scuffed-code/icu4c/source/common/locid.cpp

1101 lines
30 KiB
C++

/*
**********************************************************************
* Copyright (C) 1997-2011, International Business Machines
* Corporation and others. All Rights Reserved.
**********************************************************************
*
* File locid.cpp
*
* Created by: Richard Gillam
*
* Modification History:
*
* Date Name Description
* 02/11/97 aliu Changed gLocPath to fgDataDirectory and added
* methods to get and set it.
* 04/02/97 aliu Made operator!= inline; fixed return value
* of getName().
* 04/15/97 aliu Cleanup for AIX/Win32.
* 04/24/97 aliu Numerous changes per code review.
* 08/18/98 stephen Changed getDisplayName()
* Added SIMPLIFIED_CHINESE, TRADITIONAL_CHINESE
* Added getISOCountries(), getISOLanguages(),
* getLanguagesForCountry()
* 03/16/99 bertrand rehaul.
* 07/21/99 stephen Added U_CFUNC setDefault
* 11/09/99 weiv Added const char * getName() const;
* 04/12/00 srl removing unicodestring api's and cached hash code
* 08/10/01 grhoten Change the static Locales to accessor functions
******************************************************************************
*/
#include "unicode/locid.h"
#include "unicode/uloc.h"
#include "umutex.h"
#include "uassert.h"
#include "cmemory.h"
#include "cstring.h"
#include "uhash.h"
#include "ucln_cmn.h"
#include "ustr_imp.h"
#define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
typedef enum ELocalePos {
eENGLISH,
eFRENCH,
eGERMAN,
eITALIAN,
eJAPANESE,
eKOREAN,
eCHINESE,
eFRANCE,
eGERMANY,
eITALY,
eJAPAN,
eKOREA,
eCHINA, /* Alias for PRC */
eTAIWAN,
eUK,
eUS,
eCANADA,
eCANADA_FRENCH,
eROOT,
//eDEFAULT,
eMAX_LOCALES
} ELocalePos;
U_CFUNC int32_t locale_getKeywords(const char *localeID,
char prev,
char *keywords, int32_t keywordCapacity,
char *values, int32_t valuesCapacity, int32_t *valLen,
UBool valuesToo,
UErrorCode *status);
static U_NAMESPACE_QUALIFIER Locale *gLocaleCache = NULL;
static U_NAMESPACE_QUALIFIER Locale *gDefaultLocale = NULL;
static UHashtable *gDefaultLocalesHashT = NULL;
U_CDECL_BEGIN
//
// Deleter function for Locales owned by the default Locale hash table/
//
static void U_CALLCONV
deleteLocale(void *obj) {
delete (U_NAMESPACE_QUALIFIER Locale *) obj;
}
static UBool U_CALLCONV locale_cleanup(void)
{
U_NAMESPACE_USE
if (gLocaleCache) {
delete [] gLocaleCache;
gLocaleCache = NULL;
}
if (gDefaultLocalesHashT) {
uhash_close(gDefaultLocalesHashT); // Automatically deletes all elements, using deleter func.
gDefaultLocalesHashT = NULL;
}
else if (gDefaultLocale) {
// The cache wasn't created, and only one default locale was created.
delete gDefaultLocale;
}
gDefaultLocale = NULL;
return TRUE;
}
U_CDECL_END
U_NAMESPACE_BEGIN
//
// locale_set_default_internal.
//
void locale_set_default_internal(const char *id)
{
UErrorCode status = U_ZERO_ERROR;
UBool canonicalize = FALSE;
// If given a NULL string for the locale id, grab the default
// name from the system.
// (Different from most other locale APIs, where a null name means use
// the current ICU default locale.)
if (id == NULL) {
umtx_lock(NULL);
id = uprv_getDefaultLocaleID();
umtx_unlock(NULL);
canonicalize = TRUE; // always canonicalize host ID
}
// put the locale id into a canonical form,
// in preparation for looking up this locale in the hash table of
// already-created locale objects.
//
status = U_ZERO_ERROR;
char localeNameBuf[512];
if (canonicalize) {
uloc_canonicalize(id, localeNameBuf, sizeof(localeNameBuf)-1, &status);
} else {
uloc_getName(id, localeNameBuf, sizeof(localeNameBuf)-1, &status);
}
localeNameBuf[sizeof(localeNameBuf)-1] = 0; // Force null termination in event of
// a long name filling the buffer.
// (long names are truncated.)
// Lazy creation of the hash table itself, if needed.
UBool isOnlyLocale;
UMTX_CHECK(NULL, (gDefaultLocale == NULL), isOnlyLocale);
if (isOnlyLocale) {
// We haven't seen this locale id before.
// Create a new Locale object for it.
Locale *newFirstDefault = new Locale(Locale::eBOGUS);
if (newFirstDefault == NULL) {
// No way to report errors from here.
return;
}
newFirstDefault->init(localeNameBuf, FALSE);
umtx_lock(NULL);
if (gDefaultLocale == NULL) {
gDefaultLocale = newFirstDefault; // Assignment to gDefaultLocale must happen inside mutex
newFirstDefault = NULL;
ucln_common_registerCleanup(UCLN_COMMON_LOCALE, locale_cleanup);
}
// Else some other thread raced us through here, and set the new Locale.
// Use the hash table next.
umtx_unlock(NULL);
if (newFirstDefault == NULL) {
// We were successful in setting the locale, and we were the first one to set it.
return;
}
// else start using the hash table.
}
// Lazy creation of the hash table itself, if needed.
UBool hashTableNeedsInit;
UMTX_CHECK(NULL, (gDefaultLocalesHashT == NULL), hashTableNeedsInit);
if (hashTableNeedsInit) {
status = U_ZERO_ERROR;
UHashtable *tHashTable = uhash_open(uhash_hashChars, uhash_compareChars, NULL, &status);
if (U_FAILURE(status)) {
return;
}
uhash_setValueDeleter(tHashTable, deleteLocale);
umtx_lock(NULL);
if (gDefaultLocalesHashT == NULL) {
gDefaultLocalesHashT = tHashTable;
ucln_common_registerCleanup(UCLN_COMMON_LOCALE, locale_cleanup);
} else {
uhash_close(tHashTable);
hashTableNeedsInit = FALSE;
}
umtx_unlock(NULL);
}
// Hash table lookup, key is the locale full name
umtx_lock(NULL);
Locale *newDefault = (Locale *)uhash_get(gDefaultLocalesHashT, localeNameBuf);
if (newDefault != NULL) {
// We have the requested locale in the hash table already.
// Just set it as default. Inside the mutex lock, for those troublesome processors.
gDefaultLocale = newDefault;
umtx_unlock(NULL);
} else {
umtx_unlock(NULL);
// We haven't seen this locale id before.
// Create a new Locale object for it.
newDefault = new Locale(Locale::eBOGUS);
if (newDefault == NULL) {
// No way to report errors from here.
return;
}
newDefault->init(localeNameBuf, FALSE);
// Add newly created Locale to the hash table of default Locales
const char *key = newDefault->getName();
U_ASSERT(uprv_strcmp(key, localeNameBuf) == 0);
umtx_lock(NULL);
Locale *hashTableVal = (Locale *)uhash_get(gDefaultLocalesHashT, key);
if (hashTableVal == NULL) {
if (hashTableNeedsInit) {
// This is the second request to set the locale.
// Cache the first one.
uhash_put(gDefaultLocalesHashT, (void *)gDefaultLocale->getName(), gDefaultLocale, &status);
}
uhash_put(gDefaultLocalesHashT, (void *)key, newDefault, &status);
gDefaultLocale = newDefault;
// ignore errors from hash table insert. (Couldn't do anything anyway)
// We can still set the default Locale,
// it just wont be cached, and will eventually leak.
} else {
// Some other thread raced us through here, and got the new Locale
// into the hash table before us. Use that one.
gDefaultLocale = hashTableVal; // Assignment to gDefaultLocale must happen inside mutex
delete newDefault;
}
umtx_unlock(NULL);
}
}
U_NAMESPACE_END
/* sfb 07/21/99 */
U_CFUNC void
locale_set_default(const char *id)
{
U_NAMESPACE_USE
locale_set_default_internal(id);
}
/* end */
U_CFUNC const char *
locale_get_default(void)
{
U_NAMESPACE_USE
return Locale::getDefault().getName();
}
U_NAMESPACE_BEGIN
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(Locale)
/*Character separating the posix id fields*/
// '_'
// In the platform codepage.
#define SEP_CHAR '_'
Locale::~Locale()
{
/*if fullName is on the heap, we free it*/
if (fullName != fullNameBuffer)
{
uprv_free(fullName);
fullName = NULL;
}
if (baseName && baseName != baseNameBuffer) {
uprv_free(baseName);
baseName = NULL;
}
}
Locale::Locale()
: UObject(), fullName(fullNameBuffer), baseName(NULL)
{
init(NULL, FALSE);
}
/*
* Internal constructor to allow construction of a locale object with
* NO side effects. (Default constructor tries to get
* the default locale.)
*/
Locale::Locale(Locale::ELocaleType)
: UObject(), fullName(fullNameBuffer), baseName(NULL)
{
setToBogus();
}
Locale::Locale( const char * newLanguage,
const char * newCountry,
const char * newVariant,
const char * newKeywords)
: UObject(), fullName(fullNameBuffer), baseName(NULL)
{
if( (newLanguage==NULL) && (newCountry == NULL) && (newVariant == NULL) )
{
init(NULL, FALSE); /* shortcut */
}
else
{
MaybeStackArray<char, ULOC_FULLNAME_CAPACITY> togo;
int32_t size = 0;
int32_t lsize = 0;
int32_t csize = 0;
int32_t vsize = 0;
int32_t ksize = 0;
char *p;
// Calculate the size of the resulting string.
// Language
if ( newLanguage != NULL )
{
lsize = (int32_t)uprv_strlen(newLanguage);
size = lsize;
}
// _Country
if ( newCountry != NULL )
{
csize = (int32_t)uprv_strlen(newCountry);
size += csize;
}
// _Variant
if ( newVariant != NULL )
{
// remove leading _'s
while(newVariant[0] == SEP_CHAR)
{
newVariant++;
}
// remove trailing _'s
vsize = (int32_t)uprv_strlen(newVariant);
while( (vsize>1) && (newVariant[vsize-1] == SEP_CHAR) )
{
vsize--;
}
}
if( vsize > 0 )
{
size += vsize;
}
// Separator rules:
if ( vsize > 0 )
{
size += 2; // at least: __v
}
else if ( csize > 0 )
{
size += 1; // at least: _v
}
if ( newKeywords != NULL)
{
ksize = (int32_t)uprv_strlen(newKeywords);
size += ksize + 1;
}
// NOW we have the full locale string..
/*if the whole string is longer than our internal limit, we need
to go to the heap for temporary buffers*/
if (size >= togo.getCapacity())
{
// If togo_heap could not be created, initialize with default settings.
if (togo.resize(size+1) == NULL) {
init(NULL, FALSE);
}
}
togo[0] = 0;
// Now, copy it back.
p = togo.getAlias();
if ( lsize != 0 )
{
uprv_strcpy(p, newLanguage);
p += lsize;
}
if ( ( vsize != 0 ) || (csize != 0) ) // at least: __v
{ // ^
*p++ = SEP_CHAR;
}
if ( csize != 0 )
{
uprv_strcpy(p, newCountry);
p += csize;
}
if ( vsize != 0)
{
*p++ = SEP_CHAR; // at least: __v
uprv_strncpy(p, newVariant, vsize); // Must use strncpy because
p += vsize; // of trimming (above).
*p = 0; // terminate
}
if ( ksize != 0)
{
if (uprv_strchr(newKeywords, '=')) {
*p++ = '@'; /* keyword parsing */
}
else {
*p++ = '_'; /* Variant parsing with a script */
if ( vsize == 0) {
*p++ = '_'; /* No country found */
}
}
uprv_strcpy(p, newKeywords);
p += ksize;
}
// Parse it, because for example 'language' might really be a complete
// string.
init(togo.getAlias(), FALSE);
}
}
Locale::Locale(const Locale &other)
: UObject(other), fullName(fullNameBuffer), baseName(NULL)
{
*this = other;
}
Locale &Locale::operator=(const Locale &other)
{
if (this == &other) {
return *this;
}
if (&other == NULL) {
this->setToBogus();
return *this;
}
/* Free our current storage */
if(fullName != fullNameBuffer) {
uprv_free(fullName);
fullName = fullNameBuffer;
}
/* Allocate the full name if necessary */
if(other.fullName != other.fullNameBuffer) {
fullName = (char *)uprv_malloc(sizeof(char)*(uprv_strlen(other.fullName)+1));
if (fullName == NULL) {
return *this;
}
}
/* Copy the full name */
uprv_strcpy(fullName, other.fullName);
/* baseName is the cached result of getBaseName. if 'other' has a
baseName and it fits in baseNameBuffer, then copy it. otherwise set
it to NULL, and let the user lazy-create it (in getBaseName) if they
want it. */
if(baseName && baseName != baseNameBuffer) {
uprv_free(baseName);
}
baseName = NULL;
if(other.baseName == other.baseNameBuffer) {
uprv_strcpy(baseNameBuffer, other.baseNameBuffer);
baseName = baseNameBuffer;
}
/* Copy the language and country fields */
uprv_strcpy(language, other.language);
uprv_strcpy(script, other.script);
uprv_strcpy(country, other.country);
/* The variantBegin is an offset, just copy it */
variantBegin = other.variantBegin;
fIsBogus = other.fIsBogus;
return *this;
}
Locale *
Locale::clone() const {
return new Locale(*this);
}
UBool
Locale::operator==( const Locale& other) const
{
return (uprv_strcmp(other.fullName, fullName) == 0);
}
/*This function initializes a Locale from a C locale ID*/
Locale& Locale::init(const char* localeID, UBool canonicalize)
{
fIsBogus = FALSE;
/* Free our current storage */
if(fullName != fullNameBuffer) {
uprv_free(fullName);
fullName = fullNameBuffer;
}
if(baseName && baseName != baseNameBuffer) {
uprv_free(baseName);
baseName = NULL;
}
// not a loop:
// just an easy way to have a common error-exit
// without goto and without another function
do {
char *separator;
char *field[5] = {0};
int32_t fieldLen[5] = {0};
int32_t fieldIdx;
int32_t variantField;
int32_t length;
UErrorCode err;
if(localeID == NULL) {
// not an error, just set the default locale
return *this = getDefault();
}
/* preset all fields to empty */
language[0] = script[0] = country[0] = 0;
// "canonicalize" the locale ID to ICU/Java format
err = U_ZERO_ERROR;
length = canonicalize ?
uloc_canonicalize(localeID, fullName, sizeof(fullNameBuffer), &err) :
uloc_getName(localeID, fullName, sizeof(fullNameBuffer), &err);
if(err == U_BUFFER_OVERFLOW_ERROR || length >= (int32_t)sizeof(fullNameBuffer)) {
/*Go to heap for the fullName if necessary*/
fullName = (char *)uprv_malloc(sizeof(char)*(length + 1));
if(fullName == 0) {
fullName = fullNameBuffer;
break; // error: out of memory
}
err = U_ZERO_ERROR;
length = canonicalize ?
uloc_canonicalize(localeID, fullName, length+1, &err) :
uloc_getName(localeID, fullName, length+1, &err);
}
if(U_FAILURE(err) || err == U_STRING_NOT_TERMINATED_WARNING) {
/* should never occur */
break;
}
variantBegin = length;
/* after uloc_getName/canonicalize() we know that only '_' are separators */
separator = field[0] = fullName;
fieldIdx = 1;
while ((separator = uprv_strchr(field[fieldIdx-1], SEP_CHAR)) && fieldIdx < (int32_t)(sizeof(field)/sizeof(field[0]))-1) {
field[fieldIdx] = separator + 1;
fieldLen[fieldIdx-1] = (int32_t)(separator - field[fieldIdx-1]);
fieldIdx++;
}
// variant may contain @foo or .foo POSIX cruft; remove it
separator = uprv_strchr(field[fieldIdx-1], '@');
char* sep2 = uprv_strchr(field[fieldIdx-1], '.');
if (separator!=NULL || sep2!=NULL) {
if (separator==NULL || (sep2!=NULL && separator > sep2)) {
separator = sep2;
}
fieldLen[fieldIdx-1] = (int32_t)(separator - field[fieldIdx-1]);
} else {
fieldLen[fieldIdx-1] = length - (int32_t)(field[fieldIdx-1] - fullName);
}
if (fieldLen[0] >= (int32_t)(sizeof(language))
|| (fieldLen[1] == 4 && fieldLen[2] >= (int32_t)(sizeof(country)))
|| (fieldLen[1] != 4 && fieldLen[1] >= (int32_t)(sizeof(country))))
{
break; // error: one of the fields is too long
}
variantField = 2; /* Usually the 2nd one, except when a script is used. */
if (fieldLen[0] > 0) {
/* We have a language */
uprv_memcpy(language, fullName, fieldLen[0]);
language[fieldLen[0]] = 0;
}
if (fieldLen[1] == 4) {
/* We have at least a script */
uprv_memcpy(script, field[1], fieldLen[1]);
script[fieldLen[1]] = 0;
variantField = 3;
if (fieldLen[2] > 0) {
/* We have a country */
uprv_memcpy(country, field[2], fieldLen[2]);
country[fieldLen[2]] = 0;
}
}
else if (fieldLen[1] > 0) {
/* We have a country and no script */
uprv_memcpy(country, field[1], fieldLen[1]);
country[fieldLen[1]] = 0;
}
if (variantField > 0 && fieldLen[variantField] > 0) {
/* We have a variant */
variantBegin = (int32_t)(field[variantField] - fullName);
}
// successful end of init()
return *this;
} while(0); /*loop doesn't iterate*/
// when an error occurs, then set this object to "bogus" (there is no UErrorCode here)
setToBogus();
return *this;
}
int32_t
Locale::hashCode() const
{
return ustr_hashCharsN(fullName, uprv_strlen(fullName));
}
void
Locale::setToBogus() {
/* Free our current storage */
if(fullName != fullNameBuffer) {
uprv_free(fullName);
fullName = fullNameBuffer;
}
if(baseName && baseName != baseNameBuffer) {
uprv_free(baseName);
baseName = NULL;
}
*fullNameBuffer = 0;
*language = 0;
*script = 0;
*country = 0;
fIsBogus = TRUE;
}
const Locale& U_EXPORT2
Locale::getDefault()
{
const Locale *retLocale;
UMTX_CHECK(NULL, gDefaultLocale, retLocale);
if (retLocale == NULL) {
locale_set_default_internal(NULL);
umtx_lock(NULL);
// Need a mutex in case some other thread set a new
// default inbetween when we set and when we get the new default. For
// processors with weak memory coherency, we might not otherwise see all
// of the newly created new default locale.
retLocale = gDefaultLocale;
umtx_unlock(NULL);
}
return *retLocale;
}
void U_EXPORT2
Locale::setDefault( const Locale& newLocale,
UErrorCode& status)
{
if (U_FAILURE(status)) {
return;
}
/* Set the default from the full name string of the supplied locale.
* This is a convenient way to access the default locale caching mechanisms.
*/
const char *localeID = newLocale.getName();
locale_set_default_internal(localeID);
}
Locale U_EXPORT2
Locale::createFromName (const char *name)
{
if (name) {
Locale l("");
l.init(name, FALSE);
return l;
}
else {
return getDefault();
}
}
Locale U_EXPORT2
Locale::createCanonical(const char* name) {
Locale loc("");
loc.init(name, TRUE);
return loc;
}
const char *
Locale::getISO3Language() const
{
return uloc_getISO3Language(fullName);
}
const char *
Locale::getISO3Country() const
{
return uloc_getISO3Country(fullName);
}
/**
* Return the LCID value as specified in the "LocaleID" resource for this
* locale. The LocaleID must be expressed as a hexadecimal number, from
* one to four digits. If the LocaleID resource is not present, or is
* in an incorrect format, 0 is returned. The LocaleID is for use in
* Windows (it is an LCID), but is available on all platforms.
*/
uint32_t
Locale::getLCID() const
{
return uloc_getLCID(fullName);
}
const char* const* U_EXPORT2 Locale::getISOCountries()
{
return uloc_getISOCountries();
}
const char* const* U_EXPORT2 Locale::getISOLanguages()
{
return uloc_getISOLanguages();
}
// Set the locale's data based on a posix id.
void Locale::setFromPOSIXID(const char *posixID)
{
init(posixID, TRUE);
}
const Locale & U_EXPORT2
Locale::getRoot(void)
{
return getLocale(eROOT);
}
const Locale & U_EXPORT2
Locale::getEnglish(void)
{
return getLocale(eENGLISH);
}
const Locale & U_EXPORT2
Locale::getFrench(void)
{
return getLocale(eFRENCH);
}
const Locale & U_EXPORT2
Locale::getGerman(void)
{
return getLocale(eGERMAN);
}
const Locale & U_EXPORT2
Locale::getItalian(void)
{
return getLocale(eITALIAN);
}
const Locale & U_EXPORT2
Locale::getJapanese(void)
{
return getLocale(eJAPANESE);
}
const Locale & U_EXPORT2
Locale::getKorean(void)
{
return getLocale(eKOREAN);
}
const Locale & U_EXPORT2
Locale::getChinese(void)
{
return getLocale(eCHINESE);
}
const Locale & U_EXPORT2
Locale::getSimplifiedChinese(void)
{
return getLocale(eCHINA);
}
const Locale & U_EXPORT2
Locale::getTraditionalChinese(void)
{
return getLocale(eTAIWAN);
}
const Locale & U_EXPORT2
Locale::getFrance(void)
{
return getLocale(eFRANCE);
}
const Locale & U_EXPORT2
Locale::getGermany(void)
{
return getLocale(eGERMANY);
}
const Locale & U_EXPORT2
Locale::getItaly(void)
{
return getLocale(eITALY);
}
const Locale & U_EXPORT2
Locale::getJapan(void)
{
return getLocale(eJAPAN);
}
const Locale & U_EXPORT2
Locale::getKorea(void)
{
return getLocale(eKOREA);
}
const Locale & U_EXPORT2
Locale::getChina(void)
{
return getLocale(eCHINA);
}
const Locale & U_EXPORT2
Locale::getPRC(void)
{
return getLocale(eCHINA);
}
const Locale & U_EXPORT2
Locale::getTaiwan(void)
{
return getLocale(eTAIWAN);
}
const Locale & U_EXPORT2
Locale::getUK(void)
{
return getLocale(eUK);
}
const Locale & U_EXPORT2
Locale::getUS(void)
{
return getLocale(eUS);
}
const Locale & U_EXPORT2
Locale::getCanada(void)
{
return getLocale(eCANADA);
}
const Locale & U_EXPORT2
Locale::getCanadaFrench(void)
{
return getLocale(eCANADA_FRENCH);
}
const Locale &
Locale::getLocale(int locid)
{
Locale *localeCache = getLocaleCache();
U_ASSERT((locid < eMAX_LOCALES)&&(locid>=0));
if (localeCache == NULL) {
// Failure allocating the locale cache.
// The best we can do is return a NULL reference.
locid = 0;
}
return localeCache[locid]; /*operating on NULL*/
}
/*
This function is defined this way in order to get around static
initialization and static destruction.
*/
Locale *
Locale::getLocaleCache(void)
{
umtx_lock(NULL);
UBool needInit = (gLocaleCache == NULL);
umtx_unlock(NULL);
if (needInit) {
Locale *tLocaleCache = new Locale[(int)eMAX_LOCALES];
if (tLocaleCache == NULL) {
return NULL;
}
tLocaleCache[eROOT] = Locale("");
tLocaleCache[eENGLISH] = Locale("en");
tLocaleCache[eFRENCH] = Locale("fr");
tLocaleCache[eGERMAN] = Locale("de");
tLocaleCache[eITALIAN] = Locale("it");
tLocaleCache[eJAPANESE] = Locale("ja");
tLocaleCache[eKOREAN] = Locale("ko");
tLocaleCache[eCHINESE] = Locale("zh");
tLocaleCache[eFRANCE] = Locale("fr", "FR");
tLocaleCache[eGERMANY] = Locale("de", "DE");
tLocaleCache[eITALY] = Locale("it", "IT");
tLocaleCache[eJAPAN] = Locale("ja", "JP");
tLocaleCache[eKOREA] = Locale("ko", "KR");
tLocaleCache[eCHINA] = Locale("zh", "CN");
tLocaleCache[eTAIWAN] = Locale("zh", "TW");
tLocaleCache[eUK] = Locale("en", "GB");
tLocaleCache[eUS] = Locale("en", "US");
tLocaleCache[eCANADA] = Locale("en", "CA");
tLocaleCache[eCANADA_FRENCH] = Locale("fr", "CA");
umtx_lock(NULL);
if (gLocaleCache == NULL) {
gLocaleCache = tLocaleCache;
tLocaleCache = NULL;
ucln_common_registerCleanup(UCLN_COMMON_LOCALE, locale_cleanup);
}
umtx_unlock(NULL);
if (tLocaleCache) {
delete [] tLocaleCache; // Fancy array delete will destruct each member.
}
}
return gLocaleCache;
}
class KeywordEnumeration : public StringEnumeration {
private:
char *keywords;
char *current;
int32_t length;
UnicodeString currUSKey;
static const char fgClassID;/* Warning this is used beyond the typical RTTI usage. */
public:
static UClassID U_EXPORT2 getStaticClassID(void) { return (UClassID)&fgClassID; }
virtual UClassID getDynamicClassID(void) const { return getStaticClassID(); }
public:
KeywordEnumeration(const char *keys, int32_t keywordLen, int32_t currentIndex, UErrorCode &status)
: keywords((char *)&fgClassID), current((char *)&fgClassID), length(0) {
if(U_SUCCESS(status) && keywordLen != 0) {
if(keys == NULL || keywordLen < 0) {
status = U_ILLEGAL_ARGUMENT_ERROR;
} else {
keywords = (char *)uprv_malloc(keywordLen+1);
if (keywords == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
}
else {
uprv_memcpy(keywords, keys, keywordLen);
keywords[keywordLen] = 0;
current = keywords + currentIndex;
length = keywordLen;
}
}
}
}
virtual ~KeywordEnumeration() {
uprv_free(keywords);
}
virtual StringEnumeration * clone() const
{
UErrorCode status = U_ZERO_ERROR;
return new KeywordEnumeration(keywords, length, (int32_t)(current - keywords), status);
}
virtual int32_t count(UErrorCode &/*status*/) const {
char *kw = keywords;
int32_t result = 0;
while(*kw) {
result++;
kw += uprv_strlen(kw)+1;
}
return result;
}
virtual const char* next(int32_t* resultLength, UErrorCode& status) {
const char* result;
int32_t len;
if(U_SUCCESS(status) && *current != 0) {
result = current;
len = (int32_t)uprv_strlen(current);
current += len+1;
if(resultLength != NULL) {
*resultLength = len;
}
} else {
if(resultLength != NULL) {
*resultLength = 0;
}
result = NULL;
}
return result;
}
virtual const UnicodeString* snext(UErrorCode& status) {
int32_t resultLength = 0;
const char *s = next(&resultLength, status);
return setChars(s, resultLength, status);
}
virtual void reset(UErrorCode& /*status*/) {
current = keywords;
}
};
const char KeywordEnumeration::fgClassID = '\0';
StringEnumeration *
Locale::createKeywords(UErrorCode &status) const
{
char keywords[256];
int32_t keywordCapacity = 256;
StringEnumeration *result = NULL;
const char* variantStart = uprv_strchr(fullName, '@');
const char* assignment = uprv_strchr(fullName, '=');
if(variantStart) {
if(assignment > variantStart) {
int32_t keyLen = locale_getKeywords(variantStart+1, '@', keywords, keywordCapacity, NULL, 0, NULL, FALSE, &status);
if(keyLen) {
result = new KeywordEnumeration(keywords, keyLen, 0, status);
}
} else {
status = U_INVALID_FORMAT_ERROR;
}
}
return result;
}
int32_t
Locale::getKeywordValue(const char* keywordName, char *buffer, int32_t bufLen, UErrorCode &status) const
{
return uloc_getKeywordValue(fullName, keywordName, buffer, bufLen, &status);
}
void
Locale::setKeywordValue(const char* keywordName, const char* keywordValue, UErrorCode &status)
{
uloc_setKeywordValue(keywordName, keywordValue, fullName, ULOC_FULLNAME_CAPACITY, &status);
}
const char *
Locale::getBaseName() const
{
// lazy init
UErrorCode status = U_ZERO_ERROR;
// semantically const
if(baseName == 0) {
((Locale *)this)->baseName = ((Locale *)this)->baseNameBuffer;
int32_t baseNameSize = uloc_getBaseName(fullName, baseName, ULOC_FULLNAME_CAPACITY, &status);
if(baseNameSize >= ULOC_FULLNAME_CAPACITY) {
((Locale *)this)->baseName = (char *)uprv_malloc(sizeof(char) * baseNameSize + 1);
if (baseName == NULL) {
return baseName;
}
uloc_getBaseName(fullName, baseName, baseNameSize+1, &status);
}
baseName[baseNameSize] = 0;
// the computation of variantBegin leaves it equal to the length
// of fullName if there is no variant. It should instead be
// the length of the baseName. Patch around this for now.
if (variantBegin == (int32_t)uprv_strlen(fullName)) {
((Locale*)this)->variantBegin = baseNameSize;
}
}
return baseName;
}
//eof
U_NAMESPACE_END