other
// numeric fields (MM, dd, etc.) so the default behavior is to get locale preference for
// field length, but options bits can be used to override this.
// 2. There is a specified skeleton for the found pattern and one of the following is true:
// a) The length of the field in the skeleton (skelFieldLen) is equal to reqFieldLen.
// b) The pattern field is numeric and the skeleton field is not, or vice versa.
UnicodeString reqField = dtMatcher->skeleton.original[typeValue];
int32_t reqFieldLen = reqField.length();
int32_t adjFieldLen = reqFieldLen;
if ( (typeValue==UDATPG_HOUR_FIELD && (options & UDATPG_MATCH_HOUR_FIELD_LENGTH)==0) ||
(typeValue==UDATPG_MINUTE_FIELD && (options & UDATPG_MATCH_MINUTE_FIELD_LENGTH)==0) ||
(typeValue==UDATPG_SECOND_FIELD && (options & UDATPG_MATCH_SECOND_FIELD_LENGTH)==0) ) {
adjFieldLen = field.length();
} else if (specifiedSkeleton) {
UnicodeString skelField = specifiedSkeleton->original[typeValue];
int32_t skelFieldLen = skelField.length();
UBool patFieldIsNumeric = (row->type > 0);
UBool skelFieldIsNumeric = (specifiedSkeleton->type[typeValue] > 0);
if (skelFieldLen == reqFieldLen || (patFieldIsNumeric && !skelFieldIsNumeric) || (skelFieldIsNumeric && !patFieldIsNumeric)) {
// don't adjust the field length in the found pattern
adjFieldLen = field.length();
}
}
UChar c = (typeValue!= UDATPG_HOUR_FIELD && typeValue!= UDATPG_MONTH_FIELD)? reqField.charAt(0): field.charAt(0);
field.remove();
for (int32_t i=adjFieldLen; i>0; --i) {
field+=c;
}
}
newPattern+=field;
}
}
}
return newPattern;
}
UnicodeString
DateTimePatternGenerator::getBestAppending(int32_t missingFields, UDateTimePatternMatchOptions options) {
UnicodeString resultPattern, tempPattern, formattedPattern;
UErrorCode err=U_ZERO_ERROR;
int32_t lastMissingFieldMask=0;
if (missingFields!=0) {
resultPattern=UnicodeString();
const PtnSkeleton* specifiedSkeleton=NULL;
tempPattern = *getBestRaw(*dtMatcher, missingFields, distanceInfo, &specifiedSkeleton);
resultPattern = adjustFieldTypes(tempPattern, specifiedSkeleton, FALSE, options);
if ( distanceInfo->missingFieldMask==0 ) {
return resultPattern;
}
while (distanceInfo->missingFieldMask!=0) { // precondition: EVERY single field must work!
if ( lastMissingFieldMask == distanceInfo->missingFieldMask ) {
break; // cannot find the proper missing field
}
if (((distanceInfo->missingFieldMask & UDATPG_SECOND_AND_FRACTIONAL_MASK)==UDATPG_FRACTIONAL_MASK) &&
((missingFields & UDATPG_SECOND_AND_FRACTIONAL_MASK) == UDATPG_SECOND_AND_FRACTIONAL_MASK)) {
resultPattern = adjustFieldTypes(resultPattern, specifiedSkeleton, FALSE, options);
//resultPattern = tempPattern;
distanceInfo->missingFieldMask &= ~UDATPG_FRACTIONAL_MASK;
continue;
}
int32_t startingMask = distanceInfo->missingFieldMask;
tempPattern = *getBestRaw(*dtMatcher, distanceInfo->missingFieldMask, distanceInfo, &specifiedSkeleton);
tempPattern = adjustFieldTypes(tempPattern, specifiedSkeleton, FALSE, options);
int32_t foundMask=startingMask& ~distanceInfo->missingFieldMask;
int32_t topField=getTopBitNumber(foundMask);
UnicodeString appendName;
getAppendName((UDateTimePatternField)topField, appendName);
const Formattable formatPattern[] = {
resultPattern,
tempPattern,
appendName
};
UnicodeString emptyStr;
formattedPattern = MessageFormat::format(appendItemFormats[topField], formatPattern, 3, emptyStr, err);
lastMissingFieldMask = distanceInfo->missingFieldMask;
}
}
return formattedPattern;
}
int32_t
DateTimePatternGenerator::getTopBitNumber(int32_t foundMask) {
if ( foundMask==0 ) {
return 0;
}
int32_t i=0;
while (foundMask!=0) {
foundMask >>=1;
++i;
}
if (i-1 >UDATPG_ZONE_FIELD) {
return UDATPG_ZONE_FIELD;
}
else
return i-1;
}
void
DateTimePatternGenerator::setAvailableFormat(const UnicodeString &key, UErrorCode& err)
{
fAvailableFormatKeyHash->puti(key, 1, err);
}
UBool
DateTimePatternGenerator::isAvailableFormatSet(const UnicodeString &key) const {
return (UBool)(fAvailableFormatKeyHash->geti(key) == 1);
}
void
DateTimePatternGenerator::copyHashtable(Hashtable *other, UErrorCode &status) {
if (other == NULL) {
return;
}
if (fAvailableFormatKeyHash != NULL) {
delete fAvailableFormatKeyHash;
fAvailableFormatKeyHash = NULL;
}
initHashtable(status);
if(U_FAILURE(status)){
return;
}
int32_t pos = -1;
const UHashElement* elem = NULL;
// walk through the hash table and create a deep clone
while((elem = other->nextElement(pos))!= NULL){
const UHashTok otherKeyTok = elem->key;
UnicodeString* otherKey = (UnicodeString*)otherKeyTok.pointer;
fAvailableFormatKeyHash->puti(*otherKey, 1, status);
if(U_FAILURE(status)){
return;
}
}
}
StringEnumeration*
DateTimePatternGenerator::getSkeletons(UErrorCode& status) const {
StringEnumeration* skeletonEnumerator = new DTSkeletonEnumeration(*patternMap, DT_SKELETON, status);
return skeletonEnumerator;
}
const UnicodeString&
DateTimePatternGenerator::getPatternForSkeleton(const UnicodeString& skeleton) const {
PtnElem *curElem;
if (skeleton.length() ==0) {
return emptyString;
}
curElem = patternMap->getHeader(skeleton.charAt(0));
while ( curElem != NULL ) {
if ( curElem->skeleton->getSkeleton()==skeleton ) {
return curElem->pattern;
}
curElem=curElem->next;
}
return emptyString;
}
StringEnumeration*
DateTimePatternGenerator::getBaseSkeletons(UErrorCode& status) const {
StringEnumeration* baseSkeletonEnumerator = new DTSkeletonEnumeration(*patternMap, DT_BASESKELETON, status);
return baseSkeletonEnumerator;
}
StringEnumeration*
DateTimePatternGenerator::getRedundants(UErrorCode& status) {
StringEnumeration* output = new DTRedundantEnumeration();
const UnicodeString *pattern;
PatternMapIterator it;
for (it.set(*patternMap); it.hasNext(); ) {
DateTimeMatcher current = it.next();
pattern = patternMap->getPatternFromSkeleton(*(it.getSkeleton()));
if ( isCanonicalItem(*pattern) ) {
continue;
}
if ( skipMatcher == NULL ) {
skipMatcher = new DateTimeMatcher(current);
}
else {
*skipMatcher = current;
}
UnicodeString trial = getBestPattern(current.getPattern(), status);
if (trial == *pattern) {
((DTRedundantEnumeration *)output)->add(*pattern, status);
}
if (current.equals(skipMatcher)) {
continue;
}
}
return output;
}
UBool
DateTimePatternGenerator::isCanonicalItem(const UnicodeString& item) const {
if ( item.length() != 1 ) {
return FALSE;
}
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (item.charAt(0)==Canonical_Items[i]) {
return TRUE;
}
}
return FALSE;
}
DateTimePatternGenerator*
DateTimePatternGenerator::clone() const {
return new DateTimePatternGenerator(*this);
}
PatternMap::PatternMap() {
for (int32_t i=0; i < MAX_PATTERN_ENTRIES; ++i ) {
boot[i]=NULL;
}
isDupAllowed = TRUE;
}
void
PatternMap::copyFrom(const PatternMap& other, UErrorCode& status) {
this->isDupAllowed = other.isDupAllowed;
for (int32_t bootIndex=0; bootIndex<MAX_PATTERN_ENTRIES; ++bootIndex ) {
PtnElem *curElem, *otherElem, *prevElem=NULL;
otherElem = other.boot[bootIndex];
while (otherElem!=NULL) {
if ((curElem = new PtnElem(otherElem->basePattern, otherElem->pattern))==NULL) {
// out of memory
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
if ( this->boot[bootIndex]== NULL ) {
this->boot[bootIndex] = curElem;
}
if ((curElem->skeleton=new PtnSkeleton(*(otherElem->skeleton))) == NULL ) {
// out of memory
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
if (prevElem!=NULL) {
prevElem->next=curElem;
}
curElem->next=NULL;
prevElem = curElem;
otherElem = otherElem->next;
}
}
}
PtnElem*
PatternMap::getHeader(UChar baseChar) {
PtnElem* curElem;
if ( (baseChar >= CAP_A) && (baseChar <= CAP_Z) ) {
curElem = boot[baseChar-CAP_A];
}
else {
if ( (baseChar >=LOW_A) && (baseChar <= LOW_Z) ) {
curElem = boot[26+baseChar-LOW_A];
}
else {
return NULL;
}
}
return curElem;
}
PatternMap::~PatternMap() {
for (int32_t i=0; i < MAX_PATTERN_ENTRIES; ++i ) {
if (boot[i]!=NULL ) {
delete boot[i];
boot[i]=NULL;
}
}
} // PatternMap destructor
void
PatternMap::add(const UnicodeString& basePattern,
const PtnSkeleton& skeleton,
const UnicodeString& value,// mapped pattern value
UBool skeletonWasSpecified,
UErrorCode &status) {
UChar baseChar = basePattern.charAt(0);
PtnElem *curElem, *baseElem;
status = U_ZERO_ERROR;
// the baseChar must be A-Z or a-z
if ((baseChar >= CAP_A) && (baseChar <= CAP_Z)) {
baseElem = boot[baseChar-CAP_A];
}
else {
if ((baseChar >=LOW_A) && (baseChar <= LOW_Z)) {
baseElem = boot[26+baseChar-LOW_A];
}
else {
status = U_ILLEGAL_CHARACTER;
return;
}
}
if (baseElem == NULL) {
if ((curElem = new PtnElem(basePattern, value)) == NULL ) {
// out of memory
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
if (baseChar >= LOW_A) {
boot[26 + (baseChar-LOW_A)] = curElem;
}
else {
boot[baseChar-CAP_A] = curElem;
}
curElem->skeleton = new PtnSkeleton(skeleton);
curElem->skeletonWasSpecified = skeletonWasSpecified;
}
if ( baseElem != NULL ) {
curElem = getDuplicateElem(basePattern, skeleton, baseElem);
if (curElem == NULL) {
// add new element to the list.
curElem = baseElem;
while( curElem -> next != NULL )
{
curElem = curElem->next;
}
if ((curElem->next = new PtnElem(basePattern, value)) == NULL ) {
// out of memory
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
curElem=curElem->next;
curElem->skeleton = new PtnSkeleton(skeleton);
curElem->skeletonWasSpecified = skeletonWasSpecified;
}
else {
// Pattern exists in the list already.
if ( !isDupAllowed ) {
return;
}
// Overwrite the value.
curElem->pattern = value;
}
}
} // PatternMap::add
// Find the pattern from the given basePattern string.
const UnicodeString *
PatternMap::getPatternFromBasePattern(UnicodeString& basePattern, UBool& skeletonWasSpecified) { // key to search for
PtnElem *curElem;
if ((curElem=getHeader(basePattern.charAt(0)))==NULL) {
return NULL; // no match
}
do {
if ( basePattern.compare(curElem->basePattern)==0 ) {
skeletonWasSpecified = curElem->skeletonWasSpecified;
return &(curElem->pattern);
}
curElem=curElem->next;
}while (curElem != NULL);
return NULL;
} // PatternMap::getFromBasePattern
// Find the pattern from the given skeleton.
// At least when this is called from getBestRaw & addPattern (in which case specifiedSkeletonPtr is non-NULL),
// the comparison should be based on skeleton.original (which is unique and tied to the distance measurement in bestRaw)
// and not skeleton.baseOriginal (which is not unique); otherwise we may pick a different skeleton than the one with the
// optimum distance value in getBestRaw. When this is called from public getRedundants (specifiedSkeletonPtr is NULL),
// for now it will continue to compare based on baseOriginal so as not to change the behavior unnecessarily.
const UnicodeString *
PatternMap::getPatternFromSkeleton(PtnSkeleton& skeleton, const PtnSkeleton** specifiedSkeletonPtr) { // key to search for
PtnElem *curElem;
if (specifiedSkeletonPtr) {
*specifiedSkeletonPtr = NULL;
}
// find boot entry
UChar baseChar='\0';
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (skeleton.baseOriginal[i].length() !=0 ) {
baseChar = skeleton.baseOriginal[i].charAt(0);
break;
}
}
if ((curElem=getHeader(baseChar))==NULL) {
return NULL; // no match
}
do {
int32_t i=0;
if (specifiedSkeletonPtr != NULL) { // called from DateTimePatternGenerator::getBestRaw or addPattern, use original
for (i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (curElem->skeleton->original[i].compare(skeleton.original[i]) != 0 )
{
break;
}
}
} else { // called from DateTimePatternGenerator::getRedundants, use baseOriginal
for (i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (curElem->skeleton->baseOriginal[i].compare(skeleton.baseOriginal[i]) != 0 )
{
break;
}
}
}
if (i == UDATPG_FIELD_COUNT) {
if (specifiedSkeletonPtr && curElem->skeletonWasSpecified) {
*specifiedSkeletonPtr = curElem->skeleton;
}
return &(curElem->pattern);
}
curElem=curElem->next;
}while (curElem != NULL);
return NULL;
}
UBool
PatternMap::equals(const PatternMap& other) {
if ( this==&other ) {
return TRUE;
}
for (int32_t bootIndex=0; bootIndex<MAX_PATTERN_ENTRIES; ++bootIndex ) {
if ( boot[bootIndex]==other.boot[bootIndex] ) {
continue;
}
if ( (boot[bootIndex]==NULL)||(other.boot[bootIndex]==NULL) ) {
return FALSE;
}
PtnElem *otherElem = other.boot[bootIndex];
PtnElem *myElem = boot[bootIndex];
while ((otherElem!=NULL) || (myElem!=NULL)) {
if ( myElem == otherElem ) {
break;
}
if ((otherElem==NULL) || (myElem==NULL)) {
return FALSE;
}
if ( (myElem->basePattern != otherElem->basePattern) ||
(myElem->pattern != otherElem->pattern) ) {
return FALSE;
}
if ((myElem->skeleton!=otherElem->skeleton)&&
!myElem->skeleton->equals(*(otherElem->skeleton))) {
return FALSE;
}
myElem = myElem->next;
otherElem=otherElem->next;
}
}
return TRUE;
}
// find any key existing in the mapping table already.
// return TRUE if there is an existing key, otherwise return FALSE.
PtnElem*
PatternMap::getDuplicateElem(
const UnicodeString &basePattern,
const PtnSkeleton &skeleton,
PtnElem *baseElem) {
PtnElem *curElem;
if ( baseElem == (PtnElem *)NULL ) {
return (PtnElem*)NULL;
}
else {
curElem = baseElem;
}
do {
if ( basePattern.compare(curElem->basePattern)==0 ) {
UBool isEqual=TRUE;
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (curElem->skeleton->type[i] != skeleton.type[i] ) {
isEqual=FALSE;
break;
}
}
if (isEqual) {
return curElem;
}
}
curElem = curElem->next;
} while( curElem != (PtnElem *)NULL );
// end of the list
return (PtnElem*)NULL;
} // PatternMap::getDuplicateElem
DateTimeMatcher::DateTimeMatcher(void) {
}
DateTimeMatcher::DateTimeMatcher(const DateTimeMatcher& other) {
copyFrom(other.skeleton);
}
void
DateTimeMatcher::set(const UnicodeString& pattern, FormatParser* fp) {
PtnSkeleton localSkeleton;
return set(pattern, fp, localSkeleton);
}
void
DateTimeMatcher::set(const UnicodeString& pattern, FormatParser* fp, PtnSkeleton& skeletonResult) {
int32_t i;
for (i=0; i<UDATPG_FIELD_COUNT; ++i) {
skeletonResult.type[i]=NONE;
}
fp->set(pattern);
for (i=0; i < fp->itemNumber; i++) {
UnicodeString field = fp->items[i];
if ( field.charAt(0) == LOW_A ) {
continue; // skip 'a'
}
if ( fp->isQuoteLiteral(field) ) {
UnicodeString quoteLiteral;
fp->getQuoteLiteral(quoteLiteral, &i);
continue;
}
int32_t canonicalIndex = fp->getCanonicalIndex(field);
if (canonicalIndex < 0 ) {
continue;
}
const dtTypeElem *row = &dtTypes[canonicalIndex];
int32_t typeValue = row->field;
skeletonResult.original[typeValue]=field;
UChar repeatChar = row->patternChar;
int32_t repeatCount = row->minLen > 3 ? 3: row->minLen;
while (repeatCount-- > 0) {
skeletonResult.baseOriginal[typeValue] += repeatChar;
}
int16_t subTypeValue = row->type;
if ( row->type > 0) {
subTypeValue += field.length();
}
skeletonResult.type[typeValue] = subTypeValue;
}
copyFrom(skeletonResult);
}
void
DateTimeMatcher::getBasePattern(UnicodeString &result ) {
result.remove(); // Reset the result first.
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i ) {
if (skeleton.baseOriginal[i].length()!=0) {
result += skeleton.baseOriginal[i];
}
}
}
UnicodeString
DateTimeMatcher::getPattern() {
UnicodeString result;
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i ) {
if (skeleton.original[i].length()!=0) {
result += skeleton.original[i];
}
}
return result;
}
int32_t
DateTimeMatcher::getDistance(const DateTimeMatcher& other, int32_t includeMask, DistanceInfo& distanceInfo) {
int32_t result=0;
distanceInfo.clear();
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i ) {
int32_t myType = (includeMask&(1<<i))==0 ? 0 : skeleton.type[i];
int32_t otherType = other.skeleton.type[i];
if (myType==otherType) {
continue;
}
if (myType==0) {// and other is not
result += EXTRA_FIELD;
distanceInfo.addExtra(i);
}
else {
if (otherType==0) {
result += MISSING_FIELD;
distanceInfo.addMissing(i);
}
else {
result += abs(myType - otherType);
}
}
}
return result;
}
void
DateTimeMatcher::copyFrom(const PtnSkeleton& newSkeleton) {
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
this->skeleton.type[i]=newSkeleton.type[i];
this->skeleton.original[i]=newSkeleton.original[i];
this->skeleton.baseOriginal[i]=newSkeleton.baseOriginal[i];
}
}
void
DateTimeMatcher::copyFrom() {
// same as clear
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
this->skeleton.type[i]=0;
this->skeleton.original[i].remove();
this->skeleton.baseOriginal[i].remove();
}
}
UBool
DateTimeMatcher::equals(const DateTimeMatcher* other) const {
if (other==NULL) {
return FALSE;
}
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (this->skeleton.original[i]!=other->skeleton.original[i] ) {
return FALSE;
}
}
return TRUE;
}
int32_t
DateTimeMatcher::getFieldMask() {
int32_t result=0;
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (skeleton.type[i]!=0) {
result |= (1<<i);
}
}
return result;
}
PtnSkeleton*
DateTimeMatcher::getSkeletonPtr() {
return &skeleton;
}
FormatParser::FormatParser () {
status = START;
itemNumber=0;
}
FormatParser::~FormatParser () {
}
// Find the next token with the starting position and length
// Note: the startPos may
FormatParser::TokenStatus
FormatParser::setTokens(const UnicodeString& pattern, int32_t startPos, int32_t *len) {
int32_t curLoc = startPos;
if ( curLoc >= pattern.length()) {
return DONE;
}
// check the current char is between A-Z or a-z
do {
UChar c=pattern.charAt(curLoc);
if ( (c>=CAP_A && c<=CAP_Z) || (c>=LOW_A && c<=LOW_Z) ) {
curLoc++;
}
else {
startPos = curLoc;
*len=1;
return ADD_TOKEN;
}
if ( pattern.charAt(curLoc)!= pattern.charAt(startPos) ) {
break; // not the same token
}
} while(curLoc <= pattern.length());
*len = curLoc-startPos;
return ADD_TOKEN;
}
void
FormatParser::set(const UnicodeString& pattern) {
int32_t startPos=0;
TokenStatus result=START;
int32_t len=0;
itemNumber =0;
do {
result = setTokens( pattern, startPos, &len );
if ( result == ADD_TOKEN )
{
items[itemNumber++] = UnicodeString(pattern, startPos, len );
startPos += len;
}
else {
break;
}
} while (result==ADD_TOKEN && itemNumber < MAX_DT_TOKEN);
}
int32_t
FormatParser::getCanonicalIndex(const UnicodeString& s, UBool strict) {
int32_t len = s.length();
if (len == 0) {
return -1;
}
UChar ch = s.charAt(0);
// Verify that all are the same character.
for (int32_t l = 1; l < len; l++) {
if (ch != s.charAt(l)) {
return -1;
}
}
int32_t i = 0;
int32_t bestRow = -1;
while (dtTypes[i].patternChar != '\0') {
if ( dtTypes[i].patternChar != ch ) {
++i;
continue;
}
bestRow = i;
if (dtTypes[i].patternChar != dtTypes[i+1].patternChar) {
return i;
}
if (dtTypes[i+1].minLen <= len) {
++i;
continue;
}
return i;
}
return strict ? -1 : bestRow;
}
UBool
FormatParser::isQuoteLiteral(const UnicodeString& s) const {
return (UBool)(s.charAt(0)==SINGLE_QUOTE);
}
// This function aussumes the current itemIndex points to the quote literal.
// Please call isQuoteLiteral prior to this function.
void
FormatParser::getQuoteLiteral(UnicodeString& quote, int32_t *itemIndex) {
int32_t i=*itemIndex;
quote.remove();
if (items[i].charAt(0)==SINGLE_QUOTE) {
quote += items[i];
++i;
}
while ( i < itemNumber ) {
if ( items[i].charAt(0)==SINGLE_QUOTE ) {
if ( (i+1<itemNumber) && (items[i+1].charAt(0)==SINGLE_QUOTE)) {
// two single quotes e.g. 'o''clock'
quote += items[i++];
quote += items[i++];
continue;
}
else {
quote += items[i];
break;
}
}
else {
quote += items[i];
}
++i;
}
*itemIndex=i;
}
UBool
FormatParser::isPatternSeparator(UnicodeString& field) {
for (int32_t i=0; i<field.length(); ++i ) {
UChar c= field.charAt(i);
if ( (c==SINGLE_QUOTE) || (c==BACKSLASH) || (c==SPACE) || (c==COLON) ||
(c==QUOTATION_MARK) || (c==COMMA) || (c==HYPHEN) ||(items[i].charAt(0)==DOT) ) {
continue;
}
else {
return FALSE;
}
}
return TRUE;
}
void
DistanceInfo::setTo(DistanceInfo &other) {
missingFieldMask = other.missingFieldMask;
extraFieldMask= other.extraFieldMask;
}
PatternMapIterator::PatternMapIterator() {
bootIndex = 0;
nodePtr = NULL;
patternMap=NULL;
matcher= new DateTimeMatcher();
}
PatternMapIterator::~PatternMapIterator() {
delete matcher;
}
void
PatternMapIterator::set(PatternMap& newPatternMap) {
this->patternMap=&newPatternMap;
}
PtnSkeleton*
PatternMapIterator::getSkeleton() {
if ( nodePtr == NULL ) {
return NULL;
}
else {
return nodePtr->skeleton;
}
}
UBool
PatternMapIterator::hasNext() {
int32_t headIndex=bootIndex;
PtnElem *curPtr=nodePtr;
if (patternMap==NULL) {
return FALSE;
}
while ( headIndex < MAX_PATTERN_ENTRIES ) {
if ( curPtr != NULL ) {
if ( curPtr->next != NULL ) {
return TRUE;
}
else {
headIndex++;
curPtr=NULL;
continue;
}
}
else {
if ( patternMap->boot[headIndex] != NULL ) {
return TRUE;
}
else {
headIndex++;
continue;
}
}
}
return FALSE;
}
DateTimeMatcher&
PatternMapIterator::next() {
while ( bootIndex < MAX_PATTERN_ENTRIES ) {
if ( nodePtr != NULL ) {
if ( nodePtr->next != NULL ) {
nodePtr = nodePtr->next;
break;
}
else {
bootIndex++;
nodePtr=NULL;
continue;
}
}
else {
if ( patternMap->boot[bootIndex] != NULL ) {
nodePtr = patternMap->boot[bootIndex];
break;
}
else {
bootIndex++;
continue;
}
}
}
if (nodePtr!=NULL) {
matcher->copyFrom(*nodePtr->skeleton);
}
else {
matcher->copyFrom();
}
return *matcher;
}
PtnSkeleton::PtnSkeleton() {
}
PtnSkeleton::PtnSkeleton(const PtnSkeleton& other) {
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
this->type[i]=other.type[i];
this->original[i]=other.original[i];
this->baseOriginal[i]=other.baseOriginal[i];
}
}
UBool
PtnSkeleton::equals(const PtnSkeleton& other) {
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if ( (type[i]!= other.type[i]) ||
(original[i]!=other.original[i]) ||
(baseOriginal[i]!=other.baseOriginal[i]) ) {
return FALSE;
}
}
return TRUE;
}
UnicodeString
PtnSkeleton::getSkeleton() {
UnicodeString result;
for(int32_t i=0; i< UDATPG_FIELD_COUNT; ++i) {
if (original[i].length()!=0) {
result += original[i];
}
}
return result;
}
UnicodeString
PtnSkeleton::getBaseSkeleton() {
UnicodeString result;
for(int32_t i=0; i< UDATPG_FIELD_COUNT; ++i) {
if (baseOriginal[i].length()!=0) {
result += baseOriginal[i];
}
}
return result;
}
PtnSkeleton::~PtnSkeleton() {
}
PtnElem::PtnElem(const UnicodeString &basePat, const UnicodeString &pat) :
basePattern(basePat),
skeleton(NULL),
pattern(pat),
next(NULL)
{
}
PtnElem::~PtnElem() {
if (next!=NULL) {
delete next;
}
delete skeleton;
}
DTSkeletonEnumeration::DTSkeletonEnumeration(PatternMap &patternMap, dtStrEnum type, UErrorCode& status) {
PtnElem *curElem;
PtnSkeleton *curSkeleton;
UnicodeString s;
int32_t bootIndex;
pos=0;
fSkeletons = new UVector(status);
if (U_FAILURE(status)) {
delete fSkeletons;
return;
}
for (bootIndex=0; bootIndex<MAX_PATTERN_ENTRIES; ++bootIndex ) {
curElem = patternMap.boot[bootIndex];
while (curElem!=NULL) {
switch(type) {
case DT_BASESKELETON:
s=curElem->basePattern;
break;
case DT_PATTERN:
s=curElem->pattern;
break;
case DT_SKELETON:
curSkeleton=curElem->skeleton;
s=curSkeleton->getSkeleton();
break;
}
if ( !isCanonicalItem(s) ) {
fSkeletons->addElement(new UnicodeString(s), status);
if (U_FAILURE(status)) {
delete fSkeletons;
fSkeletons = NULL;
return;
}
}
curElem = curElem->next;
}
}
if ((bootIndex==MAX_PATTERN_ENTRIES) && (curElem!=NULL) ) {
status = U_BUFFER_OVERFLOW_ERROR;
}
}
const UnicodeString*
DTSkeletonEnumeration::snext(UErrorCode& status) {
if (U_SUCCESS(status) && pos < fSkeletons->size()) {
return (const UnicodeString*)fSkeletons->elementAt(pos++);
}
return NULL;
}
void
DTSkeletonEnumeration::reset(UErrorCode& /*status*/) {
pos=0;
}
int32_t
DTSkeletonEnumeration::count(UErrorCode& /*status*/) const {
return (fSkeletons==NULL) ? 0 : fSkeletons->size();
}
UBool
DTSkeletonEnumeration::isCanonicalItem(const UnicodeString& item) {
if ( item.length() != 1 ) {
return FALSE;
}
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (item.charAt(0)==Canonical_Items[i]) {
return TRUE;
}
}
return FALSE;
}
DTSkeletonEnumeration::~DTSkeletonEnumeration() {
UnicodeString *s;
for (int32_t i=0; i<fSkeletons->size(); ++i) {
if ((s=(UnicodeString *)fSkeletons->elementAt(i))!=NULL) {
delete s;
}
}
delete fSkeletons;
}
DTRedundantEnumeration::DTRedundantEnumeration() {
pos=0;
fPatterns = NULL;
}
void
DTRedundantEnumeration::add(const UnicodeString& pattern, UErrorCode& status) {
if (U_FAILURE(status)) return;
if (fPatterns == NULL) {
fPatterns = new UVector(status);
if (U_FAILURE(status)) {
delete fPatterns;
fPatterns = NULL;
return;
}
}
fPatterns->addElement(new UnicodeString(pattern), status);
if (U_FAILURE(status)) {
delete fPatterns;
fPatterns = NULL;
return;
}
}
const UnicodeString*
DTRedundantEnumeration::snext(UErrorCode& status) {
if (U_SUCCESS(status) && pos < fPatterns->size()) {
return (const UnicodeString*)fPatterns->elementAt(pos++);
}
return NULL;
}
void
DTRedundantEnumeration::reset(UErrorCode& /*status*/) {
pos=0;
}
int32_t
DTRedundantEnumeration::count(UErrorCode& /*status*/) const {
return (fPatterns==NULL) ? 0 : fPatterns->size();
}
UBool
DTRedundantEnumeration::isCanonicalItem(const UnicodeString& item) {
if ( item.length() != 1 ) {
return FALSE;
}
for (int32_t i=0; i<UDATPG_FIELD_COUNT; ++i) {
if (item.charAt(0)==Canonical_Items[i]) {
return TRUE;
}
}
return FALSE;
}
DTRedundantEnumeration::~DTRedundantEnumeration() {
UnicodeString *s;
for (int32_t i=0; i<fPatterns->size(); ++i) {
if ((s=(UnicodeString *)fPatterns->elementAt(i))!=NULL) {
delete s;
}
}
delete fPatterns;
}
U_NAMESPACE_END
#endif /* #if !UCONFIG_NO_FORMATTING */
//eof