scuffed-code/icu4c/source/i18n/tznames_impl.cpp
2014-07-23 21:51:09 +00:00

2020 lines
61 KiB
C++

/*
*******************************************************************************
* Copyright (C) 2011-2014, International Business Machines Corporation and
* others. All Rights Reserved.
*******************************************************************************
*
* File TZNAMES_IMPL.CPP
*
*******************************************************************************
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_FORMATTING
#include "unicode/ustring.h"
#include "unicode/timezone.h"
#include "tznames_impl.h"
#include "cmemory.h"
#include "cstring.h"
#include "uassert.h"
#include "mutex.h"
#include "uresimp.h"
#include "ureslocs.h"
#include "zonemeta.h"
#include "ucln_in.h"
#include "uvector.h"
#include "olsontz.h"
U_NAMESPACE_BEGIN
#define ZID_KEY_MAX 128
#define MZ_PREFIX_LEN 5
static const char gZoneStrings[] = "zoneStrings";
static const char gMZPrefix[] = "meta:";
static const char* KEYS[] = {"lg", "ls", "ld", "sg", "ss", "sd"};
static const int32_t KEYS_SIZE = (sizeof KEYS / sizeof KEYS[0]);
static const char gEcTag[] = "ec";
static const char EMPTY[] = "<empty>"; // place holder for empty ZNames/TZNames
static const UTimeZoneNameType ALL_NAME_TYPES[] = {
UTZNM_LONG_GENERIC, UTZNM_LONG_STANDARD, UTZNM_LONG_DAYLIGHT,
UTZNM_SHORT_GENERIC, UTZNM_SHORT_STANDARD, UTZNM_SHORT_DAYLIGHT,
UTZNM_EXEMPLAR_LOCATION,
UTZNM_UNKNOWN // unknown as the last one
};
// stuff for TZDBTimeZoneNames
static const char* TZDBNAMES_KEYS[] = {"ss", "sd"};
static const int32_t TZDBNAMES_KEYS_SIZE = (sizeof TZDBNAMES_KEYS / sizeof TZDBNAMES_KEYS[0]);
static UMutex gTZDBNamesMapLock = U_MUTEX_INITIALIZER;
static UHashtable* gTZDBNamesMap = NULL;
static icu::UInitOnce gTZDBNamesMapInitOnce = U_INITONCE_INITIALIZER;
static TextTrieMap* gTZDBNamesTrie = NULL;
static icu::UInitOnce gTZDBNamesTrieInitOnce = U_INITONCE_INITIALIZER;
U_CDECL_BEGIN
static UBool U_CALLCONV tzdbTimeZoneNames_cleanup(void) {
if (gTZDBNamesMap != NULL) {
uhash_close(gTZDBNamesMap);
gTZDBNamesMap = NULL;
}
gTZDBNamesMapInitOnce.reset();
if (gTZDBNamesTrie != NULL) {
delete gTZDBNamesTrie;
gTZDBNamesTrie = NULL;
}
gTZDBNamesTrieInitOnce.reset();
return TRUE;
}
U_CDECL_END
#define DEFAULT_CHARACTERNODE_CAPACITY 1
// ---------------------------------------------------
// CharacterNode class implementation
// ---------------------------------------------------
void CharacterNode::clear() {
uprv_memset(this, 0, sizeof(*this));
}
void CharacterNode::deleteValues(UObjectDeleter *valueDeleter) {
if (fValues == NULL) {
// Do nothing.
} else if (!fHasValuesVector) {
if (valueDeleter) {
valueDeleter(fValues);
}
} else {
delete (UVector *)fValues;
}
}
void
CharacterNode::addValue(void *value, UObjectDeleter *valueDeleter, UErrorCode &status) {
if (U_FAILURE(status)) {
if (valueDeleter) {
valueDeleter(value);
}
return;
}
if (fValues == NULL) {
fValues = value;
} else {
// At least one value already.
if (!fHasValuesVector) {
// There is only one value so far, and not in a vector yet.
// Create a vector and add the old value.
UVector *values = new UVector(valueDeleter, NULL, DEFAULT_CHARACTERNODE_CAPACITY, status);
if (U_FAILURE(status)) {
if (valueDeleter) {
valueDeleter(value);
}
return;
}
values->addElement(fValues, status);
fValues = values;
fHasValuesVector = TRUE;
}
// Add the new value.
((UVector *)fValues)->addElement(value, status);
}
}
// ---------------------------------------------------
// TextTrieMapSearchResultHandler class implementation
// ---------------------------------------------------
TextTrieMapSearchResultHandler::~TextTrieMapSearchResultHandler(){
}
// ---------------------------------------------------
// TextTrieMap class implementation
// ---------------------------------------------------
TextTrieMap::TextTrieMap(UBool ignoreCase, UObjectDeleter *valueDeleter)
: fIgnoreCase(ignoreCase), fNodes(NULL), fNodesCapacity(0), fNodesCount(0),
fLazyContents(NULL), fIsEmpty(TRUE), fValueDeleter(valueDeleter) {
}
TextTrieMap::~TextTrieMap() {
int32_t index;
for (index = 0; index < fNodesCount; ++index) {
fNodes[index].deleteValues(fValueDeleter);
}
uprv_free(fNodes);
if (fLazyContents != NULL) {
for (int32_t i=0; i<fLazyContents->size(); i+=2) {
if (fValueDeleter) {
fValueDeleter(fLazyContents->elementAt(i+1));
}
}
delete fLazyContents;
}
}
int32_t TextTrieMap::isEmpty() const {
// Use a separate field for fIsEmpty because it will remain unchanged once the
// Trie is built, while fNodes and fLazyContents change with the lazy init
// of the nodes structure. Trying to test the changing fields has
// thread safety complications.
return fIsEmpty;
}
// We defer actually building the TextTrieMap node structure until the first time a
// search is performed. put() simply saves the parameters in case we do
// eventually need to build it.
//
void
TextTrieMap::put(const UnicodeString &key, void *value, ZNStringPool &sp, UErrorCode &status) {
const UChar *s = sp.get(key, status);
put(s, value, status);
}
// This method is for designed for a persistent key, such as string key stored in
// resource bundle.
void
TextTrieMap::put(const UChar *key, void *value, UErrorCode &status) {
fIsEmpty = FALSE;
if (fLazyContents == NULL) {
fLazyContents = new UVector(status);
if (fLazyContents == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
}
if (U_FAILURE(status)) {
return;
}
U_ASSERT(fLazyContents != NULL);
UChar *s = const_cast<UChar *>(key);
fLazyContents->addElement(s, status);
fLazyContents->addElement(value, status);
}
void
TextTrieMap::putImpl(const UnicodeString &key, void *value, UErrorCode &status) {
if (fNodes == NULL) {
fNodesCapacity = 512;
fNodes = (CharacterNode *)uprv_malloc(fNodesCapacity * sizeof(CharacterNode));
fNodes[0].clear(); // Init root node.
fNodesCount = 1;
}
UnicodeString foldedKey;
const UChar *keyBuffer;
int32_t keyLength;
if (fIgnoreCase) {
// Ok to use fastCopyFrom() because we discard the copy when we return.
foldedKey.fastCopyFrom(key).foldCase();
keyBuffer = foldedKey.getBuffer();
keyLength = foldedKey.length();
} else {
keyBuffer = key.getBuffer();
keyLength = key.length();
}
CharacterNode *node = fNodes;
int32_t index;
for (index = 0; index < keyLength; ++index) {
node = addChildNode(node, keyBuffer[index], status);
}
node->addValue(value, fValueDeleter, status);
}
UBool
TextTrieMap::growNodes() {
if (fNodesCapacity == 0xffff) {
return FALSE; // We use 16-bit node indexes.
}
int32_t newCapacity = fNodesCapacity + 1000;
if (newCapacity > 0xffff) {
newCapacity = 0xffff;
}
CharacterNode *newNodes = (CharacterNode *)uprv_malloc(newCapacity * sizeof(CharacterNode));
if (newNodes == NULL) {
return FALSE;
}
uprv_memcpy(newNodes, fNodes, fNodesCount * sizeof(CharacterNode));
uprv_free(fNodes);
fNodes = newNodes;
fNodesCapacity = newCapacity;
return TRUE;
}
CharacterNode*
TextTrieMap::addChildNode(CharacterNode *parent, UChar c, UErrorCode &status) {
if (U_FAILURE(status)) {
return NULL;
}
// Linear search of the sorted list of children.
uint16_t prevIndex = 0;
uint16_t nodeIndex = parent->fFirstChild;
while (nodeIndex > 0) {
CharacterNode *current = fNodes + nodeIndex;
UChar childCharacter = current->fCharacter;
if (childCharacter == c) {
return current;
} else if (childCharacter > c) {
break;
}
prevIndex = nodeIndex;
nodeIndex = current->fNextSibling;
}
// Ensure capacity. Grow fNodes[] if needed.
if (fNodesCount == fNodesCapacity) {
int32_t parentIndex = (int32_t)(parent - fNodes);
if (!growNodes()) {
status = U_MEMORY_ALLOCATION_ERROR;
return NULL;
}
parent = fNodes + parentIndex;
}
// Insert a new child node with c in sorted order.
CharacterNode *node = fNodes + fNodesCount;
node->clear();
node->fCharacter = c;
node->fNextSibling = nodeIndex;
if (prevIndex == 0) {
parent->fFirstChild = (uint16_t)fNodesCount;
} else {
fNodes[prevIndex].fNextSibling = (uint16_t)fNodesCount;
}
++fNodesCount;
return node;
}
CharacterNode*
TextTrieMap::getChildNode(CharacterNode *parent, UChar c) const {
// Linear search of the sorted list of children.
uint16_t nodeIndex = parent->fFirstChild;
while (nodeIndex > 0) {
CharacterNode *current = fNodes + nodeIndex;
UChar childCharacter = current->fCharacter;
if (childCharacter == c) {
return current;
} else if (childCharacter > c) {
break;
}
nodeIndex = current->fNextSibling;
}
return NULL;
}
// Mutex for protecting the lazy creation of the Trie node structure on the first call to search().
static UMutex TextTrieMutex = U_MUTEX_INITIALIZER;
// buildTrie() - The Trie node structure is needed. Create it from the data that was
// saved at the time the ZoneStringFormatter was created. The Trie is only
// needed for parsing operations, which are less common than formatting,
// and the Trie is big, which is why its creation is deferred until first use.
void TextTrieMap::buildTrie(UErrorCode &status) {
if (fLazyContents != NULL) {
for (int32_t i=0; i<fLazyContents->size(); i+=2) {
const UChar *key = (UChar *)fLazyContents->elementAt(i);
void *val = fLazyContents->elementAt(i+1);
UnicodeString keyString(TRUE, key, -1); // Aliasing UnicodeString constructor.
putImpl(keyString, val, status);
}
delete fLazyContents;
fLazyContents = NULL;
}
}
void
TextTrieMap::search(const UnicodeString &text, int32_t start,
TextTrieMapSearchResultHandler *handler, UErrorCode &status) const {
{
// TODO: if locking the mutex for each check proves to be a performance problem,
// add a flag of type atomic_int32_t to class TextTrieMap, and use only
// the ICU atomic safe functions for assigning and testing.
// Don't test the pointer fLazyContents.
// Don't do unless it's really required.
Mutex lock(&TextTrieMutex);
if (fLazyContents != NULL) {
TextTrieMap *nonConstThis = const_cast<TextTrieMap *>(this);
nonConstThis->buildTrie(status);
}
}
if (fNodes == NULL) {
return;
}
search(fNodes, text, start, start, handler, status);
}
void
TextTrieMap::search(CharacterNode *node, const UnicodeString &text, int32_t start,
int32_t index, TextTrieMapSearchResultHandler *handler, UErrorCode &status) const {
if (U_FAILURE(status)) {
return;
}
if (node->hasValues()) {
if (!handler->handleMatch(index - start, node, status)) {
return;
}
if (U_FAILURE(status)) {
return;
}
}
UChar32 c = text.char32At(index);
if (fIgnoreCase) {
// size of character may grow after fold operation
UnicodeString tmp(c);
tmp.foldCase();
int32_t tmpidx = 0;
while (tmpidx < tmp.length()) {
c = tmp.char32At(tmpidx);
node = getChildNode(node, c);
if (node == NULL) {
break;
}
tmpidx = tmp.moveIndex32(tmpidx, 1);
}
} else {
node = getChildNode(node, c);
}
if (node != NULL) {
search(node, text, start, index+1, handler, status);
}
}
// ---------------------------------------------------
// ZNStringPool class implementation
// ---------------------------------------------------
static const int32_t POOL_CHUNK_SIZE = 2000;
struct ZNStringPoolChunk: public UMemory {
ZNStringPoolChunk *fNext; // Ptr to next pool chunk
int32_t fLimit; // Index to start of unused area at end of fStrings
UChar fStrings[POOL_CHUNK_SIZE]; // Strings array
ZNStringPoolChunk();
};
ZNStringPoolChunk::ZNStringPoolChunk() {
fNext = NULL;
fLimit = 0;
}
ZNStringPool::ZNStringPool(UErrorCode &status) {
fChunks = NULL;
fHash = NULL;
if (U_FAILURE(status)) {
return;
}
fChunks = new ZNStringPoolChunk;
if (fChunks == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
fHash = uhash_open(uhash_hashUChars /* keyHash */,
uhash_compareUChars /* keyComp */,
uhash_compareUChars /* valueComp */,
&status);
if (U_FAILURE(status)) {
return;
}
}
ZNStringPool::~ZNStringPool() {
if (fHash != NULL) {
uhash_close(fHash);
fHash = NULL;
}
while (fChunks != NULL) {
ZNStringPoolChunk *nextChunk = fChunks->fNext;
delete fChunks;
fChunks = nextChunk;
}
}
static const UChar EmptyString = 0;
const UChar *ZNStringPool::get(const UChar *s, UErrorCode &status) {
const UChar *pooledString;
if (U_FAILURE(status)) {
return &EmptyString;
}
pooledString = static_cast<UChar *>(uhash_get(fHash, s));
if (pooledString != NULL) {
return pooledString;
}
int32_t length = u_strlen(s);
int32_t remainingLength = POOL_CHUNK_SIZE - fChunks->fLimit;
if (remainingLength <= length) {
U_ASSERT(length < POOL_CHUNK_SIZE);
if (length >= POOL_CHUNK_SIZE) {
status = U_INTERNAL_PROGRAM_ERROR;
return &EmptyString;
}
ZNStringPoolChunk *oldChunk = fChunks;
fChunks = new ZNStringPoolChunk;
if (fChunks == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return &EmptyString;
}
fChunks->fNext = oldChunk;
}
UChar *destString = &fChunks->fStrings[fChunks->fLimit];
u_strcpy(destString, s);
fChunks->fLimit += (length + 1);
uhash_put(fHash, destString, destString, &status);
return destString;
}
//
// ZNStringPool::adopt() Put a string into the hash, but do not copy the string data
// into the pool's storage. Used for strings from resource bundles,
// which will perisist for the life of the zone string formatter, and
// therefore can be used directly without copying.
const UChar *ZNStringPool::adopt(const UChar * s, UErrorCode &status) {
const UChar *pooledString;
if (U_FAILURE(status)) {
return &EmptyString;
}
if (s != NULL) {
pooledString = static_cast<UChar *>(uhash_get(fHash, s));
if (pooledString == NULL) {
UChar *ncs = const_cast<UChar *>(s);
uhash_put(fHash, ncs, ncs, &status);
}
}
return s;
}
const UChar *ZNStringPool::get(const UnicodeString &s, UErrorCode &status) {
UnicodeString &nonConstStr = const_cast<UnicodeString &>(s);
return this->get(nonConstStr.getTerminatedBuffer(), status);
}
/*
* freeze(). Close the hash table that maps to the pooled strings.
* After freezing, the pool can not be searched or added to,
* but all existing references to pooled strings remain valid.
*
* The main purpose is to recover the storage used for the hash.
*/
void ZNStringPool::freeze() {
uhash_close(fHash);
fHash = NULL;
}
// ---------------------------------------------------
// ZNames - names common for time zone and meta zone
// ---------------------------------------------------
class ZNames : public UMemory {
public:
virtual ~ZNames();
static ZNames* createInstance(UResourceBundle* rb, const char* key);
virtual const UChar* getName(UTimeZoneNameType type);
protected:
ZNames(const UChar** names);
static const UChar** loadData(UResourceBundle* rb, const char* key);
private:
const UChar** fNames;
};
ZNames::ZNames(const UChar** names)
: fNames(names) {
}
ZNames::~ZNames() {
if (fNames != NULL) {
uprv_free(fNames);
}
}
ZNames*
ZNames::createInstance(UResourceBundle* rb, const char* key) {
const UChar** names = loadData(rb, key);
if (names == NULL) {
// No names data available
return NULL;
}
return new ZNames(names);
}
const UChar*
ZNames::getName(UTimeZoneNameType type) {
if (fNames == NULL) {
return NULL;
}
const UChar *name = NULL;
switch(type) {
case UTZNM_LONG_GENERIC:
name = fNames[0];
break;
case UTZNM_LONG_STANDARD:
name = fNames[1];
break;
case UTZNM_LONG_DAYLIGHT:
name = fNames[2];
break;
case UTZNM_SHORT_GENERIC:
name = fNames[3];
break;
case UTZNM_SHORT_STANDARD:
name = fNames[4];
break;
case UTZNM_SHORT_DAYLIGHT:
name = fNames[5];
break;
case UTZNM_EXEMPLAR_LOCATION: // implemeted by subclass
default:
name = NULL;
}
return name;
}
const UChar**
ZNames::loadData(UResourceBundle* rb, const char* key) {
if (rb == NULL || key == NULL || *key == 0) {
return NULL;
}
UErrorCode status = U_ZERO_ERROR;
const UChar **names = NULL;
UResourceBundle* rbTable = NULL;
rbTable = ures_getByKeyWithFallback(rb, key, rbTable, &status);
if (U_SUCCESS(status)) {
names = (const UChar **)uprv_malloc(sizeof(const UChar*) * KEYS_SIZE);
if (names != NULL) {
UBool isEmpty = TRUE;
for (int32_t i = 0; i < KEYS_SIZE; i++) {
status = U_ZERO_ERROR;
int32_t len = 0;
const UChar *value = ures_getStringByKeyWithFallback(rbTable, KEYS[i], &len, &status);
if (U_FAILURE(status) || len == 0) {
names[i] = NULL;
} else {
names[i] = value;
isEmpty = FALSE;
}
}
if (isEmpty) {
// No need to keep the names array
uprv_free(names);
names = NULL;
}
}
}
ures_close(rbTable);
return names;
}
// ---------------------------------------------------
// TZNames - names for a time zone
// ---------------------------------------------------
class TZNames : public ZNames {
public:
virtual ~TZNames();
static TZNames* createInstance(UResourceBundle* rb, const char* key, const UnicodeString& tzID);
virtual const UChar* getName(UTimeZoneNameType type);
private:
TZNames(const UChar** names);
const UChar* fLocationName;
UChar* fLocationNameOwned;
};
TZNames::TZNames(const UChar** names)
: ZNames(names), fLocationName(NULL), fLocationNameOwned(NULL) {
}
TZNames::~TZNames() {
if (fLocationNameOwned) {
uprv_free(fLocationNameOwned);
}
}
const UChar*
TZNames::getName(UTimeZoneNameType type) {
if (type == UTZNM_EXEMPLAR_LOCATION) {
return fLocationName;
}
return ZNames::getName(type);
}
TZNames*
TZNames::createInstance(UResourceBundle* rb, const char* key, const UnicodeString& tzID) {
if (rb == NULL || key == NULL || *key == 0) {
return NULL;
}
const UChar** names = loadData(rb, key);
const UChar* locationName = NULL;
UChar* locationNameOwned = NULL;
UErrorCode status = U_ZERO_ERROR;
int32_t len = 0;
UResourceBundle* table = ures_getByKeyWithFallback(rb, key, NULL, &status);
locationName = ures_getStringByKeyWithFallback(table, gEcTag, &len, &status);
// ignore missing resource here
status = U_ZERO_ERROR;
ures_close(table);
if (locationName == NULL) {
UnicodeString tmpName;
int32_t tmpNameLen = 0;
TimeZoneNamesImpl::getDefaultExemplarLocationName(tzID, tmpName);
tmpNameLen = tmpName.length();
if (tmpNameLen > 0) {
locationNameOwned = (UChar*) uprv_malloc(sizeof(UChar) * (tmpNameLen + 1));
if (locationNameOwned) {
tmpName.extract(locationNameOwned, tmpNameLen + 1, status);
locationName = locationNameOwned;
}
}
}
TZNames* tznames = NULL;
if (locationName != NULL || names != NULL) {
tznames = new TZNames(names);
if (tznames == NULL) {
if (locationNameOwned) {
uprv_free(locationNameOwned);
}
}
tznames->fLocationName = locationName;
tznames->fLocationNameOwned = locationNameOwned;
}
return tznames;
}
// ---------------------------------------------------
// The meta zone ID enumeration class
// ---------------------------------------------------
class MetaZoneIDsEnumeration : public StringEnumeration {
public:
MetaZoneIDsEnumeration();
MetaZoneIDsEnumeration(const UVector& mzIDs);
MetaZoneIDsEnumeration(UVector* mzIDs);
virtual ~MetaZoneIDsEnumeration();
static UClassID U_EXPORT2 getStaticClassID(void);
virtual UClassID getDynamicClassID(void) const;
virtual const UnicodeString* snext(UErrorCode& status);
virtual void reset(UErrorCode& status);
virtual int32_t count(UErrorCode& status) const;
private:
int32_t fLen;
int32_t fPos;
const UVector* fMetaZoneIDs;
UVector *fLocalVector;
};
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(MetaZoneIDsEnumeration)
MetaZoneIDsEnumeration::MetaZoneIDsEnumeration()
: fLen(0), fPos(0), fMetaZoneIDs(NULL), fLocalVector(NULL) {
}
MetaZoneIDsEnumeration::MetaZoneIDsEnumeration(const UVector& mzIDs)
: fPos(0), fMetaZoneIDs(&mzIDs), fLocalVector(NULL) {
fLen = fMetaZoneIDs->size();
}
MetaZoneIDsEnumeration::MetaZoneIDsEnumeration(UVector *mzIDs)
: fLen(0), fPos(0), fMetaZoneIDs(mzIDs), fLocalVector(mzIDs) {
if (fMetaZoneIDs) {
fLen = fMetaZoneIDs->size();
}
}
const UnicodeString*
MetaZoneIDsEnumeration::snext(UErrorCode& status) {
if (U_SUCCESS(status) && fMetaZoneIDs != NULL && fPos < fLen) {
unistr.setTo((const UChar*)fMetaZoneIDs->elementAt(fPos++), -1);
return &unistr;
}
return NULL;
}
void
MetaZoneIDsEnumeration::reset(UErrorCode& /*status*/) {
fPos = 0;
}
int32_t
MetaZoneIDsEnumeration::count(UErrorCode& /*status*/) const {
return fLen;
}
MetaZoneIDsEnumeration::~MetaZoneIDsEnumeration() {
if (fLocalVector) {
delete fLocalVector;
}
}
U_CDECL_BEGIN
/**
* ZNameInfo stores zone name information in the trie
*/
typedef struct ZNameInfo {
UTimeZoneNameType type;
const UChar* tzID;
const UChar* mzID;
} ZNameInfo;
/**
* ZMatchInfo stores zone name match information used by find method
*/
typedef struct ZMatchInfo {
const ZNameInfo* znameInfo;
int32_t matchLength;
} ZMatchInfo;
U_CDECL_END
// ---------------------------------------------------
// ZNameSearchHandler
// ---------------------------------------------------
class ZNameSearchHandler : public TextTrieMapSearchResultHandler {
public:
ZNameSearchHandler(uint32_t types);
virtual ~ZNameSearchHandler();
UBool handleMatch(int32_t matchLength, const CharacterNode *node, UErrorCode &status);
TimeZoneNames::MatchInfoCollection* getMatches(int32_t& maxMatchLen);
private:
uint32_t fTypes;
int32_t fMaxMatchLen;
TimeZoneNames::MatchInfoCollection* fResults;
};
ZNameSearchHandler::ZNameSearchHandler(uint32_t types)
: fTypes(types), fMaxMatchLen(0), fResults(NULL) {
}
ZNameSearchHandler::~ZNameSearchHandler() {
if (fResults != NULL) {
delete fResults;
}
}
UBool
ZNameSearchHandler::handleMatch(int32_t matchLength, const CharacterNode *node, UErrorCode &status) {
if (U_FAILURE(status)) {
return FALSE;
}
if (node->hasValues()) {
int32_t valuesCount = node->countValues();
for (int32_t i = 0; i < valuesCount; i++) {
ZNameInfo *nameinfo = (ZNameInfo *)node->getValue(i);
if (nameinfo == NULL) {
continue;
}
if ((nameinfo->type & fTypes) != 0) {
// matches a requested type
if (fResults == NULL) {
fResults = new TimeZoneNames::MatchInfoCollection();
if (fResults == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
}
if (U_SUCCESS(status)) {
U_ASSERT(fResults != NULL);
if (nameinfo->tzID) {
fResults->addZone(nameinfo->type, matchLength, UnicodeString(nameinfo->tzID, -1), status);
} else {
U_ASSERT(nameinfo->mzID);
fResults->addMetaZone(nameinfo->type, matchLength, UnicodeString(nameinfo->mzID, -1), status);
}
if (U_SUCCESS(status) && matchLength > fMaxMatchLen) {
fMaxMatchLen = matchLength;
}
}
}
}
}
return TRUE;
}
TimeZoneNames::MatchInfoCollection*
ZNameSearchHandler::getMatches(int32_t& maxMatchLen) {
// give the ownership to the caller
TimeZoneNames::MatchInfoCollection* results = fResults;
maxMatchLen = fMaxMatchLen;
// reset
fResults = NULL;
fMaxMatchLen = 0;
return results;
}
// ---------------------------------------------------
// TimeZoneNamesImpl
//
// TimeZoneNames implementation class. This is the main
// part of this module.
// ---------------------------------------------------
U_CDECL_BEGIN
/**
* Deleter for ZNames
*/
static void U_CALLCONV
deleteZNames(void *obj) {
if (obj != EMPTY) {
delete (ZNames *)obj;
}
}
/**
* Deleter for TZNames
*/
static void U_CALLCONV
deleteTZNames(void *obj) {
if (obj != EMPTY) {
delete (TZNames *)obj;
}
}
/**
* Deleter for ZNameInfo
*/
static void U_CALLCONV
deleteZNameInfo(void *obj) {
uprv_free(obj);
}
U_CDECL_END
static UMutex gLock = U_MUTEX_INITIALIZER;
TimeZoneNamesImpl::TimeZoneNamesImpl(const Locale& locale, UErrorCode& status)
: fLocale(locale),
fZoneStrings(NULL),
fTZNamesMap(NULL),
fMZNamesMap(NULL),
fNamesTrieFullyLoaded(FALSE),
fNamesTrie(TRUE, deleteZNameInfo) {
initialize(locale, status);
}
void
TimeZoneNamesImpl::initialize(const Locale& locale, UErrorCode& status) {
if (U_FAILURE(status)) {
return;
}
// Load zoneStrings bundle
UErrorCode tmpsts = U_ZERO_ERROR; // OK with fallback warning..
fZoneStrings = ures_open(U_ICUDATA_ZONE, locale.getName(), &tmpsts);
fZoneStrings = ures_getByKeyWithFallback(fZoneStrings, gZoneStrings, fZoneStrings, &tmpsts);
if (U_FAILURE(tmpsts)) {
status = tmpsts;
cleanup();
return;
}
// Initialize hashtables holding time zone/meta zone names
fMZNamesMap = uhash_open(uhash_hashUChars, uhash_compareUChars, NULL, &status);
fTZNamesMap = uhash_open(uhash_hashUChars, uhash_compareUChars, NULL, &status);
if (U_FAILURE(status)) {
cleanup();
return;
}
uhash_setValueDeleter(fMZNamesMap, deleteZNames);
uhash_setValueDeleter(fTZNamesMap, deleteTZNames);
// no key deleters for name maps
// preload zone strings for the default zone
TimeZone *tz = TimeZone::createDefault();
const UChar *tzID = ZoneMeta::getCanonicalCLDRID(*tz);
if (tzID != NULL) {
loadStrings(UnicodeString(tzID));
}
delete tz;
return;
}
/*
* This method updates the cache and must be called with a lock,
* except initializer.
*/
void
TimeZoneNamesImpl::loadStrings(const UnicodeString& tzCanonicalID) {
loadTimeZoneNames(tzCanonicalID);
UErrorCode status = U_ZERO_ERROR;
StringEnumeration *mzIDs = getAvailableMetaZoneIDs(tzCanonicalID, status);
if (U_SUCCESS(status) && mzIDs != NULL) {
const UnicodeString *mzID;
while ((mzID = mzIDs->snext(status))) {
if (U_FAILURE(status)) {
break;
}
loadMetaZoneNames(*mzID);
}
delete mzIDs;
}
}
TimeZoneNamesImpl::~TimeZoneNamesImpl() {
cleanup();
}
void
TimeZoneNamesImpl::cleanup() {
if (fZoneStrings != NULL) {
ures_close(fZoneStrings);
fZoneStrings = NULL;
}
if (fMZNamesMap != NULL) {
uhash_close(fMZNamesMap);
fMZNamesMap = NULL;
}
if (fTZNamesMap != NULL) {
uhash_close(fTZNamesMap);
fTZNamesMap = NULL;
}
}
UBool
TimeZoneNamesImpl::operator==(const TimeZoneNames& other) const {
if (this == &other) {
return TRUE;
}
// No implementation for now
return FALSE;
}
TimeZoneNames*
TimeZoneNamesImpl::clone() const {
UErrorCode status = U_ZERO_ERROR;
return new TimeZoneNamesImpl(fLocale, status);
}
StringEnumeration*
TimeZoneNamesImpl::getAvailableMetaZoneIDs(UErrorCode& status) const {
return TimeZoneNamesImpl::_getAvailableMetaZoneIDs(status);
}
// static implementation of getAvailableMetaZoneIDs(UErrorCode&)
StringEnumeration*
TimeZoneNamesImpl::_getAvailableMetaZoneIDs(UErrorCode& status) {
if (U_FAILURE(status)) {
return NULL;
}
const UVector* mzIDs = ZoneMeta::getAvailableMetazoneIDs();
if (mzIDs == NULL) {
return new MetaZoneIDsEnumeration();
}
return new MetaZoneIDsEnumeration(*mzIDs);
}
StringEnumeration*
TimeZoneNamesImpl::getAvailableMetaZoneIDs(const UnicodeString& tzID, UErrorCode& status) const {
return TimeZoneNamesImpl::_getAvailableMetaZoneIDs(tzID, status);
}
// static implementation of getAvailableMetaZoneIDs(const UnicodeString&, UErrorCode&)
StringEnumeration*
TimeZoneNamesImpl::_getAvailableMetaZoneIDs(const UnicodeString& tzID, UErrorCode& status) {
if (U_FAILURE(status)) {
return NULL;
}
const UVector* mappings = ZoneMeta::getMetazoneMappings(tzID);
if (mappings == NULL) {
return new MetaZoneIDsEnumeration();
}
MetaZoneIDsEnumeration *senum = NULL;
UVector* mzIDs = new UVector(NULL, uhash_compareUChars, status);
if (mzIDs == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
if (U_SUCCESS(status)) {
U_ASSERT(mzIDs != NULL);
for (int32_t i = 0; U_SUCCESS(status) && i < mappings->size(); i++) {
OlsonToMetaMappingEntry *map = (OlsonToMetaMappingEntry *)mappings->elementAt(i);
const UChar *mzID = map->mzid;
if (!mzIDs->contains((void *)mzID)) {
mzIDs->addElement((void *)mzID, status);
}
}
if (U_SUCCESS(status)) {
senum = new MetaZoneIDsEnumeration(mzIDs);
} else {
delete mzIDs;
}
}
return senum;
}
UnicodeString&
TimeZoneNamesImpl::getMetaZoneID(const UnicodeString& tzID, UDate date, UnicodeString& mzID) const {
return TimeZoneNamesImpl::_getMetaZoneID(tzID, date, mzID);
}
// static implementation of getMetaZoneID
UnicodeString&
TimeZoneNamesImpl::_getMetaZoneID(const UnicodeString& tzID, UDate date, UnicodeString& mzID) {
ZoneMeta::getMetazoneID(tzID, date, mzID);
return mzID;
}
UnicodeString&
TimeZoneNamesImpl::getReferenceZoneID(const UnicodeString& mzID, const char* region, UnicodeString& tzID) const {
return TimeZoneNamesImpl::_getReferenceZoneID(mzID, region, tzID);
}
// static implementaion of getReferenceZoneID
UnicodeString&
TimeZoneNamesImpl::_getReferenceZoneID(const UnicodeString& mzID, const char* region, UnicodeString& tzID) {
ZoneMeta::getZoneIdByMetazone(mzID, UnicodeString(region, -1, US_INV), tzID);
return tzID;
}
UnicodeString&
TimeZoneNamesImpl::getMetaZoneDisplayName(const UnicodeString& mzID,
UTimeZoneNameType type,
UnicodeString& name) const {
name.setToBogus(); // cleanup result.
if (mzID.isEmpty()) {
return name;
}
ZNames *znames = NULL;
TimeZoneNamesImpl *nonConstThis = const_cast<TimeZoneNamesImpl *>(this);
umtx_lock(&gLock);
{
znames = nonConstThis->loadMetaZoneNames(mzID);
}
umtx_unlock(&gLock);
if (znames != NULL) {
const UChar* s = znames->getName(type);
if (s != NULL) {
name.setTo(TRUE, s, -1);
}
}
return name;
}
UnicodeString&
TimeZoneNamesImpl::getTimeZoneDisplayName(const UnicodeString& tzID, UTimeZoneNameType type, UnicodeString& name) const {
name.setToBogus(); // cleanup result.
if (tzID.isEmpty()) {
return name;
}
TZNames *tznames = NULL;
TimeZoneNamesImpl *nonConstThis = const_cast<TimeZoneNamesImpl *>(this);
umtx_lock(&gLock);
{
tznames = nonConstThis->loadTimeZoneNames(tzID);
}
umtx_unlock(&gLock);
if (tznames != NULL) {
const UChar *s = tznames->getName(type);
if (s != NULL) {
name.setTo(TRUE, s, -1);
}
}
return name;
}
UnicodeString&
TimeZoneNamesImpl::getExemplarLocationName(const UnicodeString& tzID, UnicodeString& name) const {
name.setToBogus(); // cleanup result.
const UChar* locName = NULL;
TZNames *tznames = NULL;
TimeZoneNamesImpl *nonConstThis = const_cast<TimeZoneNamesImpl *>(this);
umtx_lock(&gLock);
{
tznames = nonConstThis->loadTimeZoneNames(tzID);
}
umtx_unlock(&gLock);
if (tznames != NULL) {
locName = tznames->getName(UTZNM_EXEMPLAR_LOCATION);
}
if (locName != NULL) {
name.setTo(TRUE, locName, -1);
}
return name;
}
// Merge the MZ_PREFIX and mzId
static void mergeTimeZoneKey(const UnicodeString& mzID, char* result) {
if (mzID.isEmpty()) {
result[0] = '\0';
return;
}
char mzIdChar[ZID_KEY_MAX + 1];
int32_t keyLen;
int32_t prefixLen = uprv_strlen(gMZPrefix);
keyLen = mzID.extract(0, mzID.length(), mzIdChar, ZID_KEY_MAX + 1, US_INV);
uprv_memcpy((void *)result, (void *)gMZPrefix, prefixLen);
uprv_memcpy((void *)(result + prefixLen), (void *)mzIdChar, keyLen);
result[keyLen + prefixLen] = '\0';
}
/*
* This method updates the cache and must be called with a lock
*/
ZNames*
TimeZoneNamesImpl::loadMetaZoneNames(const UnicodeString& mzID) {
if (mzID.length() > (ZID_KEY_MAX - MZ_PREFIX_LEN)) {
return NULL;
}
ZNames *znames = NULL;
UErrorCode status = U_ZERO_ERROR;
UChar mzIDKey[ZID_KEY_MAX + 1];
mzID.extract(mzIDKey, ZID_KEY_MAX + 1, status);
U_ASSERT(status == U_ZERO_ERROR); // already checked length above
mzIDKey[mzID.length()] = 0;
void *cacheVal = uhash_get(fMZNamesMap, mzIDKey);
if (cacheVal == NULL) {
char key[ZID_KEY_MAX + 1];
mergeTimeZoneKey(mzID, key);
znames = ZNames::createInstance(fZoneStrings, key);
if (znames == NULL) {
cacheVal = (void *)EMPTY;
} else {
cacheVal = znames;
}
// Use the persistent ID as the resource key, so we can
// avoid duplications.
const UChar* newKey = ZoneMeta::findMetaZoneID(mzID);
if (newKey != NULL) {
uhash_put(fMZNamesMap, (void *)newKey, cacheVal, &status);
if (U_FAILURE(status)) {
if (znames != NULL) {
delete znames;
znames = NULL;
}
} else if (znames != NULL) {
// put the name info into the trie
for (int32_t i = 0; ALL_NAME_TYPES[i] != UTZNM_UNKNOWN; i++) {
const UChar* name = znames->getName(ALL_NAME_TYPES[i]);
if (name != NULL) {
ZNameInfo *nameinfo = (ZNameInfo *)uprv_malloc(sizeof(ZNameInfo));
if (nameinfo != NULL) {
nameinfo->type = ALL_NAME_TYPES[i];
nameinfo->tzID = NULL;
nameinfo->mzID = newKey;
fNamesTrie.put(name, nameinfo, status);
}
}
}
}
} else {
// Should never happen with a valid input
if (znames != NULL) {
// It's not possible that we get a valid ZNames with unknown ID.
// But just in case..
delete znames;
znames = NULL;
}
}
} else if (cacheVal != EMPTY) {
znames = (ZNames *)cacheVal;
}
return znames;
}
/*
* This method updates the cache and must be called with a lock
*/
TZNames*
TimeZoneNamesImpl::loadTimeZoneNames(const UnicodeString& tzID) {
if (tzID.length() > ZID_KEY_MAX) {
return NULL;
}
TZNames *tznames = NULL;
UErrorCode status = U_ZERO_ERROR;
UChar tzIDKey[ZID_KEY_MAX + 1];
int32_t tzIDKeyLen = tzID.extract(tzIDKey, ZID_KEY_MAX + 1, status);
U_ASSERT(status == U_ZERO_ERROR); // already checked length above
tzIDKey[tzIDKeyLen] = 0;
void *cacheVal = uhash_get(fTZNamesMap, tzIDKey);
if (cacheVal == NULL) {
char key[ZID_KEY_MAX + 1];
UErrorCode status = U_ZERO_ERROR;
// Replace "/" with ":".
UnicodeString uKey(tzID);
for (int32_t i = 0; i < uKey.length(); i++) {
if (uKey.charAt(i) == (UChar)0x2F) {
uKey.setCharAt(i, (UChar)0x3A);
}
}
uKey.extract(0, uKey.length(), key, sizeof(key), US_INV);
tznames = TZNames::createInstance(fZoneStrings, key, tzID);
if (tznames == NULL) {
cacheVal = (void *)EMPTY;
} else {
cacheVal = tznames;
}
// Use the persistent ID as the resource key, so we can
// avoid duplications.
const UChar* newKey = ZoneMeta::findTimeZoneID(tzID);
if (newKey != NULL) {
uhash_put(fTZNamesMap, (void *)newKey, cacheVal, &status);
if (U_FAILURE(status)) {
if (tznames != NULL) {
delete tznames;
tznames = NULL;
}
} else if (tznames != NULL) {
// put the name info into the trie
for (int32_t i = 0; ALL_NAME_TYPES[i] != UTZNM_UNKNOWN; i++) {
const UChar* name = tznames->getName(ALL_NAME_TYPES[i]);
if (name != NULL) {
ZNameInfo *nameinfo = (ZNameInfo *)uprv_malloc(sizeof(ZNameInfo));
if (nameinfo != NULL) {
nameinfo->type = ALL_NAME_TYPES[i];
nameinfo->tzID = newKey;
nameinfo->mzID = NULL;
fNamesTrie.put(name, nameinfo, status);
}
}
}
}
} else {
// Should never happen with a valid input
if (tznames != NULL) {
// It's not possible that we get a valid TZNames with unknown ID.
// But just in case..
delete tznames;
tznames = NULL;
}
}
} else if (cacheVal != EMPTY) {
tznames = (TZNames *)cacheVal;
}
return tznames;
}
TimeZoneNames::MatchInfoCollection*
TimeZoneNamesImpl::find(const UnicodeString& text, int32_t start, uint32_t types, UErrorCode& status) const {
ZNameSearchHandler handler(types);
TimeZoneNamesImpl *nonConstThis = const_cast<TimeZoneNamesImpl *>(this);
umtx_lock(&gLock);
{
fNamesTrie.search(text, start, (TextTrieMapSearchResultHandler *)&handler, status);
}
umtx_unlock(&gLock);
if (U_FAILURE(status)) {
return NULL;
}
int32_t maxLen = 0;
TimeZoneNames::MatchInfoCollection* matches = handler.getMatches(maxLen);
if (matches != NULL && ((maxLen == (text.length() - start)) || fNamesTrieFullyLoaded)) {
// perfect match
return matches;
}
delete matches;
// All names are not yet loaded into the trie
umtx_lock(&gLock);
{
if (!fNamesTrieFullyLoaded) {
const UnicodeString *id;
// load strings for all zones
StringEnumeration *tzIDs = TimeZone::createTimeZoneIDEnumeration(UCAL_ZONE_TYPE_CANONICAL, NULL, NULL, status);
if (U_SUCCESS(status)) {
while ((id = tzIDs->snext(status))) {
if (U_FAILURE(status)) {
break;
}
// loadStrings also load related metazone strings
nonConstThis->loadStrings(*id);
}
}
if (tzIDs != NULL) {
delete tzIDs;
}
if (U_SUCCESS(status)) {
nonConstThis->fNamesTrieFullyLoaded = TRUE;
}
}
}
umtx_unlock(&gLock);
if (U_FAILURE(status)) {
return NULL;
}
umtx_lock(&gLock);
{
// now try it again
fNamesTrie.search(text, start, (TextTrieMapSearchResultHandler *)&handler, status);
}
umtx_unlock(&gLock);
return handler.getMatches(maxLen);
}
static const UChar gEtcPrefix[] = { 0x45, 0x74, 0x63, 0x2F }; // "Etc/"
static const int32_t gEtcPrefixLen = 4;
static const UChar gSystemVPrefix[] = { 0x53, 0x79, 0x73, 0x74, 0x65, 0x6D, 0x56, 0x2F }; // "SystemV/
static const int32_t gSystemVPrefixLen = 8;
static const UChar gRiyadh8[] = { 0x52, 0x69, 0x79, 0x61, 0x64, 0x68, 0x38 }; // "Riyadh8"
static const int32_t gRiyadh8Len = 7;
UnicodeString& U_EXPORT2
TimeZoneNamesImpl::getDefaultExemplarLocationName(const UnicodeString& tzID, UnicodeString& name) {
if (tzID.isEmpty() || tzID.startsWith(gEtcPrefix, gEtcPrefixLen)
|| tzID.startsWith(gSystemVPrefix, gSystemVPrefixLen) || tzID.indexOf(gRiyadh8, gRiyadh8Len, 0) > 0) {
name.setToBogus();
return name;
}
int32_t sep = tzID.lastIndexOf((UChar)0x2F /* '/' */);
if (sep > 0 && sep + 1 < tzID.length()) {
name.setTo(tzID, sep + 1);
name.findAndReplace(UnicodeString((UChar)0x5f /* _ */),
UnicodeString((UChar)0x20 /* space */));
} else {
name.setToBogus();
}
return name;
}
// ---------------------------------------------------
// TZDBTimeZoneNames and its supporting classes
//
// TZDBTimeZoneNames is an implementation class of
// TimeZoneNames holding the IANA tz database abbreviations.
// ---------------------------------------------------
class TZDBNames : public UMemory {
public:
virtual ~TZDBNames();
static TZDBNames* createInstance(UResourceBundle* rb, const char* key);
const UChar* getName(UTimeZoneNameType type) const;
const char** getParseRegions(int32_t& numRegions) const;
protected:
TZDBNames(const UChar** names, char** regions, int32_t numRegions);
private:
const UChar** fNames;
char** fRegions;
int32_t fNumRegions;
};
TZDBNames::TZDBNames(const UChar** names, char** regions, int32_t numRegions)
: fNames(names),
fRegions(regions),
fNumRegions(numRegions) {
}
TZDBNames::~TZDBNames() {
if (fNames != NULL) {
uprv_free(fNames);
}
if (fRegions != NULL) {
char **p = fRegions;
for (int32_t i = 0; i < fNumRegions; p++, i++) {
uprv_free(*p);
}
uprv_free(fRegions);
}
}
TZDBNames*
TZDBNames::createInstance(UResourceBundle* rb, const char* key) {
if (rb == NULL || key == NULL || *key == 0) {
return NULL;
}
UErrorCode status = U_ZERO_ERROR;
const UChar **names = NULL;
char** regions = NULL;
int32_t numRegions = 0;
int32_t len = 0;
UResourceBundle* rbTable = NULL;
rbTable = ures_getByKey(rb, key, rbTable, &status);
if (U_FAILURE(status)) {
return NULL;
}
names = (const UChar **)uprv_malloc(sizeof(const UChar*) * TZDBNAMES_KEYS_SIZE);
UBool isEmpty = TRUE;
if (names != NULL) {
for (int32_t i = 0; i < TZDBNAMES_KEYS_SIZE; i++) {
status = U_ZERO_ERROR;
const UChar *value = ures_getStringByKey(rbTable, TZDBNAMES_KEYS[i], &len, &status);
if (U_FAILURE(status) || len == 0) {
names[i] = NULL;
} else {
names[i] = value;
isEmpty = FALSE;
}
}
}
if (isEmpty) {
if (names != NULL) {
uprv_free(names);
}
return NULL;
}
UResourceBundle *regionsRes = ures_getByKey(rbTable, "parseRegions", NULL, &status);
UBool regionError = FALSE;
if (U_SUCCESS(status)) {
numRegions = ures_getSize(regionsRes);
if (numRegions > 0) {
regions = (char**)uprv_malloc(sizeof(char*) * numRegions);
if (regions != NULL) {
char **pRegion = regions;
for (int32_t i = 0; i < numRegions; i++, pRegion++) {
*pRegion = NULL;
}
// filling regions
pRegion = regions;
for (int32_t i = 0; i < numRegions; i++, pRegion++) {
status = U_ZERO_ERROR;
const UChar *uregion = ures_getStringByIndex(regionsRes, i, &len, &status);
if (U_FAILURE(status)) {
regionError = TRUE;
break;
}
*pRegion = (char*)uprv_malloc(sizeof(char) * (len + 1));
if (*pRegion == NULL) {
regionError = TRUE;
break;
}
u_UCharsToChars(uregion, *pRegion, len);
(*pRegion)[len] = 0;
}
}
}
}
ures_close(regionsRes);
ures_close(rbTable);
if (regionError) {
if (names != NULL) {
uprv_free(names);
}
if (regions != NULL) {
char **p = regions;
for (int32_t i = 0; i < numRegions; p++, i++) {
uprv_free(p);
}
uprv_free(regions);
}
return NULL;
}
return new TZDBNames(names, regions, numRegions);
}
const UChar*
TZDBNames::getName(UTimeZoneNameType type) const {
if (fNames == NULL) {
return NULL;
}
const UChar *name = NULL;
switch(type) {
case UTZNM_SHORT_STANDARD:
name = fNames[0];
break;
case UTZNM_SHORT_DAYLIGHT:
name = fNames[1];
break;
default:
name = NULL;
}
return name;
}
const char**
TZDBNames::getParseRegions(int32_t& numRegions) const {
if (fRegions == NULL) {
numRegions = 0;
} else {
numRegions = fNumRegions;
}
return (const char**)fRegions;
}
U_CDECL_BEGIN
/**
* TZDBNameInfo stores metazone name information for the IANA abbreviations
* in the trie
*/
typedef struct TZDBNameInfo {
const UChar* mzID;
UTimeZoneNameType type;
UBool ambiguousType;
const char** parseRegions;
int32_t nRegions;
} TZDBNameInfo;
U_CDECL_END
class TZDBNameSearchHandler : public TextTrieMapSearchResultHandler {
public:
TZDBNameSearchHandler(uint32_t types, const char* region);
virtual ~TZDBNameSearchHandler();
UBool handleMatch(int32_t matchLength, const CharacterNode *node, UErrorCode &status);
TimeZoneNames::MatchInfoCollection* getMatches(int32_t& maxMatchLen);
private:
uint32_t fTypes;
int32_t fMaxMatchLen;
TimeZoneNames::MatchInfoCollection* fResults;
const char* fRegion;
};
TZDBNameSearchHandler::TZDBNameSearchHandler(uint32_t types, const char* region)
: fTypes(types), fMaxMatchLen(0), fResults(NULL), fRegion(region) {
}
TZDBNameSearchHandler::~TZDBNameSearchHandler() {
if (fResults != NULL) {
delete fResults;
}
}
UBool
TZDBNameSearchHandler::handleMatch(int32_t matchLength, const CharacterNode *node, UErrorCode &status) {
if (U_FAILURE(status)) {
return FALSE;
}
TZDBNameInfo *match = NULL;
TZDBNameInfo *defaultRegionMatch = NULL;
if (node->hasValues()) {
int32_t valuesCount = node->countValues();
for (int32_t i = 0; i < valuesCount; i++) {
TZDBNameInfo *ninfo = (TZDBNameInfo *)node->getValue(i);
if (ninfo == NULL) {
continue;
}
if ((ninfo->type & fTypes) != 0) {
// Some tz database abbreviations are ambiguous. For example,
// CST means either Central Standard Time or China Standard Time.
// Unlike CLDR time zone display names, this implementation
// does not use unique names. And TimeZoneFormat does not expect
// multiple results returned for the same time zone type.
// For this reason, this implementation resolve one among same
// zone type with a same name at this level.
if (ninfo->parseRegions == NULL) {
// parseRegions == null means this is the default metazone
// mapping for the abbreviation.
if (defaultRegionMatch == NULL) {
match = defaultRegionMatch = ninfo;
}
} else {
UBool matchRegion = FALSE;
// non-default metazone mapping for an abbreviation
// comes with applicable regions. For example, the default
// metazone mapping for "CST" is America_Central,
// but if region is one of CN/MO/TW, "CST" is parsed
// as metazone China (China Standard Time).
for (int32_t i = 0; i < ninfo->nRegions; i++) {
const char *region = ninfo->parseRegions[i];
if (uprv_strcmp(fRegion, region) == 0) {
match = ninfo;
matchRegion = TRUE;
break;
}
}
if (matchRegion) {
break;
}
if (match == NULL) {
match = ninfo;
}
}
}
}
if (match != NULL) {
UTimeZoneNameType ntype = match->type;
// Note: Workaround for duplicated standard/daylight names
// The tz database contains a few zones sharing a
// same name for both standard time and daylight saving
// time. For example, Australia/Sydney observes DST,
// but "EST" is used for both standard and daylight.
// When both SHORT_STANDARD and SHORT_DAYLIGHT are included
// in the find operation, we cannot tell which one was
// actually matched.
// TimeZoneFormat#parse returns a matched name type (standard
// or daylight) and DateFormat implementation uses the info to
// to adjust actual time. To avoid false type information,
// this implementation replaces the name type with SHORT_GENERIC.
if (match->ambiguousType
&& (ntype == UTZNM_SHORT_STANDARD || ntype == UTZNM_SHORT_DAYLIGHT)
&& (fTypes & UTZNM_SHORT_STANDARD) != 0
&& (fTypes & UTZNM_SHORT_DAYLIGHT) != 0) {
ntype = UTZNM_SHORT_GENERIC;
}
if (fResults == NULL) {
fResults = new TimeZoneNames::MatchInfoCollection();
if (fResults == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
}
if (U_SUCCESS(status)) {
U_ASSERT(fResults != NULL);
U_ASSERT(match->mzID != NULL);
fResults->addMetaZone(ntype, matchLength, UnicodeString(match->mzID, -1), status);
if (U_SUCCESS(status) && matchLength > fMaxMatchLen) {
fMaxMatchLen = matchLength;
}
}
}
}
return TRUE;
}
TimeZoneNames::MatchInfoCollection*
TZDBNameSearchHandler::getMatches(int32_t& maxMatchLen) {
// give the ownership to the caller
TimeZoneNames::MatchInfoCollection* results = fResults;
maxMatchLen = fMaxMatchLen;
// reset
fResults = NULL;
fMaxMatchLen = 0;
return results;
}
U_CDECL_BEGIN
/**
* Deleter for TZDBNames
*/
static void U_CALLCONV
deleteTZDBNames(void *obj) {
if (obj != EMPTY) {
delete (TZDBNames *)obj;
}
}
static void U_CALLCONV initTZDBNamesMap(UErrorCode &status) {
gTZDBNamesMap = uhash_open(uhash_hashUChars, uhash_compareUChars, NULL, &status);
if (U_FAILURE(status)) {
gTZDBNamesMap = NULL;
return;
}
// no key deleters for tzdb name maps
uhash_setValueDeleter(gTZDBNamesMap, deleteTZDBNames);
ucln_i18n_registerCleanup(UCLN_I18N_TZDBTIMEZONENAMES, tzdbTimeZoneNames_cleanup);
}
/**
* Deleter for TZDBNameInfo
*/
static void U_CALLCONV
deleteTZDBNameInfo(void *obj) {
if (obj != NULL) {
uprv_free(obj);
}
}
static void U_CALLCONV prepareFind(UErrorCode &status) {
if (U_FAILURE(status)) {
return;
}
gTZDBNamesTrie = new TextTrieMap(TRUE, deleteTZDBNameInfo);
if (gTZDBNamesTrie == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
const UnicodeString *mzID;
StringEnumeration *mzIDs = TimeZoneNamesImpl::_getAvailableMetaZoneIDs(status);
if (U_SUCCESS(status)) {
while ((mzID = mzIDs->snext(status)) && U_SUCCESS(status)) {
const TZDBNames *names = TZDBTimeZoneNames::getMetaZoneNames(*mzID, status);
if (names == NULL) {
continue;
}
const UChar *std = names->getName(UTZNM_SHORT_STANDARD);
const UChar *dst = names->getName(UTZNM_SHORT_DAYLIGHT);
if (std == NULL && dst == NULL) {
continue;
}
int32_t numRegions = 0;
const char **parseRegions = names->getParseRegions(numRegions);
// The tz database contains a few zones sharing a
// same name for both standard time and daylight saving
// time. For example, Australia/Sydney observes DST,
// but "EST" is used for both standard and daylight.
// we need to store the information for later processing.
UBool ambiguousType = (std != NULL && dst != NULL && u_strcmp(std, dst) == 0);
const UChar *uMzID = ZoneMeta::findMetaZoneID(*mzID);
if (std != NULL) {
TZDBNameInfo *stdInf = (TZDBNameInfo *)uprv_malloc(sizeof(TZDBNameInfo));
if (stdInf == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
break;
}
stdInf->mzID = uMzID;
stdInf->type = UTZNM_SHORT_STANDARD;
stdInf->ambiguousType = ambiguousType;
stdInf->parseRegions = parseRegions;
stdInf->nRegions = numRegions;
gTZDBNamesTrie->put(std, stdInf, status);
}
if (U_SUCCESS(status) && dst != NULL) {
TZDBNameInfo *dstInf = (TZDBNameInfo *)uprv_malloc(sizeof(TZDBNameInfo));
if (dstInf == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
break;
}
dstInf->mzID = uMzID;
dstInf->type = UTZNM_SHORT_DAYLIGHT;
dstInf->ambiguousType = ambiguousType;
dstInf->parseRegions = parseRegions;
dstInf->nRegions = numRegions;
gTZDBNamesTrie->put(dst, dstInf, status);
}
}
}
delete mzIDs;
if (U_FAILURE(status)) {
delete gTZDBNamesTrie;
gTZDBNamesTrie = NULL;
return;
}
ucln_i18n_registerCleanup(UCLN_I18N_TZDBTIMEZONENAMES, tzdbTimeZoneNames_cleanup);
}
U_CDECL_END
TZDBTimeZoneNames::TZDBTimeZoneNames(const Locale& locale)
: fLocale(locale) {
UBool useWorld = TRUE;
const char* region = fLocale.getCountry();
int32_t regionLen = uprv_strlen(region);
if (regionLen == 0) {
UErrorCode status = U_ZERO_ERROR;
char loc[ULOC_FULLNAME_CAPACITY];
uloc_addLikelySubtags(fLocale.getName(), loc, sizeof(loc), &status);
regionLen = uloc_getCountry(loc, fRegion, sizeof(fRegion), &status);
if (U_SUCCESS(status) && regionLen < (int32_t)sizeof(fRegion)) {
useWorld = FALSE;
}
} else if (regionLen < (int32_t)sizeof(fRegion)) {
uprv_strcpy(fRegion, region);
useWorld = FALSE;
}
if (useWorld) {
uprv_strcpy(fRegion, "001");
}
}
TZDBTimeZoneNames::~TZDBTimeZoneNames() {
}
UBool
TZDBTimeZoneNames::operator==(const TimeZoneNames& other) const {
if (this == &other) {
return TRUE;
}
// No implementation for now
return FALSE;
}
TimeZoneNames*
TZDBTimeZoneNames::clone() const {
return new TZDBTimeZoneNames(fLocale);
}
StringEnumeration*
TZDBTimeZoneNames::getAvailableMetaZoneIDs(UErrorCode& status) const {
return TimeZoneNamesImpl::_getAvailableMetaZoneIDs(status);
}
StringEnumeration*
TZDBTimeZoneNames::getAvailableMetaZoneIDs(const UnicodeString& tzID, UErrorCode& status) const {
return TimeZoneNamesImpl::_getAvailableMetaZoneIDs(tzID, status);
}
UnicodeString&
TZDBTimeZoneNames::getMetaZoneID(const UnicodeString& tzID, UDate date, UnicodeString& mzID) const {
return TimeZoneNamesImpl::_getMetaZoneID(tzID, date, mzID);
}
UnicodeString&
TZDBTimeZoneNames::getReferenceZoneID(const UnicodeString& mzID, const char* region, UnicodeString& tzID) const {
return TimeZoneNamesImpl::_getReferenceZoneID(mzID, region, tzID);
}
UnicodeString&
TZDBTimeZoneNames::getMetaZoneDisplayName(const UnicodeString& mzID,
UTimeZoneNameType type,
UnicodeString& name) const {
name.setToBogus();
if (mzID.isEmpty()) {
return name;
}
UErrorCode status = U_ZERO_ERROR;
const TZDBNames *tzdbNames = TZDBTimeZoneNames::getMetaZoneNames(mzID, status);
if (U_SUCCESS(status)) {
const UChar *s = tzdbNames->getName(type);
if (s != NULL) {
name.setTo(TRUE, s, -1);
}
}
return name;
}
UnicodeString&
TZDBTimeZoneNames::getTimeZoneDisplayName(const UnicodeString& /* tzID */, UTimeZoneNameType /* type */, UnicodeString& name) const {
// No abbreviations associated a zone directly for now.
name.setToBogus();
return name;
}
TZDBTimeZoneNames::MatchInfoCollection*
TZDBTimeZoneNames::find(const UnicodeString& text, int32_t start, uint32_t types, UErrorCode& status) const {
umtx_initOnce(gTZDBNamesTrieInitOnce, &prepareFind, status);
if (U_FAILURE(status)) {
return NULL;
}
TZDBNameSearchHandler handler(types, fRegion);
gTZDBNamesTrie->search(text, start, (TextTrieMapSearchResultHandler *)&handler, status);
if (U_FAILURE(status)) {
return NULL;
}
int32_t maxLen = 0;
return handler.getMatches(maxLen);
}
const TZDBNames*
TZDBTimeZoneNames::getMetaZoneNames(const UnicodeString& mzID, UErrorCode& status) {
umtx_initOnce(gTZDBNamesMapInitOnce, &initTZDBNamesMap, status);
if (U_FAILURE(status)) {
return NULL;
}
TZDBNames* tzdbNames = NULL;
UChar mzIDKey[ZID_KEY_MAX + 1];
mzID.extract(mzIDKey, ZID_KEY_MAX + 1, status);
U_ASSERT(status == U_ZERO_ERROR); // already checked length above
mzIDKey[mzID.length()] = 0;
umtx_lock(&gTZDBNamesMapLock);
{
void *cacheVal = uhash_get(gTZDBNamesMap, mzIDKey);
if (cacheVal == NULL) {
UResourceBundle *zoneStringsRes = ures_openDirect(U_ICUDATA_ZONE, "tzdbNames", &status);
zoneStringsRes = ures_getByKey(zoneStringsRes, gZoneStrings, zoneStringsRes, &status);
if (U_SUCCESS(status)) {
char key[ZID_KEY_MAX + 1];
mergeTimeZoneKey(mzID, key);
tzdbNames = TZDBNames::createInstance(zoneStringsRes, key);
if (tzdbNames == NULL) {
cacheVal = (void *)EMPTY;
} else {
cacheVal = tzdbNames;
}
// Use the persistent ID as the resource key, so we can
// avoid duplications.
const UChar* newKey = ZoneMeta::findMetaZoneID(mzID);
if (newKey != NULL) {
uhash_put(gTZDBNamesMap, (void *)newKey, cacheVal, &status);
if (U_FAILURE(status)) {
if (tzdbNames != NULL) {
delete tzdbNames;
tzdbNames = NULL;
}
}
} else {
// Should never happen with a valid input
if (tzdbNames != NULL) {
// It's not possible that we get a valid tzdbNames with unknown ID.
// But just in case..
delete tzdbNames;
tzdbNames = NULL;
}
}
}
ures_close(zoneStringsRes);
} else if (cacheVal != EMPTY) {
tzdbNames = (TZDBNames *)cacheVal;
}
}
umtx_unlock(&gTZDBNamesMapLock);
return tzdbNames;
}
U_NAMESPACE_END
#endif /* #if !UCONFIG_NO_FORMATTING */
//eof