git-svn-id: https://svn.wxwidgets.org/svn/wx/wxWidgets/trunk@58757 c3d73ce0-8a6f-49c7-b76d-6d57e0e08775
3565 lines
124 KiB
C++
3565 lines
124 KiB
C++
/////////////////////////////////////////////////////////////////////////////
|
|
// Name: src/common/intl.cpp
|
|
// Purpose: Internationalization and localisation for wxWidgets
|
|
// Author: Vadim Zeitlin
|
|
// Modified by: Michael N. Filippov <michael@idisys.iae.nsk.su>
|
|
// (2003/09/30 - PluralForms support)
|
|
// Created: 29/01/98
|
|
// RCS-ID: $Id$
|
|
// Copyright: (c) 1998 Vadim Zeitlin <zeitlin@dptmaths.ens-cachan.fr>
|
|
// Licence: wxWindows licence
|
|
/////////////////////////////////////////////////////////////////////////////
|
|
|
|
// ============================================================================
|
|
// declaration
|
|
// ============================================================================
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// headers
|
|
// ----------------------------------------------------------------------------
|
|
|
|
// For compilers that support precompilation, includes "wx.h".
|
|
#include "wx/wxprec.h"
|
|
|
|
#ifdef __BORLANDC__
|
|
#pragma hdrstop
|
|
#endif
|
|
|
|
#ifdef __EMX__
|
|
// The following define is needed by Innotek's libc to
|
|
// make the definition of struct localeconv available.
|
|
#define __INTERNAL_DEFS
|
|
#endif
|
|
|
|
#if wxUSE_INTL
|
|
|
|
#ifndef WX_PRECOMP
|
|
#include "wx/dynarray.h"
|
|
#include "wx/string.h"
|
|
#include "wx/intl.h"
|
|
#include "wx/log.h"
|
|
#include "wx/utils.h"
|
|
#include "wx/app.h"
|
|
#include "wx/hashmap.h"
|
|
#include "wx/module.h"
|
|
#endif // WX_PRECOMP
|
|
|
|
#ifndef __WXWINCE__
|
|
#include <locale.h>
|
|
#endif
|
|
|
|
// standard headers
|
|
#include <ctype.h>
|
|
#include <stdlib.h>
|
|
#ifdef HAVE_LANGINFO_H
|
|
#include <langinfo.h>
|
|
#endif
|
|
|
|
#ifdef __WIN32__
|
|
#include "wx/msw/private.h"
|
|
#elif defined(__UNIX_LIKE__)
|
|
#include "wx/fontmap.h" // for CharsetToEncoding()
|
|
#endif
|
|
|
|
#include "wx/file.h"
|
|
#include "wx/filename.h"
|
|
#include "wx/tokenzr.h"
|
|
#include "wx/fontmap.h"
|
|
#include "wx/encconv.h"
|
|
#include "wx/scopedptr.h"
|
|
#include "wx/apptrait.h"
|
|
#include "wx/stdpaths.h"
|
|
#include "wx/hashset.h"
|
|
#include "wx/filesys.h"
|
|
|
|
#if defined(__DARWIN__)
|
|
#include "wx/osx/core/cfref.h"
|
|
#include <CoreFoundation/CFLocale.h>
|
|
#include "wx/osx/core/cfstring.h"
|
|
#endif
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// simple types
|
|
// ----------------------------------------------------------------------------
|
|
|
|
// this should *not* be wxChar, this type must have exactly 8 bits!
|
|
typedef wxUint8 size_t8;
|
|
typedef wxUint32 size_t32;
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// constants
|
|
// ----------------------------------------------------------------------------
|
|
|
|
// magic number identifying the .mo format file
|
|
const size_t32 MSGCATALOG_MAGIC = 0x950412de;
|
|
const size_t32 MSGCATALOG_MAGIC_SW = 0xde120495;
|
|
|
|
// the constants describing the format of ll_CC locale string
|
|
static const size_t LEN_LANG = 2;
|
|
static const size_t LEN_SUBLANG = 2;
|
|
static const size_t LEN_FULL = LEN_LANG + 1 + LEN_SUBLANG; // 1 for '_'
|
|
|
|
#define TRACE_I18N wxS("i18n")
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// global functions
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#ifdef __WXDEBUG__
|
|
|
|
// small class to suppress the translation erros until exit from current scope
|
|
class NoTransErr
|
|
{
|
|
public:
|
|
NoTransErr() { ms_suppressCount++; }
|
|
~NoTransErr() { ms_suppressCount--; }
|
|
|
|
static bool Suppress() { return ms_suppressCount > 0; }
|
|
|
|
private:
|
|
static size_t ms_suppressCount;
|
|
};
|
|
|
|
size_t NoTransErr::ms_suppressCount = 0;
|
|
|
|
#else // !Debug
|
|
|
|
class NoTransErr
|
|
{
|
|
public:
|
|
NoTransErr() { }
|
|
~NoTransErr() { }
|
|
};
|
|
|
|
#endif // Debug/!Debug
|
|
|
|
static wxLocale *wxSetLocale(wxLocale *pLocale);
|
|
|
|
namespace
|
|
{
|
|
|
|
// get just the language part
|
|
inline wxString ExtractLang(const wxString& langFull)
|
|
{
|
|
return langFull.Left(LEN_LANG);
|
|
}
|
|
|
|
// helper functions of GetSystemLanguage()
|
|
#ifdef __UNIX__
|
|
|
|
// get everything else (including the leading '_')
|
|
inline wxString ExtractNotLang(const wxString& langFull)
|
|
{
|
|
return langFull.Mid(LEN_LANG);
|
|
}
|
|
|
|
#endif // __UNIX__
|
|
|
|
} // anonymous namespace
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// Plural forms parser
|
|
// ----------------------------------------------------------------------------
|
|
|
|
/*
|
|
Simplified Grammar
|
|
|
|
Expression:
|
|
LogicalOrExpression '?' Expression ':' Expression
|
|
LogicalOrExpression
|
|
|
|
LogicalOrExpression:
|
|
LogicalAndExpression "||" LogicalOrExpression // to (a || b) || c
|
|
LogicalAndExpression
|
|
|
|
LogicalAndExpression:
|
|
EqualityExpression "&&" LogicalAndExpression // to (a && b) && c
|
|
EqualityExpression
|
|
|
|
EqualityExpression:
|
|
RelationalExpression "==" RelationalExperession
|
|
RelationalExpression "!=" RelationalExperession
|
|
RelationalExpression
|
|
|
|
RelationalExpression:
|
|
MultiplicativeExpression '>' MultiplicativeExpression
|
|
MultiplicativeExpression '<' MultiplicativeExpression
|
|
MultiplicativeExpression ">=" MultiplicativeExpression
|
|
MultiplicativeExpression "<=" MultiplicativeExpression
|
|
MultiplicativeExpression
|
|
|
|
MultiplicativeExpression:
|
|
PmExpression '%' PmExpression
|
|
PmExpression
|
|
|
|
PmExpression:
|
|
N
|
|
Number
|
|
'(' Expression ')'
|
|
*/
|
|
|
|
class wxPluralFormsToken
|
|
{
|
|
public:
|
|
enum Type
|
|
{
|
|
T_ERROR, T_EOF, T_NUMBER, T_N, T_PLURAL, T_NPLURALS, T_EQUAL, T_ASSIGN,
|
|
T_GREATER, T_GREATER_OR_EQUAL, T_LESS, T_LESS_OR_EQUAL,
|
|
T_REMINDER, T_NOT_EQUAL,
|
|
T_LOGICAL_AND, T_LOGICAL_OR, T_QUESTION, T_COLON, T_SEMICOLON,
|
|
T_LEFT_BRACKET, T_RIGHT_BRACKET
|
|
};
|
|
Type type() const { return m_type; }
|
|
void setType(Type type) { m_type = type; }
|
|
// for T_NUMBER only
|
|
typedef int Number;
|
|
Number number() const { return m_number; }
|
|
void setNumber(Number num) { m_number = num; }
|
|
private:
|
|
Type m_type;
|
|
Number m_number;
|
|
};
|
|
|
|
|
|
class wxPluralFormsScanner
|
|
{
|
|
public:
|
|
wxPluralFormsScanner(const char* s);
|
|
const wxPluralFormsToken& token() const { return m_token; }
|
|
bool nextToken(); // returns false if error
|
|
private:
|
|
const char* m_s;
|
|
wxPluralFormsToken m_token;
|
|
};
|
|
|
|
wxPluralFormsScanner::wxPluralFormsScanner(const char* s) : m_s(s)
|
|
{
|
|
nextToken();
|
|
}
|
|
|
|
bool wxPluralFormsScanner::nextToken()
|
|
{
|
|
wxPluralFormsToken::Type type = wxPluralFormsToken::T_ERROR;
|
|
while (isspace((unsigned char) *m_s))
|
|
{
|
|
++m_s;
|
|
}
|
|
if (*m_s == 0)
|
|
{
|
|
type = wxPluralFormsToken::T_EOF;
|
|
}
|
|
else if (isdigit((unsigned char) *m_s))
|
|
{
|
|
wxPluralFormsToken::Number number = *m_s++ - '0';
|
|
while (isdigit((unsigned char) *m_s))
|
|
{
|
|
number = number * 10 + (*m_s++ - '0');
|
|
}
|
|
m_token.setNumber(number);
|
|
type = wxPluralFormsToken::T_NUMBER;
|
|
}
|
|
else if (isalpha((unsigned char) *m_s))
|
|
{
|
|
const char* begin = m_s++;
|
|
while (isalnum((unsigned char) *m_s))
|
|
{
|
|
++m_s;
|
|
}
|
|
size_t size = m_s - begin;
|
|
if (size == 1 && memcmp(begin, "n", size) == 0)
|
|
{
|
|
type = wxPluralFormsToken::T_N;
|
|
}
|
|
else if (size == 6 && memcmp(begin, "plural", size) == 0)
|
|
{
|
|
type = wxPluralFormsToken::T_PLURAL;
|
|
}
|
|
else if (size == 8 && memcmp(begin, "nplurals", size) == 0)
|
|
{
|
|
type = wxPluralFormsToken::T_NPLURALS;
|
|
}
|
|
}
|
|
else if (*m_s == '=')
|
|
{
|
|
++m_s;
|
|
if (*m_s == '=')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_EQUAL;
|
|
}
|
|
else
|
|
{
|
|
type = wxPluralFormsToken::T_ASSIGN;
|
|
}
|
|
}
|
|
else if (*m_s == '>')
|
|
{
|
|
++m_s;
|
|
if (*m_s == '=')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_GREATER_OR_EQUAL;
|
|
}
|
|
else
|
|
{
|
|
type = wxPluralFormsToken::T_GREATER;
|
|
}
|
|
}
|
|
else if (*m_s == '<')
|
|
{
|
|
++m_s;
|
|
if (*m_s == '=')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_LESS_OR_EQUAL;
|
|
}
|
|
else
|
|
{
|
|
type = wxPluralFormsToken::T_LESS;
|
|
}
|
|
}
|
|
else if (*m_s == '%')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_REMINDER;
|
|
}
|
|
else if (*m_s == '!' && m_s[1] == '=')
|
|
{
|
|
m_s += 2;
|
|
type = wxPluralFormsToken::T_NOT_EQUAL;
|
|
}
|
|
else if (*m_s == '&' && m_s[1] == '&')
|
|
{
|
|
m_s += 2;
|
|
type = wxPluralFormsToken::T_LOGICAL_AND;
|
|
}
|
|
else if (*m_s == '|' && m_s[1] == '|')
|
|
{
|
|
m_s += 2;
|
|
type = wxPluralFormsToken::T_LOGICAL_OR;
|
|
}
|
|
else if (*m_s == '?')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_QUESTION;
|
|
}
|
|
else if (*m_s == ':')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_COLON;
|
|
} else if (*m_s == ';') {
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_SEMICOLON;
|
|
}
|
|
else if (*m_s == '(')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_LEFT_BRACKET;
|
|
}
|
|
else if (*m_s == ')')
|
|
{
|
|
++m_s;
|
|
type = wxPluralFormsToken::T_RIGHT_BRACKET;
|
|
}
|
|
m_token.setType(type);
|
|
return type != wxPluralFormsToken::T_ERROR;
|
|
}
|
|
|
|
class wxPluralFormsNode;
|
|
|
|
// NB: Can't use wxDEFINE_SCOPED_PTR_TYPE because wxPluralFormsNode is not
|
|
// fully defined yet:
|
|
class wxPluralFormsNodePtr
|
|
{
|
|
public:
|
|
wxPluralFormsNodePtr(wxPluralFormsNode *p = NULL) : m_p(p) {}
|
|
~wxPluralFormsNodePtr();
|
|
wxPluralFormsNode& operator*() const { return *m_p; }
|
|
wxPluralFormsNode* operator->() const { return m_p; }
|
|
wxPluralFormsNode* get() const { return m_p; }
|
|
wxPluralFormsNode* release();
|
|
void reset(wxPluralFormsNode *p);
|
|
|
|
private:
|
|
wxPluralFormsNode *m_p;
|
|
};
|
|
|
|
class wxPluralFormsNode
|
|
{
|
|
public:
|
|
wxPluralFormsNode(const wxPluralFormsToken& token) : m_token(token) {}
|
|
const wxPluralFormsToken& token() const { return m_token; }
|
|
const wxPluralFormsNode* node(size_t i) const
|
|
{ return m_nodes[i].get(); }
|
|
void setNode(size_t i, wxPluralFormsNode* n);
|
|
wxPluralFormsNode* releaseNode(size_t i);
|
|
wxPluralFormsToken::Number evaluate(wxPluralFormsToken::Number n) const;
|
|
|
|
private:
|
|
wxPluralFormsToken m_token;
|
|
wxPluralFormsNodePtr m_nodes[3];
|
|
};
|
|
|
|
wxPluralFormsNodePtr::~wxPluralFormsNodePtr()
|
|
{
|
|
delete m_p;
|
|
}
|
|
wxPluralFormsNode* wxPluralFormsNodePtr::release()
|
|
{
|
|
wxPluralFormsNode *p = m_p;
|
|
m_p = NULL;
|
|
return p;
|
|
}
|
|
void wxPluralFormsNodePtr::reset(wxPluralFormsNode *p)
|
|
{
|
|
if (p != m_p)
|
|
{
|
|
delete m_p;
|
|
m_p = p;
|
|
}
|
|
}
|
|
|
|
|
|
void wxPluralFormsNode::setNode(size_t i, wxPluralFormsNode* n)
|
|
{
|
|
m_nodes[i].reset(n);
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsNode::releaseNode(size_t i)
|
|
{
|
|
return m_nodes[i].release();
|
|
}
|
|
|
|
wxPluralFormsToken::Number
|
|
wxPluralFormsNode::evaluate(wxPluralFormsToken::Number n) const
|
|
{
|
|
switch (token().type())
|
|
{
|
|
// leaf
|
|
case wxPluralFormsToken::T_NUMBER:
|
|
return token().number();
|
|
case wxPluralFormsToken::T_N:
|
|
return n;
|
|
// 2 args
|
|
case wxPluralFormsToken::T_EQUAL:
|
|
return node(0)->evaluate(n) == node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_NOT_EQUAL:
|
|
return node(0)->evaluate(n) != node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_GREATER:
|
|
return node(0)->evaluate(n) > node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_GREATER_OR_EQUAL:
|
|
return node(0)->evaluate(n) >= node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_LESS:
|
|
return node(0)->evaluate(n) < node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_LESS_OR_EQUAL:
|
|
return node(0)->evaluate(n) <= node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_REMINDER:
|
|
{
|
|
wxPluralFormsToken::Number number = node(1)->evaluate(n);
|
|
if (number != 0)
|
|
{
|
|
return node(0)->evaluate(n) % number;
|
|
}
|
|
else
|
|
{
|
|
return 0;
|
|
}
|
|
}
|
|
case wxPluralFormsToken::T_LOGICAL_AND:
|
|
return node(0)->evaluate(n) && node(1)->evaluate(n);
|
|
case wxPluralFormsToken::T_LOGICAL_OR:
|
|
return node(0)->evaluate(n) || node(1)->evaluate(n);
|
|
// 3 args
|
|
case wxPluralFormsToken::T_QUESTION:
|
|
return node(0)->evaluate(n)
|
|
? node(1)->evaluate(n)
|
|
: node(2)->evaluate(n);
|
|
default:
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
|
|
class wxPluralFormsCalculator
|
|
{
|
|
public:
|
|
wxPluralFormsCalculator() : m_nplurals(0), m_plural(0) {}
|
|
|
|
// input: number, returns msgstr index
|
|
int evaluate(int n) const;
|
|
|
|
// input: text after "Plural-Forms:" (e.g. "nplurals=2; plural=(n != 1);"),
|
|
// if s == 0, creates default handler
|
|
// returns 0 if error
|
|
static wxPluralFormsCalculator* make(const char* s = 0);
|
|
|
|
~wxPluralFormsCalculator() {}
|
|
|
|
void init(wxPluralFormsToken::Number nplurals, wxPluralFormsNode* plural);
|
|
|
|
private:
|
|
wxPluralFormsToken::Number m_nplurals;
|
|
wxPluralFormsNodePtr m_plural;
|
|
};
|
|
|
|
wxDEFINE_SCOPED_PTR_TYPE(wxPluralFormsCalculator)
|
|
|
|
void wxPluralFormsCalculator::init(wxPluralFormsToken::Number nplurals,
|
|
wxPluralFormsNode* plural)
|
|
{
|
|
m_nplurals = nplurals;
|
|
m_plural.reset(plural);
|
|
}
|
|
|
|
int wxPluralFormsCalculator::evaluate(int n) const
|
|
{
|
|
if (m_plural.get() == 0)
|
|
{
|
|
return 0;
|
|
}
|
|
wxPluralFormsToken::Number number = m_plural->evaluate(n);
|
|
if (number < 0 || number > m_nplurals)
|
|
{
|
|
return 0;
|
|
}
|
|
return number;
|
|
}
|
|
|
|
|
|
class wxPluralFormsParser
|
|
{
|
|
public:
|
|
wxPluralFormsParser(wxPluralFormsScanner& scanner) : m_scanner(scanner) {}
|
|
bool parse(wxPluralFormsCalculator& rCalculator);
|
|
|
|
private:
|
|
wxPluralFormsNode* parsePlural();
|
|
// stops at T_SEMICOLON, returns 0 if error
|
|
wxPluralFormsScanner& m_scanner;
|
|
const wxPluralFormsToken& token() const;
|
|
bool nextToken();
|
|
|
|
wxPluralFormsNode* expression();
|
|
wxPluralFormsNode* logicalOrExpression();
|
|
wxPluralFormsNode* logicalAndExpression();
|
|
wxPluralFormsNode* equalityExpression();
|
|
wxPluralFormsNode* multiplicativeExpression();
|
|
wxPluralFormsNode* relationalExpression();
|
|
wxPluralFormsNode* pmExpression();
|
|
};
|
|
|
|
bool wxPluralFormsParser::parse(wxPluralFormsCalculator& rCalculator)
|
|
{
|
|
if (token().type() != wxPluralFormsToken::T_NPLURALS)
|
|
return false;
|
|
if (!nextToken())
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_ASSIGN)
|
|
return false;
|
|
if (!nextToken())
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_NUMBER)
|
|
return false;
|
|
wxPluralFormsToken::Number nplurals = token().number();
|
|
if (!nextToken())
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_SEMICOLON)
|
|
return false;
|
|
if (!nextToken())
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_PLURAL)
|
|
return false;
|
|
if (!nextToken())
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_ASSIGN)
|
|
return false;
|
|
if (!nextToken())
|
|
return false;
|
|
wxPluralFormsNode* plural = parsePlural();
|
|
if (plural == 0)
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_SEMICOLON)
|
|
return false;
|
|
if (!nextToken())
|
|
return false;
|
|
if (token().type() != wxPluralFormsToken::T_EOF)
|
|
return false;
|
|
rCalculator.init(nplurals, plural);
|
|
return true;
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::parsePlural()
|
|
{
|
|
wxPluralFormsNode* p = expression();
|
|
if (p == NULL)
|
|
{
|
|
return NULL;
|
|
}
|
|
wxPluralFormsNodePtr n(p);
|
|
if (token().type() != wxPluralFormsToken::T_SEMICOLON)
|
|
{
|
|
return NULL;
|
|
}
|
|
return n.release();
|
|
}
|
|
|
|
const wxPluralFormsToken& wxPluralFormsParser::token() const
|
|
{
|
|
return m_scanner.token();
|
|
}
|
|
|
|
bool wxPluralFormsParser::nextToken()
|
|
{
|
|
if (!m_scanner.nextToken())
|
|
return false;
|
|
return true;
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::expression()
|
|
{
|
|
wxPluralFormsNode* p = logicalOrExpression();
|
|
if (p == NULL)
|
|
return NULL;
|
|
wxPluralFormsNodePtr n(p);
|
|
if (token().type() == wxPluralFormsToken::T_QUESTION)
|
|
{
|
|
wxPluralFormsNodePtr qn(new wxPluralFormsNode(token()));
|
|
if (!nextToken())
|
|
{
|
|
return 0;
|
|
}
|
|
p = expression();
|
|
if (p == 0)
|
|
{
|
|
return 0;
|
|
}
|
|
qn->setNode(1, p);
|
|
if (token().type() != wxPluralFormsToken::T_COLON)
|
|
{
|
|
return 0;
|
|
}
|
|
if (!nextToken())
|
|
{
|
|
return 0;
|
|
}
|
|
p = expression();
|
|
if (p == 0)
|
|
{
|
|
return 0;
|
|
}
|
|
qn->setNode(2, p);
|
|
qn->setNode(0, n.release());
|
|
return qn.release();
|
|
}
|
|
return n.release();
|
|
}
|
|
|
|
wxPluralFormsNode*wxPluralFormsParser::logicalOrExpression()
|
|
{
|
|
wxPluralFormsNode* p = logicalAndExpression();
|
|
if (p == NULL)
|
|
return NULL;
|
|
wxPluralFormsNodePtr ln(p);
|
|
if (token().type() == wxPluralFormsToken::T_LOGICAL_OR)
|
|
{
|
|
wxPluralFormsNodePtr un(new wxPluralFormsNode(token()));
|
|
if (!nextToken())
|
|
{
|
|
return 0;
|
|
}
|
|
p = logicalOrExpression();
|
|
if (p == 0)
|
|
{
|
|
return 0;
|
|
}
|
|
wxPluralFormsNodePtr rn(p); // right
|
|
if (rn->token().type() == wxPluralFormsToken::T_LOGICAL_OR)
|
|
{
|
|
// see logicalAndExpression comment
|
|
un->setNode(0, ln.release());
|
|
un->setNode(1, rn->releaseNode(0));
|
|
rn->setNode(0, un.release());
|
|
return rn.release();
|
|
}
|
|
|
|
|
|
un->setNode(0, ln.release());
|
|
un->setNode(1, rn.release());
|
|
return un.release();
|
|
}
|
|
return ln.release();
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::logicalAndExpression()
|
|
{
|
|
wxPluralFormsNode* p = equalityExpression();
|
|
if (p == NULL)
|
|
return NULL;
|
|
wxPluralFormsNodePtr ln(p); // left
|
|
if (token().type() == wxPluralFormsToken::T_LOGICAL_AND)
|
|
{
|
|
wxPluralFormsNodePtr un(new wxPluralFormsNode(token())); // up
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
p = logicalAndExpression();
|
|
if (p == 0)
|
|
{
|
|
return NULL;
|
|
}
|
|
wxPluralFormsNodePtr rn(p); // right
|
|
if (rn->token().type() == wxPluralFormsToken::T_LOGICAL_AND)
|
|
{
|
|
// transform 1 && (2 && 3) -> (1 && 2) && 3
|
|
// u r
|
|
// l r -> u 3
|
|
// 2 3 l 2
|
|
un->setNode(0, ln.release());
|
|
un->setNode(1, rn->releaseNode(0));
|
|
rn->setNode(0, un.release());
|
|
return rn.release();
|
|
}
|
|
|
|
un->setNode(0, ln.release());
|
|
un->setNode(1, rn.release());
|
|
return un.release();
|
|
}
|
|
return ln.release();
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::equalityExpression()
|
|
{
|
|
wxPluralFormsNode* p = relationalExpression();
|
|
if (p == NULL)
|
|
return NULL;
|
|
wxPluralFormsNodePtr n(p);
|
|
if (token().type() == wxPluralFormsToken::T_EQUAL
|
|
|| token().type() == wxPluralFormsToken::T_NOT_EQUAL)
|
|
{
|
|
wxPluralFormsNodePtr qn(new wxPluralFormsNode(token()));
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
p = relationalExpression();
|
|
if (p == NULL)
|
|
{
|
|
return NULL;
|
|
}
|
|
qn->setNode(1, p);
|
|
qn->setNode(0, n.release());
|
|
return qn.release();
|
|
}
|
|
return n.release();
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::relationalExpression()
|
|
{
|
|
wxPluralFormsNode* p = multiplicativeExpression();
|
|
if (p == NULL)
|
|
return NULL;
|
|
wxPluralFormsNodePtr n(p);
|
|
if (token().type() == wxPluralFormsToken::T_GREATER
|
|
|| token().type() == wxPluralFormsToken::T_LESS
|
|
|| token().type() == wxPluralFormsToken::T_GREATER_OR_EQUAL
|
|
|| token().type() == wxPluralFormsToken::T_LESS_OR_EQUAL)
|
|
{
|
|
wxPluralFormsNodePtr qn(new wxPluralFormsNode(token()));
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
p = multiplicativeExpression();
|
|
if (p == NULL)
|
|
{
|
|
return NULL;
|
|
}
|
|
qn->setNode(1, p);
|
|
qn->setNode(0, n.release());
|
|
return qn.release();
|
|
}
|
|
return n.release();
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::multiplicativeExpression()
|
|
{
|
|
wxPluralFormsNode* p = pmExpression();
|
|
if (p == NULL)
|
|
return NULL;
|
|
wxPluralFormsNodePtr n(p);
|
|
if (token().type() == wxPluralFormsToken::T_REMINDER)
|
|
{
|
|
wxPluralFormsNodePtr qn(new wxPluralFormsNode(token()));
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
p = pmExpression();
|
|
if (p == NULL)
|
|
{
|
|
return NULL;
|
|
}
|
|
qn->setNode(1, p);
|
|
qn->setNode(0, n.release());
|
|
return qn.release();
|
|
}
|
|
return n.release();
|
|
}
|
|
|
|
wxPluralFormsNode* wxPluralFormsParser::pmExpression()
|
|
{
|
|
wxPluralFormsNodePtr n;
|
|
if (token().type() == wxPluralFormsToken::T_N
|
|
|| token().type() == wxPluralFormsToken::T_NUMBER)
|
|
{
|
|
n.reset(new wxPluralFormsNode(token()));
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
}
|
|
else if (token().type() == wxPluralFormsToken::T_LEFT_BRACKET) {
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
wxPluralFormsNode* p = expression();
|
|
if (p == NULL)
|
|
{
|
|
return NULL;
|
|
}
|
|
n.reset(p);
|
|
if (token().type() != wxPluralFormsToken::T_RIGHT_BRACKET)
|
|
{
|
|
return NULL;
|
|
}
|
|
if (!nextToken())
|
|
{
|
|
return NULL;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
return NULL;
|
|
}
|
|
return n.release();
|
|
}
|
|
|
|
wxPluralFormsCalculator* wxPluralFormsCalculator::make(const char* s)
|
|
{
|
|
wxPluralFormsCalculatorPtr calculator(new wxPluralFormsCalculator);
|
|
if (s != NULL)
|
|
{
|
|
wxPluralFormsScanner scanner(s);
|
|
wxPluralFormsParser p(scanner);
|
|
if (!p.parse(*calculator))
|
|
{
|
|
return NULL;
|
|
}
|
|
}
|
|
return calculator.release();
|
|
}
|
|
|
|
|
|
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxMsgCatalogFile corresponds to one disk-file message catalog.
|
|
//
|
|
// This is a "low-level" class and is used only by wxMsgCatalog
|
|
// NOTE: for the documentation of the binary catalog (.MO) files refer to
|
|
// the GNU gettext manual:
|
|
// http://www.gnu.org/software/autoconf/manual/gettext/MO-Files.html
|
|
// ----------------------------------------------------------------------------
|
|
|
|
WX_DECLARE_EXPORTED_STRING_HASH_MAP(wxString, wxMessagesHash);
|
|
|
|
class wxMsgCatalogFile
|
|
{
|
|
public:
|
|
// ctor & dtor
|
|
wxMsgCatalogFile();
|
|
~wxMsgCatalogFile();
|
|
|
|
// load the catalog from disk (szDirPrefix corresponds to language)
|
|
bool Load(const wxString& szDirPrefix, const wxString& szName,
|
|
wxPluralFormsCalculatorPtr& rPluralFormsCalculator);
|
|
|
|
// fills the hash with string-translation pairs
|
|
bool FillHash(wxMessagesHash& hash,
|
|
const wxString& msgIdCharset,
|
|
bool convertEncoding) const;
|
|
|
|
// return the charset of the strings in this catalog or empty string if
|
|
// none/unknown
|
|
wxString GetCharset() const { return m_charset; }
|
|
|
|
private:
|
|
// this implementation is binary compatible with GNU gettext() version 0.10
|
|
|
|
// an entry in the string table
|
|
struct wxMsgTableEntry
|
|
{
|
|
size_t32 nLen; // length of the string
|
|
size_t32 ofsString; // pointer to the string
|
|
};
|
|
|
|
// header of a .mo file
|
|
struct wxMsgCatalogHeader
|
|
{
|
|
size_t32 magic, // offset +00: magic id
|
|
revision, // +04: revision
|
|
numStrings; // +08: number of strings in the file
|
|
size_t32 ofsOrigTable, // +0C: start of original string table
|
|
ofsTransTable; // +10: start of translated string table
|
|
size_t32 nHashSize, // +14: hash table size
|
|
ofsHashTable; // +18: offset of hash table start
|
|
};
|
|
|
|
// all data is stored here
|
|
wxMemoryBuffer m_data;
|
|
|
|
// data description
|
|
size_t32 m_numStrings; // number of strings in this domain
|
|
wxMsgTableEntry *m_pOrigTable, // pointer to original strings
|
|
*m_pTransTable; // translated
|
|
|
|
wxString m_charset; // from the message catalog header
|
|
|
|
|
|
// swap the 2 halves of 32 bit integer if needed
|
|
size_t32 Swap(size_t32 ui) const
|
|
{
|
|
return m_bSwapped ? (ui << 24) | ((ui & 0xff00) << 8) |
|
|
((ui >> 8) & 0xff00) | (ui >> 24)
|
|
: ui;
|
|
}
|
|
|
|
// just return the pointer to the start of the data as "char *" to
|
|
// facilitate doing pointer arithmetic with it
|
|
char *StringData() const
|
|
{
|
|
return static_cast<char *>(m_data.GetData());
|
|
}
|
|
|
|
const char *StringAtOfs(wxMsgTableEntry *pTable, size_t32 n) const
|
|
{
|
|
const wxMsgTableEntry * const ent = pTable + n;
|
|
|
|
// this check could fail for a corrupt message catalog
|
|
size_t32 ofsString = Swap(ent->ofsString);
|
|
if ( ofsString + Swap(ent->nLen) > m_data.GetDataLen())
|
|
{
|
|
return NULL;
|
|
}
|
|
|
|
return StringData() + ofsString;
|
|
}
|
|
|
|
bool m_bSwapped; // wrong endianness?
|
|
|
|
wxDECLARE_NO_COPY_CLASS(wxMsgCatalogFile);
|
|
};
|
|
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxMsgCatalog corresponds to one loaded message catalog.
|
|
//
|
|
// This is a "low-level" class and is used only by wxLocale (that's why
|
|
// it's designed to be stored in a linked list)
|
|
// ----------------------------------------------------------------------------
|
|
|
|
class wxMsgCatalog
|
|
{
|
|
public:
|
|
#if !wxUSE_UNICODE
|
|
wxMsgCatalog() { m_conv = NULL; }
|
|
~wxMsgCatalog();
|
|
#endif
|
|
|
|
// load the catalog from disk (szDirPrefix corresponds to language)
|
|
bool Load(const wxString& dirPrefix, const wxString& name,
|
|
const wxString& msgIdCharset, bool bConvertEncoding = false);
|
|
|
|
// get name of the catalog
|
|
wxString GetName() const { return m_name; }
|
|
|
|
// get the translated string: returns NULL if not found
|
|
const wxString *GetString(const wxString& sz, size_t n = size_t(-1)) const;
|
|
|
|
// public variable pointing to the next element in a linked list (or NULL)
|
|
wxMsgCatalog *m_pNext;
|
|
|
|
private:
|
|
wxMessagesHash m_messages; // all messages in the catalog
|
|
wxString m_name; // name of the domain
|
|
|
|
#if !wxUSE_UNICODE
|
|
// the conversion corresponding to this catalog charset if we installed it
|
|
// as the global one
|
|
wxCSConv *m_conv;
|
|
#endif
|
|
|
|
wxPluralFormsCalculatorPtr m_pluralFormsCalculator;
|
|
};
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// global variables
|
|
// ----------------------------------------------------------------------------
|
|
|
|
// the list of the directories to search for message catalog files
|
|
static wxArrayString gs_searchPrefixes;
|
|
|
|
// ============================================================================
|
|
// implementation
|
|
// ============================================================================
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxLanguageInfo
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#ifdef __WXMSW__
|
|
|
|
// helper used by wxLanguageInfo::GetLocaleName() and elsewhere to determine
|
|
// whether the locale is Unicode-only (it is if this function returns empty
|
|
// string)
|
|
static wxString wxGetANSICodePageForLocale(LCID lcid)
|
|
{
|
|
wxString cp;
|
|
|
|
wxChar buffer[16];
|
|
if ( ::GetLocaleInfo(lcid, LOCALE_IDEFAULTANSICODEPAGE,
|
|
buffer, WXSIZEOF(buffer)) > 0 )
|
|
{
|
|
if ( buffer[0] != _T('0') || buffer[1] != _T('\0') )
|
|
cp = buffer;
|
|
//else: this locale doesn't use ANSI code page
|
|
}
|
|
|
|
return cp;
|
|
}
|
|
|
|
wxUint32 wxLanguageInfo::GetLCID() const
|
|
{
|
|
return MAKELCID(MAKELANGID(WinLang, WinSublang), SORT_DEFAULT);
|
|
}
|
|
|
|
wxString wxLanguageInfo::GetLocaleName() const
|
|
{
|
|
wxString locale;
|
|
|
|
const LCID lcid = GetLCID();
|
|
|
|
wxChar buffer[256];
|
|
buffer[0] = _T('\0');
|
|
if ( !::GetLocaleInfo(lcid, LOCALE_SENGLANGUAGE, buffer, WXSIZEOF(buffer)) )
|
|
{
|
|
wxLogLastError(_T("GetLocaleInfo(LOCALE_SENGLANGUAGE)"));
|
|
return locale;
|
|
}
|
|
|
|
locale << buffer;
|
|
if ( ::GetLocaleInfo(lcid, LOCALE_SENGCOUNTRY,
|
|
buffer, WXSIZEOF(buffer)) > 0 )
|
|
{
|
|
locale << _T('_') << buffer;
|
|
}
|
|
|
|
const wxString cp = wxGetANSICodePageForLocale(lcid);
|
|
if ( !cp.empty() )
|
|
{
|
|
locale << _T('.') << cp;
|
|
}
|
|
|
|
return locale;
|
|
}
|
|
|
|
#endif // __WXMSW__
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxMsgCatalogFile class
|
|
// ----------------------------------------------------------------------------
|
|
|
|
wxMsgCatalogFile::wxMsgCatalogFile()
|
|
{
|
|
}
|
|
|
|
wxMsgCatalogFile::~wxMsgCatalogFile()
|
|
{
|
|
}
|
|
|
|
// return the directories to search for message catalogs under the given
|
|
// prefix, separated by wxPATH_SEP
|
|
static
|
|
wxString GetMsgCatalogSubdirs(const wxString& prefix, const wxString& lang)
|
|
{
|
|
// Search first in Unix-standard prefix/lang/LC_MESSAGES, then in
|
|
// prefix/lang and finally in just prefix.
|
|
//
|
|
// Note that we use LC_MESSAGES on all platforms and not just Unix, because
|
|
// it doesn't cost much to look into one more directory and doing it this
|
|
// way has two important benefits:
|
|
// a) we don't break compatibility with wx-2.6 and older by stopping to
|
|
// look in a directory where the catalogs used to be and thus silently
|
|
// breaking apps after they are recompiled against the latest wx
|
|
// b) it makes it possible to package app's support files in the same
|
|
// way on all target platforms
|
|
const wxString pathPrefix = wxFileName(prefix, lang).GetFullPath();
|
|
|
|
wxString searchPath;
|
|
searchPath.reserve(4*pathPrefix.length());
|
|
searchPath << pathPrefix << wxFILE_SEP_PATH << "LC_MESSAGES" << wxPATH_SEP
|
|
<< prefix << wxFILE_SEP_PATH << wxPATH_SEP
|
|
<< pathPrefix;
|
|
|
|
return searchPath;
|
|
}
|
|
|
|
// construct the search path for the given language
|
|
static wxString GetFullSearchPath(const wxString& lang)
|
|
{
|
|
// first take the entries explicitly added by the program
|
|
wxArrayString paths;
|
|
paths.reserve(gs_searchPrefixes.size() + 1);
|
|
size_t n,
|
|
count = gs_searchPrefixes.size();
|
|
for ( n = 0; n < count; n++ )
|
|
{
|
|
paths.Add(GetMsgCatalogSubdirs(gs_searchPrefixes[n], lang));
|
|
}
|
|
|
|
|
|
#if wxUSE_STDPATHS
|
|
// then look in the standard location
|
|
const wxString stdp = wxStandardPaths::Get().
|
|
GetLocalizedResourcesDir(lang, wxStandardPaths::ResourceCat_Messages);
|
|
|
|
if ( paths.Index(stdp) == wxNOT_FOUND )
|
|
paths.Add(stdp);
|
|
#endif // wxUSE_STDPATHS
|
|
|
|
// last look in default locations
|
|
#ifdef __UNIX__
|
|
// LC_PATH is a standard env var containing the search path for the .mo
|
|
// files
|
|
const char *pszLcPath = wxGetenv("LC_PATH");
|
|
if ( pszLcPath )
|
|
{
|
|
const wxString lcp = GetMsgCatalogSubdirs(pszLcPath, lang);
|
|
if ( paths.Index(lcp) == wxNOT_FOUND )
|
|
paths.Add(lcp);
|
|
}
|
|
|
|
// also add the one from where wxWin was installed:
|
|
wxString wxp = wxGetInstallPrefix();
|
|
if ( !wxp.empty() )
|
|
{
|
|
wxp = GetMsgCatalogSubdirs(wxp + wxS("/share/locale"), lang);
|
|
if ( paths.Index(wxp) == wxNOT_FOUND )
|
|
paths.Add(wxp);
|
|
}
|
|
#endif // __UNIX__
|
|
|
|
|
|
// finally construct the full search path
|
|
wxString searchPath;
|
|
searchPath.reserve(500);
|
|
count = paths.size();
|
|
for ( n = 0; n < count; n++ )
|
|
{
|
|
searchPath += paths[n];
|
|
if ( n != count - 1 )
|
|
searchPath += wxPATH_SEP;
|
|
}
|
|
|
|
return searchPath;
|
|
}
|
|
|
|
// open disk file and read in it's contents
|
|
bool wxMsgCatalogFile::Load(const wxString& szDirPrefix, const wxString& szName,
|
|
wxPluralFormsCalculatorPtr& rPluralFormsCalculator)
|
|
{
|
|
wxCHECK_MSG( szDirPrefix.length() >= LEN_LANG, false,
|
|
"invalid language specification" );
|
|
|
|
wxString searchPath;
|
|
|
|
#if wxUSE_FONTMAP
|
|
// first look for the catalog for this language and the current locale:
|
|
// notice that we don't use the system name for the locale as this would
|
|
// force us to install catalogs in different locations depending on the
|
|
// system but always use the canonical name
|
|
wxFontEncoding encSys = wxLocale::GetSystemEncoding();
|
|
if ( encSys != wxFONTENCODING_SYSTEM )
|
|
{
|
|
wxString fullname(szDirPrefix);
|
|
fullname << wxS('.') << wxFontMapperBase::GetEncodingName(encSys);
|
|
searchPath << GetFullSearchPath(fullname) << wxPATH_SEP;
|
|
}
|
|
#endif // wxUSE_FONTMAP
|
|
|
|
|
|
searchPath += GetFullSearchPath(szDirPrefix);
|
|
if ( szDirPrefix.length() > LEN_LANG && szDirPrefix[LEN_LANG] == wxS('_') )
|
|
{
|
|
// also add just base locale name: for things like "fr_BE" (Belgium
|
|
// French) we should use fall back on plain "fr" if no Belgium-specific
|
|
// message catalogs exist
|
|
searchPath << wxPATH_SEP
|
|
<< GetFullSearchPath(ExtractLang(szDirPrefix));
|
|
}
|
|
|
|
// don't give translation errors here because the wxstd catalog might
|
|
// not yet be loaded (and it's normal)
|
|
//
|
|
// (we're using an object because we have several return paths)
|
|
|
|
NoTransErr noTransErr;
|
|
wxLogVerbose(_("looking for catalog '%s' in path '%s'."),
|
|
szName, searchPath.c_str());
|
|
wxLogTrace(TRACE_I18N, wxS("Looking for \"%s.mo\" in \"%s\""),
|
|
szName, searchPath.c_str());
|
|
|
|
wxFileName fn(szName);
|
|
fn.SetExt(wxS("mo"));
|
|
|
|
wxString strFullName;
|
|
#if wxUSE_FILESYSTEM
|
|
wxFileSystem fileSys;
|
|
if ( !fileSys.FindFileInPath(&strFullName, searchPath, fn.GetFullPath()) )
|
|
#else // !wxUSE_FILESYSTEM
|
|
if ( !wxFindFileInPath(&strFullName, searchPath, fn.GetFullPath()) )
|
|
#endif // wxUSE_FILESYSTEM/!wxUSE_FILESYSTEM
|
|
{
|
|
wxLogVerbose(_("catalog file for domain '%s' not found."), szName);
|
|
wxLogTrace(TRACE_I18N, wxS("Catalog \"%s.mo\" not found"), szName);
|
|
return false;
|
|
}
|
|
|
|
// open file and read its data
|
|
wxLogVerbose(_("using catalog '%s' from '%s'."), szName, strFullName.c_str());
|
|
wxLogTrace(TRACE_I18N, wxS("Using catalog \"%s\"."), strFullName.c_str());
|
|
|
|
#if wxUSE_FILESYSTEM
|
|
wxFSFile * const fileMsg = fileSys.OpenFile(strFullName);
|
|
if ( !fileMsg )
|
|
return false;
|
|
|
|
wxInputStream *fileStream = fileMsg->GetStream();
|
|
m_data.SetDataLen(0);
|
|
|
|
static const size_t chunkSize = 4096;
|
|
while ( !fileStream->Eof() ) {
|
|
fileStream->Read(m_data.GetAppendBuf(chunkSize), chunkSize);
|
|
m_data.UngetAppendBuf(fileStream->LastRead());
|
|
}
|
|
|
|
delete fileMsg;
|
|
#else // !wxUSE_FILESYSTEM
|
|
wxFile fileMsg(strFullName);
|
|
if ( !fileMsg.IsOpened() )
|
|
return false;
|
|
|
|
// get the file size (assume it is less than 4Gb...)
|
|
wxFileOffset lenFile = fileMsg.Length();
|
|
if ( lenFile == wxInvalidOffset )
|
|
return false;
|
|
|
|
size_t nSize = wx_truncate_cast(size_t, lenFile);
|
|
wxASSERT_MSG( nSize == lenFile + size_t(0), wxS("message catalog bigger than 4GB?") );
|
|
|
|
// read the whole file in memory
|
|
if ( fileMsg.Read(m_data.GetWriteBuf(nSize), nSize) != lenFile )
|
|
return false;
|
|
|
|
m_data.UngetWriteBuf(nSize);
|
|
#endif // wxUSE_FILESYSTEM/!wxUSE_FILESYSTEM
|
|
|
|
|
|
// examine header
|
|
bool bValid = m_data.GetDataLen() > sizeof(wxMsgCatalogHeader);
|
|
|
|
const wxMsgCatalogHeader *pHeader = (wxMsgCatalogHeader *)m_data.GetData();
|
|
if ( bValid ) {
|
|
// we'll have to swap all the integers if it's true
|
|
m_bSwapped = pHeader->magic == MSGCATALOG_MAGIC_SW;
|
|
|
|
// check the magic number
|
|
bValid = m_bSwapped || pHeader->magic == MSGCATALOG_MAGIC;
|
|
}
|
|
|
|
if ( !bValid ) {
|
|
// it's either too short or has incorrect magic number
|
|
wxLogWarning(_("'%s' is not a valid message catalog."), strFullName.c_str());
|
|
|
|
return false;
|
|
}
|
|
|
|
// initialize
|
|
m_numStrings = Swap(pHeader->numStrings);
|
|
m_pOrigTable = (wxMsgTableEntry *)(StringData() +
|
|
Swap(pHeader->ofsOrigTable));
|
|
m_pTransTable = (wxMsgTableEntry *)(StringData() +
|
|
Swap(pHeader->ofsTransTable));
|
|
|
|
// now parse catalog's header and try to extract catalog charset and
|
|
// plural forms formula from it:
|
|
|
|
const char* headerData = StringAtOfs(m_pOrigTable, 0);
|
|
if ( headerData && headerData[0] == '\0' )
|
|
{
|
|
// Extract the charset:
|
|
const char * const header = StringAtOfs(m_pTransTable, 0);
|
|
const char *
|
|
cset = strstr(header, "Content-Type: text/plain; charset=");
|
|
if ( cset )
|
|
{
|
|
cset += 34; // strlen("Content-Type: text/plain; charset=")
|
|
|
|
const char * const csetEnd = strchr(cset, '\n');
|
|
if ( csetEnd )
|
|
{
|
|
m_charset = wxString(cset, csetEnd - cset);
|
|
if ( m_charset == wxS("CHARSET") )
|
|
{
|
|
// "CHARSET" is not valid charset, but lazy translator
|
|
m_charset.empty();
|
|
}
|
|
}
|
|
}
|
|
// else: incorrectly filled Content-Type header
|
|
|
|
// Extract plural forms:
|
|
const char * plurals = strstr(header, "Plural-Forms:");
|
|
if ( plurals )
|
|
{
|
|
plurals += 13; // strlen("Plural-Forms:")
|
|
const char * const pluralsEnd = strchr(plurals, '\n');
|
|
if ( pluralsEnd )
|
|
{
|
|
const size_t pluralsLen = pluralsEnd - plurals;
|
|
wxCharBuffer buf(pluralsLen);
|
|
strncpy(buf.data(), plurals, pluralsLen);
|
|
wxPluralFormsCalculator * const
|
|
pCalculator = wxPluralFormsCalculator::make(buf);
|
|
if ( pCalculator )
|
|
{
|
|
rPluralFormsCalculator.reset(pCalculator);
|
|
}
|
|
else
|
|
{
|
|
wxLogVerbose(_("Failed to parse Plural-Forms: '%s'"),
|
|
buf.data());
|
|
}
|
|
}
|
|
}
|
|
|
|
if ( !rPluralFormsCalculator.get() )
|
|
rPluralFormsCalculator.reset(wxPluralFormsCalculator::make());
|
|
}
|
|
|
|
// everything is fine
|
|
return true;
|
|
}
|
|
|
|
bool wxMsgCatalogFile::FillHash(wxMessagesHash& hash,
|
|
const wxString& msgIdCharset,
|
|
bool convertEncoding) const
|
|
{
|
|
#if wxUSE_UNICODE
|
|
// this parameter doesn't make sense, we always must convert encoding in
|
|
// Unicode build
|
|
convertEncoding = true;
|
|
#elif wxUSE_FONTMAP
|
|
if ( convertEncoding )
|
|
{
|
|
// determine if we need any conversion at all
|
|
wxFontEncoding encCat = wxFontMapperBase::GetEncodingFromName(m_charset);
|
|
if ( encCat == wxLocale::GetSystemEncoding() )
|
|
{
|
|
// no need to convert
|
|
convertEncoding = false;
|
|
}
|
|
}
|
|
#endif // wxUSE_UNICODE/wxUSE_FONTMAP
|
|
|
|
#if wxUSE_WCHAR_T
|
|
// conversion to use to convert catalog strings to the GUI encoding
|
|
wxMBConv *inputConv,
|
|
*inputConvPtr = NULL; // same as inputConv but safely deleteable
|
|
if ( convertEncoding && !m_charset.empty() )
|
|
{
|
|
inputConvPtr =
|
|
inputConv = new wxCSConv(m_charset);
|
|
}
|
|
else // no need or not possible to convert the encoding
|
|
{
|
|
#if wxUSE_UNICODE
|
|
// we must somehow convert the narrow strings in the message catalog to
|
|
// wide strings, so use the default conversion if we have no charset
|
|
inputConv = wxConvCurrent;
|
|
#else // !wxUSE_UNICODE
|
|
inputConv = NULL;
|
|
#endif // wxUSE_UNICODE/!wxUSE_UNICODE
|
|
}
|
|
|
|
// conversion to apply to msgid strings before looking them up: we only
|
|
// need it if the msgids are neither in 7 bit ASCII nor in the same
|
|
// encoding as the catalog
|
|
wxCSConv *sourceConv = msgIdCharset.empty() || (msgIdCharset == m_charset)
|
|
? NULL
|
|
: new wxCSConv(msgIdCharset);
|
|
|
|
#elif wxUSE_FONTMAP
|
|
wxASSERT_MSG( msgIdCharset.empty(),
|
|
wxS("non-ASCII msgid languages only supported if wxUSE_WCHAR_T=1") );
|
|
|
|
wxEncodingConverter converter;
|
|
if ( convertEncoding )
|
|
{
|
|
wxFontEncoding targetEnc = wxFONTENCODING_SYSTEM;
|
|
wxFontEncoding enc = wxFontMapperBase::Get()->CharsetToEncoding(m_charset, false);
|
|
if ( enc == wxFONTENCODING_SYSTEM )
|
|
{
|
|
convertEncoding = false; // unknown encoding
|
|
}
|
|
else
|
|
{
|
|
targetEnc = wxLocale::GetSystemEncoding();
|
|
if (targetEnc == wxFONTENCODING_SYSTEM)
|
|
{
|
|
wxFontEncodingArray a = wxEncodingConverter::GetPlatformEquivalents(enc);
|
|
if (a[0] == enc)
|
|
// no conversion needed, locale uses native encoding
|
|
convertEncoding = false;
|
|
if (a.GetCount() == 0)
|
|
// we don't know common equiv. under this platform
|
|
convertEncoding = false;
|
|
targetEnc = a[0];
|
|
}
|
|
}
|
|
|
|
if ( convertEncoding )
|
|
{
|
|
converter.Init(enc, targetEnc);
|
|
}
|
|
}
|
|
#endif // wxUSE_WCHAR_T/!wxUSE_WCHAR_T
|
|
(void)convertEncoding; // get rid of warnings about unused parameter
|
|
|
|
for (size_t32 i = 0; i < m_numStrings; i++)
|
|
{
|
|
const char *data = StringAtOfs(m_pOrigTable, i);
|
|
if (!data)
|
|
return false; // may happen for invalid MO files
|
|
|
|
wxString msgid;
|
|
#if wxUSE_UNICODE
|
|
msgid = wxString(data, *inputConv);
|
|
#else // ASCII
|
|
#if wxUSE_WCHAR_T
|
|
if ( inputConv && sourceConv )
|
|
msgid = wxString(inputConv->cMB2WC(data), *sourceConv);
|
|
else
|
|
#endif
|
|
msgid = data;
|
|
#endif // wxUSE_UNICODE
|
|
|
|
data = StringAtOfs(m_pTransTable, i);
|
|
if (!data)
|
|
return false; // may happen for invalid MO files
|
|
|
|
size_t length = Swap(m_pTransTable[i].nLen);
|
|
size_t offset = 0;
|
|
size_t index = 0;
|
|
while (offset < length)
|
|
{
|
|
const char * const str = data + offset;
|
|
|
|
wxString msgstr;
|
|
#if wxUSE_UNICODE
|
|
msgstr = wxString(str, *inputConv);
|
|
#elif wxUSE_WCHAR_T
|
|
if ( inputConv )
|
|
msgstr = wxString(inputConv->cMB2WC(str), *wxConvUI);
|
|
else
|
|
msgstr = str;
|
|
#else // !wxUSE_WCHAR_T
|
|
#if wxUSE_FONTMAP
|
|
if ( bConvertEncoding )
|
|
msgstr = wxString(converter.Convert(str));
|
|
else
|
|
#endif
|
|
msgstr = str;
|
|
#endif // wxUSE_WCHAR_T/!wxUSE_WCHAR_T
|
|
|
|
if ( !msgstr.empty() )
|
|
{
|
|
hash[index == 0 ? msgid : msgid + wxChar(index)] = msgstr;
|
|
}
|
|
|
|
// skip this string
|
|
// IMPORTANT: accesses to the 'data' pointer are valid only for
|
|
// the first 'length+1' bytes (GNU specs says that the
|
|
// final NUL is not counted in length); using wxStrnlen()
|
|
// we make sure we don't access memory beyond the valid range
|
|
// (which otherwise may happen for invalid MO files):
|
|
offset += wxStrnlen(str, length - offset) + 1;
|
|
++index;
|
|
}
|
|
}
|
|
|
|
#if wxUSE_WCHAR_T
|
|
delete sourceConv;
|
|
delete inputConvPtr;
|
|
#endif // wxUSE_WCHAR_T
|
|
|
|
return true;
|
|
}
|
|
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxMsgCatalog class
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#if !wxUSE_UNICODE
|
|
wxMsgCatalog::~wxMsgCatalog()
|
|
{
|
|
if ( m_conv )
|
|
{
|
|
if ( wxConvUI == m_conv )
|
|
{
|
|
// we only change wxConvUI if it points to wxConvLocal so we reset
|
|
// it back to it too
|
|
wxConvUI = &wxConvLocal;
|
|
}
|
|
|
|
delete m_conv;
|
|
}
|
|
}
|
|
#endif // !wxUSE_UNICODE
|
|
|
|
bool wxMsgCatalog::Load(const wxString& dirPrefix, const wxString& name,
|
|
const wxString& msgIdCharset, bool bConvertEncoding)
|
|
{
|
|
wxMsgCatalogFile file;
|
|
|
|
m_name = name;
|
|
|
|
if ( !file.Load(dirPrefix, name, m_pluralFormsCalculator) )
|
|
return false;
|
|
|
|
if ( !file.FillHash(m_messages, msgIdCharset, bConvertEncoding) )
|
|
return false;
|
|
|
|
#if !wxUSE_UNICODE
|
|
// we should use a conversion compatible with the message catalog encoding
|
|
// in the GUI if we don't convert the strings to the current conversion but
|
|
// as the encoding is global, only change it once, otherwise we could get
|
|
// into trouble if we use several message catalogs with different encodings
|
|
//
|
|
// this is, of course, a hack but it at least allows the program to use
|
|
// message catalogs in any encodings without asking the user to change his
|
|
// locale
|
|
if ( !bConvertEncoding &&
|
|
!file.GetCharset().empty() &&
|
|
wxConvUI == &wxConvLocal )
|
|
{
|
|
wxConvUI =
|
|
m_conv = new wxCSConv(file.GetCharset());
|
|
}
|
|
#endif // !wxUSE_UNICODE
|
|
|
|
return true;
|
|
}
|
|
|
|
const wxString *wxMsgCatalog::GetString(const wxString& str, size_t n) const
|
|
{
|
|
int index = 0;
|
|
if (n != size_t(-1))
|
|
{
|
|
index = m_pluralFormsCalculator->evaluate(n);
|
|
}
|
|
wxMessagesHash::const_iterator i;
|
|
if (index != 0)
|
|
{
|
|
i = m_messages.find(wxString(str) + wxChar(index)); // plural
|
|
}
|
|
else
|
|
{
|
|
i = m_messages.find(str);
|
|
}
|
|
|
|
if ( i != m_messages.end() )
|
|
{
|
|
return &i->second;
|
|
}
|
|
else
|
|
return NULL;
|
|
}
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxLocale
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#include "wx/arrimpl.cpp"
|
|
WX_DECLARE_EXPORTED_OBJARRAY(wxLanguageInfo, wxLanguageInfoArray);
|
|
WX_DEFINE_OBJARRAY(wxLanguageInfoArray)
|
|
|
|
wxLanguageInfoArray *wxLocale::ms_languagesDB = NULL;
|
|
|
|
/*static*/ void wxLocale::CreateLanguagesDB()
|
|
{
|
|
if (ms_languagesDB == NULL)
|
|
{
|
|
ms_languagesDB = new wxLanguageInfoArray;
|
|
InitLanguagesDB();
|
|
}
|
|
}
|
|
|
|
/*static*/ void wxLocale::DestroyLanguagesDB()
|
|
{
|
|
delete ms_languagesDB;
|
|
ms_languagesDB = NULL;
|
|
}
|
|
|
|
|
|
void wxLocale::DoCommonInit()
|
|
{
|
|
m_pszOldLocale = NULL;
|
|
|
|
m_pOldLocale = wxSetLocale(this);
|
|
|
|
m_pMsgCat = NULL;
|
|
m_language = wxLANGUAGE_UNKNOWN;
|
|
m_initialized = false;
|
|
}
|
|
|
|
// NB: this function has (desired) side effect of changing current locale
|
|
bool wxLocale::Init(const wxString& name,
|
|
const wxString& shortName,
|
|
const wxString& locale,
|
|
bool bLoadDefault,
|
|
bool bConvertEncoding)
|
|
{
|
|
wxASSERT_MSG( !m_initialized,
|
|
wxS("you can't call wxLocale::Init more than once") );
|
|
|
|
m_initialized = true;
|
|
m_strLocale = name;
|
|
m_strShort = shortName;
|
|
m_bConvertEncoding = bConvertEncoding;
|
|
m_language = wxLANGUAGE_UNKNOWN;
|
|
|
|
// change current locale (default: same as long name)
|
|
wxString szLocale(locale);
|
|
if ( szLocale.empty() )
|
|
{
|
|
// the argument to setlocale()
|
|
szLocale = shortName;
|
|
|
|
wxCHECK_MSG( !szLocale.empty(), false,
|
|
wxS("no locale to set in wxLocale::Init()") );
|
|
}
|
|
|
|
const char *oldLocale = wxSetlocale(LC_ALL, szLocale);
|
|
if ( oldLocale )
|
|
m_pszOldLocale = wxStrdup(oldLocale);
|
|
else
|
|
m_pszOldLocale = NULL;
|
|
|
|
if ( m_pszOldLocale == NULL )
|
|
wxLogError(_("locale '%s' can not be set."), szLocale);
|
|
|
|
// the short name will be used to look for catalog files as well,
|
|
// so we need something here
|
|
if ( m_strShort.empty() ) {
|
|
// FIXME I don't know how these 2 letter abbreviations are formed,
|
|
// this wild guess is surely wrong
|
|
if ( !szLocale.empty() )
|
|
{
|
|
m_strShort += (wxChar)wxTolower(szLocale[0]);
|
|
if ( szLocale.length() > 1 )
|
|
m_strShort += (wxChar)wxTolower(szLocale[1]);
|
|
}
|
|
}
|
|
|
|
// load the default catalog with wxWidgets standard messages
|
|
m_pMsgCat = NULL;
|
|
bool bOk = true;
|
|
if ( bLoadDefault )
|
|
{
|
|
bOk = AddCatalog(wxS("wxstd"));
|
|
|
|
// there may be a catalog with toolkit specific overrides, it is not
|
|
// an error if this does not exist
|
|
if ( bOk )
|
|
{
|
|
wxString port(wxPlatformInfo::Get().GetPortIdName());
|
|
if ( !port.empty() )
|
|
{
|
|
AddCatalog(port.BeforeFirst(wxS('/')).MakeLower());
|
|
}
|
|
}
|
|
}
|
|
|
|
return bOk;
|
|
}
|
|
|
|
|
|
#if defined(__UNIX__) && wxUSE_UNICODE && !defined(__WXMAC__)
|
|
static const char *wxSetlocaleTryUTF8(int c, const wxString& lc)
|
|
{
|
|
const char *l = NULL;
|
|
|
|
// NB: We prefer to set UTF-8 locale if it's possible and only fall back to
|
|
// non-UTF-8 locale if it fails
|
|
|
|
if ( !lc.empty() )
|
|
{
|
|
wxString buf(lc);
|
|
wxString buf2;
|
|
buf2 = buf + wxS(".UTF-8");
|
|
l = wxSetlocale(c, buf2);
|
|
if ( !l )
|
|
{
|
|
buf2 = buf + wxS(".utf-8");
|
|
l = wxSetlocale(c, buf2);
|
|
}
|
|
if ( !l )
|
|
{
|
|
buf2 = buf + wxS(".UTF8");
|
|
l = wxSetlocale(c, buf2);
|
|
}
|
|
if ( !l )
|
|
{
|
|
buf2 = buf + wxS(".utf8");
|
|
l = wxSetlocale(c, buf2);
|
|
}
|
|
}
|
|
|
|
// if we can't set UTF-8 locale, try non-UTF-8 one:
|
|
if ( !l )
|
|
l = wxSetlocale(c, lc);
|
|
|
|
return l;
|
|
}
|
|
#else
|
|
#define wxSetlocaleTryUTF8(c, lc) wxSetlocale(c, lc)
|
|
#endif
|
|
|
|
bool wxLocale::Init(int language, int flags)
|
|
{
|
|
bool ret = true;
|
|
|
|
int lang = language;
|
|
if (lang == wxLANGUAGE_DEFAULT)
|
|
{
|
|
// auto detect the language
|
|
lang = GetSystemLanguage();
|
|
}
|
|
|
|
// We failed to detect system language, so we will use English:
|
|
if (lang == wxLANGUAGE_UNKNOWN)
|
|
{
|
|
return false;
|
|
}
|
|
|
|
const wxLanguageInfo *info = GetLanguageInfo(lang);
|
|
|
|
// Unknown language:
|
|
if (info == NULL)
|
|
{
|
|
wxLogError(wxS("Unknown language %i."), lang);
|
|
return false;
|
|
}
|
|
|
|
wxString name = info->Description;
|
|
wxString canonical = info->CanonicalName;
|
|
wxString locale;
|
|
|
|
// Set the locale:
|
|
#if defined(__OS2__)
|
|
const char *retloc = wxSetlocale(LC_ALL , wxEmptyString);
|
|
#elif defined(__UNIX__) && !defined(__WXMAC__)
|
|
if (language != wxLANGUAGE_DEFAULT)
|
|
locale = info->CanonicalName;
|
|
|
|
const char *retloc = wxSetlocaleTryUTF8(LC_ALL, locale);
|
|
|
|
const wxString langOnly = ExtractLang(locale);
|
|
if ( !retloc )
|
|
{
|
|
// Some C libraries don't like xx_YY form and require xx only
|
|
retloc = wxSetlocaleTryUTF8(LC_ALL, langOnly);
|
|
}
|
|
|
|
#if wxUSE_FONTMAP
|
|
// some systems (e.g. FreeBSD and HP-UX) don't have xx_YY aliases but
|
|
// require the full xx_YY.encoding form, so try using UTF-8 because this is
|
|
// the only thing we can do generically
|
|
//
|
|
// TODO: add encodings applicable to each language to the lang DB and try
|
|
// them all in turn here
|
|
if ( !retloc )
|
|
{
|
|
const wxChar **names =
|
|
wxFontMapperBase::GetAllEncodingNames(wxFONTENCODING_UTF8);
|
|
while ( *names )
|
|
{
|
|
retloc = wxSetlocale(LC_ALL, locale + wxS('.') + *names++);
|
|
if ( retloc )
|
|
break;
|
|
}
|
|
}
|
|
#endif // wxUSE_FONTMAP
|
|
|
|
if ( !retloc )
|
|
{
|
|
// Some C libraries (namely glibc) still use old ISO 639,
|
|
// so will translate the abbrev for them
|
|
wxString localeAlt;
|
|
if ( langOnly == wxS("he") )
|
|
localeAlt = wxS("iw") + ExtractNotLang(locale);
|
|
else if ( langOnly == wxS("id") )
|
|
localeAlt = wxS("in") + ExtractNotLang(locale);
|
|
else if ( langOnly == wxS("yi") )
|
|
localeAlt = wxS("ji") + ExtractNotLang(locale);
|
|
else if ( langOnly == wxS("nb") )
|
|
localeAlt = wxS("no_NO");
|
|
else if ( langOnly == wxS("nn") )
|
|
localeAlt = wxS("no_NY");
|
|
|
|
if ( !localeAlt.empty() )
|
|
{
|
|
retloc = wxSetlocaleTryUTF8(LC_ALL, localeAlt);
|
|
if ( !retloc )
|
|
retloc = wxSetlocaleTryUTF8(LC_ALL, ExtractLang(localeAlt));
|
|
}
|
|
}
|
|
|
|
if ( !retloc )
|
|
ret = false;
|
|
|
|
#ifdef __AIX__
|
|
// at least in AIX 5.2 libc is buggy and the string returned from
|
|
// setlocale(LC_ALL) can't be passed back to it because it returns 6
|
|
// strings (one for each locale category), i.e. for C locale we get back
|
|
// "C C C C C C"
|
|
//
|
|
// this contradicts IBM own docs but this is not of much help, so just work
|
|
// around it in the crudest possible manner
|
|
char* p = const_cast<char*>(wxStrchr(retloc, ' '));
|
|
if ( p )
|
|
*p = '\0';
|
|
#endif // __AIX__
|
|
|
|
#elif defined(__WIN32__)
|
|
const char *retloc = "C";
|
|
if ( language != wxLANGUAGE_DEFAULT )
|
|
{
|
|
if ( info->WinLang == 0 )
|
|
{
|
|
wxLogWarning(wxS("Locale '%s' not supported by OS."), name.c_str());
|
|
// retloc already set to "C"
|
|
}
|
|
else // language supported by Windows
|
|
{
|
|
// Windows CE doesn't have SetThreadLocale() and there doesn't seem
|
|
// to be any equivalent
|
|
#ifndef __WXWINCE__
|
|
const wxUint32 lcid = info->GetLCID();
|
|
|
|
// change locale used by Windows functions
|
|
::SetThreadLocale(lcid);
|
|
#endif
|
|
|
|
// and also call setlocale() to change locale used by the CRT
|
|
locale = info->GetLocaleName();
|
|
if ( locale.empty() )
|
|
{
|
|
ret = false;
|
|
}
|
|
else // have a valid locale
|
|
{
|
|
retloc = wxSetlocale(LC_ALL, locale);
|
|
}
|
|
}
|
|
}
|
|
else // language == wxLANGUAGE_DEFAULT
|
|
{
|
|
retloc = wxSetlocale(LC_ALL, wxEmptyString);
|
|
}
|
|
|
|
#if wxUSE_UNICODE && (defined(__VISUALC__) || defined(__MINGW32__))
|
|
// VC++ setlocale() (also used by Mingw) can't set locale to languages that
|
|
// can only be written using Unicode, therefore wxSetlocale() call fails
|
|
// for such languages but we don't want to report it as an error -- so that
|
|
// at least message catalogs can be used.
|
|
if ( !retloc )
|
|
{
|
|
if ( wxGetANSICodePageForLocale(LOCALE_USER_DEFAULT).empty() )
|
|
{
|
|
// we set the locale to a Unicode-only language, don't treat the
|
|
// inability of CRT to use it as an error
|
|
retloc = "C";
|
|
}
|
|
}
|
|
#endif // CRT not handling Unicode-only languages
|
|
|
|
if ( !retloc )
|
|
ret = false;
|
|
#elif defined(__WXMAC__)
|
|
if (lang == wxLANGUAGE_DEFAULT)
|
|
locale = wxEmptyString;
|
|
else
|
|
locale = info->CanonicalName;
|
|
|
|
const char *retloc = wxSetlocale(LC_ALL, locale);
|
|
|
|
if ( !retloc )
|
|
{
|
|
// Some C libraries don't like xx_YY form and require xx only
|
|
retloc = wxSetlocale(LC_ALL, ExtractLang(locale));
|
|
}
|
|
#else
|
|
wxUnusedVar(flags);
|
|
return false;
|
|
#define WX_NO_LOCALE_SUPPORT
|
|
#endif
|
|
|
|
#ifndef WX_NO_LOCALE_SUPPORT
|
|
if ( !ret )
|
|
{
|
|
wxLogWarning(_("Cannot set locale to language \"%s\"."), name.c_str());
|
|
|
|
// continue nevertheless and try to load at least the translations for
|
|
// this language
|
|
}
|
|
|
|
if ( !Init(name, canonical, retloc,
|
|
(flags & wxLOCALE_LOAD_DEFAULT) != 0,
|
|
(flags & wxLOCALE_CONV_ENCODING) != 0) )
|
|
{
|
|
ret = false;
|
|
}
|
|
|
|
if (IsOk()) // setlocale() succeeded
|
|
m_language = lang;
|
|
|
|
return ret;
|
|
#endif // !WX_NO_LOCALE_SUPPORT
|
|
}
|
|
|
|
|
|
|
|
void wxLocale::AddCatalogLookupPathPrefix(const wxString& prefix)
|
|
{
|
|
if ( gs_searchPrefixes.Index(prefix) == wxNOT_FOUND )
|
|
{
|
|
gs_searchPrefixes.Add(prefix);
|
|
}
|
|
//else: already have it
|
|
}
|
|
|
|
/*static*/ int wxLocale::GetSystemLanguage()
|
|
{
|
|
CreateLanguagesDB();
|
|
|
|
// init i to avoid compiler warning
|
|
size_t i = 0,
|
|
count = ms_languagesDB->GetCount();
|
|
|
|
#if defined(__UNIX__)
|
|
// first get the string identifying the language from the environment
|
|
wxString langFull;
|
|
#ifdef __WXMAC__
|
|
wxCFRef<CFLocaleRef> userLocaleRef(CFLocaleCopyCurrent());
|
|
|
|
// because the locale identifier (kCFLocaleIdentifier) is formatted a little bit differently, eg
|
|
// az_Cyrl_AZ@calendar=buddhist;currency=JPY we just recreate the base info as expected by wx here
|
|
|
|
wxCFStringRef str(wxCFRetain((CFStringRef)CFLocaleGetValue(userLocaleRef, kCFLocaleLanguageCode)));
|
|
langFull = str.AsString()+"_";
|
|
str.reset(wxCFRetain((CFStringRef)CFLocaleGetValue(userLocaleRef, kCFLocaleCountryCode)));
|
|
langFull += str.AsString();
|
|
#else
|
|
if (!wxGetEnv(wxS("LC_ALL"), &langFull) &&
|
|
!wxGetEnv(wxS("LC_MESSAGES"), &langFull) &&
|
|
!wxGetEnv(wxS("LANG"), &langFull))
|
|
{
|
|
// no language specified, treat it as English
|
|
return wxLANGUAGE_ENGLISH_US;
|
|
}
|
|
|
|
if ( langFull == wxS("C") || langFull == wxS("POSIX") )
|
|
{
|
|
// default C locale is English too
|
|
return wxLANGUAGE_ENGLISH_US;
|
|
}
|
|
#endif
|
|
|
|
// the language string has the following form
|
|
//
|
|
// lang[_LANG][.encoding][@modifier]
|
|
//
|
|
// (see environ(5) in the Open Unix specification)
|
|
//
|
|
// where lang is the primary language, LANG is a sublang/territory,
|
|
// encoding is the charset to use and modifier "allows the user to select
|
|
// a specific instance of localization data within a single category"
|
|
//
|
|
// for example, the following strings are valid:
|
|
// fr
|
|
// fr_FR
|
|
// de_DE.iso88591
|
|
// de_DE@euro
|
|
// de_DE.iso88591@euro
|
|
|
|
// for now we don't use the encoding, although we probably should (doing
|
|
// translations of the msg catalogs on the fly as required) (TODO)
|
|
//
|
|
// we need the modified for languages like Valencian: ca_ES@valencia
|
|
// though, remember it
|
|
wxString modifier;
|
|
size_t posModifier = langFull.find_first_of(wxS("@"));
|
|
if ( posModifier != wxString::npos )
|
|
modifier = langFull.Mid(posModifier);
|
|
|
|
size_t posEndLang = langFull.find_first_of(wxS("@."));
|
|
if ( posEndLang != wxString::npos )
|
|
{
|
|
langFull.Truncate(posEndLang);
|
|
}
|
|
|
|
// in addition to the format above, we also can have full language names
|
|
// in LANG env var - for example, SuSE is known to use LANG="german" - so
|
|
// check for this
|
|
|
|
// do we have just the language (or sublang too)?
|
|
bool justLang = langFull.length() == LEN_LANG;
|
|
if ( justLang ||
|
|
(langFull.length() == LEN_FULL && langFull[LEN_LANG] == wxS('_')) )
|
|
{
|
|
// 0. Make sure the lang is according to latest ISO 639
|
|
// (this is necessary because glibc uses iw and in instead
|
|
// of he and id respectively).
|
|
|
|
// the language itself (second part is the dialect/sublang)
|
|
wxString langOrig = ExtractLang(langFull);
|
|
|
|
wxString lang;
|
|
if ( langOrig == wxS("iw"))
|
|
lang = wxS("he");
|
|
else if (langOrig == wxS("in"))
|
|
lang = wxS("id");
|
|
else if (langOrig == wxS("ji"))
|
|
lang = wxS("yi");
|
|
else if (langOrig == wxS("no_NO"))
|
|
lang = wxS("nb_NO");
|
|
else if (langOrig == wxS("no_NY"))
|
|
lang = wxS("nn_NO");
|
|
else if (langOrig == wxS("no"))
|
|
lang = wxS("nb_NO");
|
|
else
|
|
lang = langOrig;
|
|
|
|
// did we change it?
|
|
if ( lang != langOrig )
|
|
{
|
|
langFull = lang + ExtractNotLang(langFull);
|
|
}
|
|
|
|
// 1. Try to find the language either as is:
|
|
// a) With modifier if set
|
|
if ( !modifier.empty() )
|
|
{
|
|
wxString langFullWithModifier = langFull + modifier;
|
|
for ( i = 0; i < count; i++ )
|
|
{
|
|
if ( ms_languagesDB->Item(i).CanonicalName == langFullWithModifier )
|
|
break;
|
|
}
|
|
}
|
|
|
|
// b) Without modifier
|
|
if ( modifier.empty() || i == count )
|
|
{
|
|
for ( i = 0; i < count; i++ )
|
|
{
|
|
if ( ms_languagesDB->Item(i).CanonicalName == langFull )
|
|
break;
|
|
}
|
|
}
|
|
|
|
// 2. If langFull is of the form xx_YY, try to find xx:
|
|
if ( i == count && !justLang )
|
|
{
|
|
for ( i = 0; i < count; i++ )
|
|
{
|
|
if ( ms_languagesDB->Item(i).CanonicalName == lang )
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
// 3. If langFull is of the form xx, try to find any xx_YY record:
|
|
if ( i == count && justLang )
|
|
{
|
|
for ( i = 0; i < count; i++ )
|
|
{
|
|
if ( ExtractLang(ms_languagesDB->Item(i).CanonicalName)
|
|
== langFull )
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else // not standard format
|
|
{
|
|
// try to find the name in verbose description
|
|
for ( i = 0; i < count; i++ )
|
|
{
|
|
if (ms_languagesDB->Item(i).Description.CmpNoCase(langFull) == 0)
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
#elif defined(__WIN32__)
|
|
LCID lcid = GetUserDefaultLCID();
|
|
if ( lcid != 0 )
|
|
{
|
|
wxUint32 lang = PRIMARYLANGID(LANGIDFROMLCID(lcid));
|
|
wxUint32 sublang = SUBLANGID(LANGIDFROMLCID(lcid));
|
|
|
|
for ( i = 0; i < count; i++ )
|
|
{
|
|
if (ms_languagesDB->Item(i).WinLang == lang &&
|
|
ms_languagesDB->Item(i).WinSublang == sublang)
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
//else: leave wxlang == wxLANGUAGE_UNKNOWN
|
|
#endif // Unix/Win32
|
|
|
|
if ( i < count )
|
|
{
|
|
// we did find a matching entry, use it
|
|
return ms_languagesDB->Item(i).Language;
|
|
}
|
|
|
|
// no info about this language in the database
|
|
return wxLANGUAGE_UNKNOWN;
|
|
}
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// encoding stuff
|
|
// ----------------------------------------------------------------------------
|
|
|
|
// this is a bit strange as under Windows we get the encoding name using its
|
|
// numeric value and under Unix we do it the other way round, but this just
|
|
// reflects the way different systems provide the encoding info
|
|
|
|
/* static */
|
|
wxString wxLocale::GetSystemEncodingName()
|
|
{
|
|
wxString encname;
|
|
|
|
#if defined(__WIN32__) && !defined(__WXMICROWIN__)
|
|
// FIXME: what is the error return value for GetACP()?
|
|
UINT codepage = ::GetACP();
|
|
encname.Printf(wxS("windows-%u"), codepage);
|
|
#elif defined(__WXMAC__)
|
|
// default is just empty string, this resolves to the default system
|
|
// encoding later
|
|
#elif defined(__UNIX_LIKE__)
|
|
|
|
#if defined(HAVE_LANGINFO_H) && defined(CODESET)
|
|
// GNU libc provides current character set this way (this conforms
|
|
// to Unix98)
|
|
char *oldLocale = strdup(setlocale(LC_CTYPE, NULL));
|
|
setlocale(LC_CTYPE, "");
|
|
const char *alang = nl_langinfo(CODESET);
|
|
setlocale(LC_CTYPE, oldLocale);
|
|
free(oldLocale);
|
|
|
|
if ( alang )
|
|
{
|
|
encname = wxString::FromAscii( alang );
|
|
}
|
|
else // nl_langinfo() failed
|
|
#endif // HAVE_LANGINFO_H
|
|
{
|
|
// if we can't get at the character set directly, try to see if it's in
|
|
// the environment variables (in most cases this won't work, but I was
|
|
// out of ideas)
|
|
char *lang = getenv( "LC_ALL");
|
|
char *dot = lang ? strchr(lang, '.') : NULL;
|
|
if (!dot)
|
|
{
|
|
lang = getenv( "LC_CTYPE" );
|
|
if ( lang )
|
|
dot = strchr(lang, '.' );
|
|
}
|
|
if (!dot)
|
|
{
|
|
lang = getenv( "LANG");
|
|
if ( lang )
|
|
dot = strchr(lang, '.');
|
|
}
|
|
|
|
if ( dot )
|
|
{
|
|
encname = wxString::FromAscii( dot+1 );
|
|
}
|
|
}
|
|
#endif // Win32/Unix
|
|
|
|
return encname;
|
|
}
|
|
|
|
/* static */
|
|
wxFontEncoding wxLocale::GetSystemEncoding()
|
|
{
|
|
#if defined(__WIN32__) && !defined(__WXMICROWIN__)
|
|
UINT codepage = ::GetACP();
|
|
|
|
// wxWidgets only knows about CP1250-1257, 874, 932, 936, 949, 950
|
|
if ( codepage >= 1250 && codepage <= 1257 )
|
|
{
|
|
return (wxFontEncoding)(wxFONTENCODING_CP1250 + codepage - 1250);
|
|
}
|
|
|
|
if ( codepage == 874 )
|
|
{
|
|
return wxFONTENCODING_CP874;
|
|
}
|
|
|
|
if ( codepage == 932 )
|
|
{
|
|
return wxFONTENCODING_CP932;
|
|
}
|
|
|
|
if ( codepage == 936 )
|
|
{
|
|
return wxFONTENCODING_CP936;
|
|
}
|
|
|
|
if ( codepage == 949 )
|
|
{
|
|
return wxFONTENCODING_CP949;
|
|
}
|
|
|
|
if ( codepage == 950 )
|
|
{
|
|
return wxFONTENCODING_CP950;
|
|
}
|
|
#elif defined(__WXMAC__)
|
|
CFStringEncoding encoding = 0 ;
|
|
encoding = CFStringGetSystemEncoding() ;
|
|
return wxMacGetFontEncFromSystemEnc( encoding ) ;
|
|
#elif defined(__UNIX_LIKE__) && wxUSE_FONTMAP
|
|
const wxString encname = GetSystemEncodingName();
|
|
if ( !encname.empty() )
|
|
{
|
|
wxFontEncoding enc = wxFontMapperBase::GetEncodingFromName(encname);
|
|
|
|
// on some modern Linux systems (RedHat 8) the default system locale
|
|
// is UTF8 -- but it isn't supported by wxGTK1 in ANSI build at all so
|
|
// don't even try to use it in this case
|
|
#if !wxUSE_UNICODE && \
|
|
((defined(__WXGTK__) && !defined(__WXGTK20__)) || defined(__WXMOTIF__))
|
|
if ( enc == wxFONTENCODING_UTF8 )
|
|
{
|
|
// the most similar supported encoding...
|
|
enc = wxFONTENCODING_ISO8859_1;
|
|
}
|
|
#endif // !wxUSE_UNICODE
|
|
|
|
// GetEncodingFromName() returns wxFONTENCODING_DEFAULT for C locale
|
|
// (a.k.a. US-ASCII) which is arguably a bug but keep it like this for
|
|
// backwards compatibility and just take care to not return
|
|
// wxFONTENCODING_DEFAULT from here as this surely doesn't make sense
|
|
if ( enc == wxFONTENCODING_DEFAULT )
|
|
{
|
|
// we don't have wxFONTENCODING_ASCII, so use the closest one
|
|
return wxFONTENCODING_ISO8859_1;
|
|
}
|
|
|
|
if ( enc != wxFONTENCODING_MAX )
|
|
{
|
|
return enc;
|
|
}
|
|
//else: return wxFONTENCODING_SYSTEM below
|
|
}
|
|
#endif // Win32/Unix
|
|
|
|
return wxFONTENCODING_SYSTEM;
|
|
}
|
|
|
|
/* static */
|
|
void wxLocale::AddLanguage(const wxLanguageInfo& info)
|
|
{
|
|
CreateLanguagesDB();
|
|
ms_languagesDB->Add(info);
|
|
}
|
|
|
|
/* static */
|
|
const wxLanguageInfo *wxLocale::GetLanguageInfo(int lang)
|
|
{
|
|
CreateLanguagesDB();
|
|
|
|
// calling GetLanguageInfo(wxLANGUAGE_DEFAULT) is a natural thing to do, so
|
|
// make it work
|
|
if ( lang == wxLANGUAGE_DEFAULT )
|
|
lang = GetSystemLanguage();
|
|
|
|
const size_t count = ms_languagesDB->GetCount();
|
|
for ( size_t i = 0; i < count; i++ )
|
|
{
|
|
if ( ms_languagesDB->Item(i).Language == lang )
|
|
{
|
|
// We need to create a temporary here in order to make this work with BCC in final build mode
|
|
wxLanguageInfo *ptr = &ms_languagesDB->Item(i);
|
|
return ptr;
|
|
}
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
/* static */
|
|
wxString wxLocale::GetLanguageName(int lang)
|
|
{
|
|
const wxLanguageInfo *info = GetLanguageInfo(lang);
|
|
if ( !info )
|
|
return wxEmptyString;
|
|
else
|
|
return info->Description;
|
|
}
|
|
|
|
/* static */
|
|
const wxLanguageInfo *wxLocale::FindLanguageInfo(const wxString& locale)
|
|
{
|
|
CreateLanguagesDB();
|
|
|
|
const wxLanguageInfo *infoRet = NULL;
|
|
|
|
const size_t count = ms_languagesDB->GetCount();
|
|
for ( size_t i = 0; i < count; i++ )
|
|
{
|
|
const wxLanguageInfo *info = &ms_languagesDB->Item(i);
|
|
|
|
if ( wxStricmp(locale, info->CanonicalName) == 0 ||
|
|
wxStricmp(locale, info->Description) == 0 )
|
|
{
|
|
// exact match, stop searching
|
|
infoRet = info;
|
|
break;
|
|
}
|
|
|
|
if ( wxStricmp(locale, info->CanonicalName.BeforeFirst(wxS('_'))) == 0 )
|
|
{
|
|
// a match -- but maybe we'll find an exact one later, so continue
|
|
// looking
|
|
//
|
|
// OTOH, maybe we had already found a language match and in this
|
|
// case don't overwrite it because the entry for the default
|
|
// country always appears first in ms_languagesDB
|
|
if ( !infoRet )
|
|
infoRet = info;
|
|
}
|
|
}
|
|
|
|
return infoRet;
|
|
}
|
|
|
|
wxString wxLocale::GetSysName() const
|
|
{
|
|
return wxSetlocale(LC_ALL, NULL);
|
|
}
|
|
|
|
// clean up
|
|
wxLocale::~wxLocale()
|
|
{
|
|
// free memory
|
|
wxMsgCatalog *pTmpCat;
|
|
while ( m_pMsgCat != NULL ) {
|
|
pTmpCat = m_pMsgCat;
|
|
m_pMsgCat = m_pMsgCat->m_pNext;
|
|
delete pTmpCat;
|
|
}
|
|
|
|
// restore old locale pointer
|
|
wxSetLocale(m_pOldLocale);
|
|
|
|
wxSetlocale(LC_ALL, m_pszOldLocale);
|
|
free((wxChar *)m_pszOldLocale); // const_cast
|
|
}
|
|
|
|
// get the translation of given string in current locale
|
|
const wxString& wxLocale::GetString(const wxString& origString,
|
|
const wxString& domain) const
|
|
{
|
|
return GetString(origString, origString, size_t(-1), domain);
|
|
}
|
|
|
|
const wxString& wxLocale::GetString(const wxString& origString,
|
|
const wxString& origString2,
|
|
size_t n,
|
|
const wxString& domain) const
|
|
{
|
|
if ( origString.empty() )
|
|
return GetUntranslatedString(origString);
|
|
|
|
const wxString *trans = NULL;
|
|
wxMsgCatalog *pMsgCat;
|
|
|
|
if ( !domain.empty() )
|
|
{
|
|
pMsgCat = FindCatalog(domain);
|
|
|
|
// does the catalog exist?
|
|
if ( pMsgCat != NULL )
|
|
trans = pMsgCat->GetString(origString, n);
|
|
}
|
|
else
|
|
{
|
|
// search in all domains
|
|
for ( pMsgCat = m_pMsgCat; pMsgCat != NULL; pMsgCat = pMsgCat->m_pNext )
|
|
{
|
|
trans = pMsgCat->GetString(origString, n);
|
|
if ( trans != NULL ) // take the first found
|
|
break;
|
|
}
|
|
}
|
|
|
|
if ( trans == NULL )
|
|
{
|
|
#ifdef __WXDEBUG__
|
|
if ( !NoTransErr::Suppress() )
|
|
{
|
|
NoTransErr noTransErr;
|
|
|
|
wxLogTrace(TRACE_I18N,
|
|
wxS("string \"%s\"[%ld] not found in %slocale '%s'."),
|
|
origString, (long)n,
|
|
wxString::Format(wxS("domain '%s' "), domain).c_str(),
|
|
m_strLocale.c_str());
|
|
}
|
|
#endif // __WXDEBUG__
|
|
|
|
if (n == size_t(-1))
|
|
return GetUntranslatedString(origString);
|
|
else
|
|
return GetUntranslatedString(n == 1 ? origString : origString2);
|
|
}
|
|
|
|
return *trans;
|
|
}
|
|
|
|
WX_DECLARE_HASH_SET(wxString, wxStringHash, wxStringEqual,
|
|
wxLocaleUntranslatedStrings);
|
|
|
|
/* static */
|
|
const wxString& wxLocale::GetUntranslatedString(const wxString& str)
|
|
{
|
|
static wxLocaleUntranslatedStrings s_strings;
|
|
|
|
wxLocaleUntranslatedStrings::iterator i = s_strings.find(str);
|
|
if ( i == s_strings.end() )
|
|
return *s_strings.insert(str).first;
|
|
|
|
return *i;
|
|
}
|
|
|
|
wxString wxLocale::GetHeaderValue(const wxString& header,
|
|
const wxString& domain) const
|
|
{
|
|
if ( header.empty() )
|
|
return wxEmptyString;
|
|
|
|
const wxString *trans = NULL;
|
|
wxMsgCatalog *pMsgCat;
|
|
|
|
if ( !domain.empty() )
|
|
{
|
|
pMsgCat = FindCatalog(domain);
|
|
|
|
// does the catalog exist?
|
|
if ( pMsgCat == NULL )
|
|
return wxEmptyString;
|
|
|
|
trans = pMsgCat->GetString(wxEmptyString, (size_t)-1);
|
|
}
|
|
else
|
|
{
|
|
// search in all domains
|
|
for ( pMsgCat = m_pMsgCat; pMsgCat != NULL; pMsgCat = pMsgCat->m_pNext )
|
|
{
|
|
trans = pMsgCat->GetString(wxEmptyString, (size_t)-1);
|
|
if ( trans != NULL ) // take the first found
|
|
break;
|
|
}
|
|
}
|
|
|
|
if ( !trans || trans->empty() )
|
|
return wxEmptyString;
|
|
|
|
size_t found = trans->find(header);
|
|
if ( found == wxString::npos )
|
|
return wxEmptyString;
|
|
|
|
found += header.length() + 2 /* ': ' */;
|
|
|
|
// Every header is separated by \n
|
|
|
|
size_t endLine = trans->find(wxS('\n'), found);
|
|
size_t len = (endLine == wxString::npos) ?
|
|
wxString::npos : (endLine - found);
|
|
|
|
return trans->substr(found, len);
|
|
}
|
|
|
|
|
|
// find catalog by name in a linked list, return NULL if !found
|
|
wxMsgCatalog *wxLocale::FindCatalog(const wxString& domain) const
|
|
{
|
|
// linear search in the linked list
|
|
wxMsgCatalog *pMsgCat;
|
|
for ( pMsgCat = m_pMsgCat; pMsgCat != NULL; pMsgCat = pMsgCat->m_pNext )
|
|
{
|
|
if ( pMsgCat->GetName() == domain )
|
|
return pMsgCat;
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
// check if the given locale is provided by OS and C run time
|
|
/* static */
|
|
bool wxLocale::IsAvailable(int lang)
|
|
{
|
|
const wxLanguageInfo *info = wxLocale::GetLanguageInfo(lang);
|
|
wxCHECK_MSG( info, false, wxS("invalid language") );
|
|
|
|
#if defined(__WIN32__)
|
|
if ( !info->WinLang )
|
|
return false;
|
|
|
|
if ( !::IsValidLocale(info->GetLCID(), LCID_INSTALLED) )
|
|
return false;
|
|
|
|
#elif defined(__UNIX__)
|
|
|
|
// Test if setting the locale works, then set it back.
|
|
const char *oldLocale = wxSetlocale(LC_ALL, "");
|
|
const char *tmp = wxSetlocaleTryUTF8(LC_ALL, info->CanonicalName);
|
|
if ( !tmp )
|
|
{
|
|
// Some C libraries don't like xx_YY form and require xx only
|
|
tmp = wxSetlocaleTryUTF8(LC_ALL, ExtractLang(info->CanonicalName));
|
|
if ( !tmp )
|
|
return false;
|
|
}
|
|
// restore the original locale
|
|
wxSetlocale(LC_ALL, oldLocale);
|
|
#endif
|
|
|
|
return true;
|
|
}
|
|
|
|
// check if the given catalog is loaded
|
|
bool wxLocale::IsLoaded(const wxString& szDomain) const
|
|
{
|
|
return FindCatalog(szDomain) != NULL;
|
|
}
|
|
|
|
// add a catalog to our linked list
|
|
bool wxLocale::AddCatalog(const wxString& szDomain)
|
|
{
|
|
return AddCatalog(szDomain, wxLANGUAGE_ENGLISH_US, wxEmptyString);
|
|
}
|
|
|
|
// add a catalog to our linked list
|
|
bool wxLocale::AddCatalog(const wxString& szDomain,
|
|
wxLanguage msgIdLanguage,
|
|
const wxString& msgIdCharset)
|
|
|
|
{
|
|
wxCHECK_MSG( !m_strShort.empty(), false, "must initialize catalog first" );
|
|
|
|
|
|
// It is OK to not load catalog if the msgid language and m_language match,
|
|
// in which case we can directly display the texts embedded in program's
|
|
// source code:
|
|
if ( msgIdLanguage == m_language )
|
|
return true;
|
|
|
|
|
|
wxMsgCatalog *pMsgCat = new wxMsgCatalog;
|
|
|
|
if ( pMsgCat->Load(m_strShort, szDomain, msgIdCharset, m_bConvertEncoding) )
|
|
{
|
|
// add it to the head of the list so that in GetString it will
|
|
// be searched before the catalogs added earlier
|
|
pMsgCat->m_pNext = m_pMsgCat;
|
|
m_pMsgCat = pMsgCat;
|
|
|
|
return true;
|
|
}
|
|
|
|
// don't add it because it couldn't be loaded anyway
|
|
delete pMsgCat;
|
|
|
|
|
|
// If there's no exact match, we may still get partial match where the
|
|
// (basic) language is same, but the country differs. For example, it's
|
|
// permitted to use en_US strings from sources even if m_language is en_GB:
|
|
const wxLanguageInfo *msgIdLangInfo = GetLanguageInfo(msgIdLanguage);
|
|
if ( msgIdLangInfo &&
|
|
ExtractLang(msgIdLangInfo->CanonicalName) == ExtractLang(m_strShort) )
|
|
{
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// accessors for locale-dependent data
|
|
// ----------------------------------------------------------------------------
|
|
|
|
#if defined(__WXMSW__)
|
|
|
|
/* static */
|
|
wxString wxLocale::GetInfo(wxLocaleInfo index, wxLocaleCategory WXUNUSED(cat))
|
|
{
|
|
wxUint32 lcid = LOCALE_USER_DEFAULT;
|
|
|
|
if (wxGetLocale())
|
|
{
|
|
const wxLanguageInfo *info = GetLanguageInfo(wxGetLocale()->GetLanguage());
|
|
if ( info )
|
|
lcid = info->GetLCID();
|
|
}
|
|
|
|
wxString str;
|
|
wxChar buffer[256];
|
|
size_t count;
|
|
buffer[0] = wxS('\0');
|
|
switch (index)
|
|
{
|
|
case wxLOCALE_DECIMAL_POINT:
|
|
count = ::GetLocaleInfo(lcid, LOCALE_SDECIMAL, buffer, 256);
|
|
if (!count)
|
|
str << wxS(".");
|
|
else
|
|
str << buffer;
|
|
break;
|
|
#if 0
|
|
case wxSYS_LIST_SEPARATOR:
|
|
count = ::GetLocaleInfo(lcid, LOCALE_SLIST, buffer, 256);
|
|
if (!count)
|
|
str << wxS(",");
|
|
else
|
|
str << buffer;
|
|
break;
|
|
case wxSYS_LEADING_ZERO: // 0 means no leading zero, 1 means leading zero
|
|
count = ::GetLocaleInfo(lcid, LOCALE_ILZERO, buffer, 256);
|
|
if (!count)
|
|
str << wxS("0");
|
|
else
|
|
str << buffer;
|
|
break;
|
|
#endif
|
|
default:
|
|
wxFAIL_MSG(wxS("Unknown System String !"));
|
|
}
|
|
return str;
|
|
}
|
|
|
|
#elif defined(__DARWIN__)
|
|
|
|
/* static */
|
|
wxString wxLocale::GetInfo(wxLocaleInfo index, wxLocaleCategory WXUNUSED(cat))
|
|
{
|
|
CFLocaleRef userLocaleRefRaw;
|
|
if ( wxGetLocale() )
|
|
{
|
|
userLocaleRefRaw = CFLocaleCreate
|
|
(
|
|
kCFAllocatorDefault,
|
|
wxCFStringRef(wxGetLocale()->GetCanonicalName())
|
|
);
|
|
}
|
|
else // no current locale, use the default one
|
|
{
|
|
userLocaleRefRaw = CFLocaleCopyCurrent();
|
|
}
|
|
|
|
wxCFRef<CFLocaleRef> userLocaleRef(userLocaleRefRaw);
|
|
|
|
CFStringRef cfstr = 0;
|
|
switch ( index )
|
|
{
|
|
case wxLOCALE_THOUSANDS_SEP:
|
|
cfstr = (CFStringRef) CFLocaleGetValue(userLocaleRef, kCFLocaleGroupingSeparator);
|
|
break;
|
|
|
|
case wxLOCALE_DECIMAL_POINT:
|
|
cfstr = (CFStringRef) CFLocaleGetValue(userLocaleRef, kCFLocaleDecimalSeparator);
|
|
break;
|
|
|
|
default:
|
|
wxFAIL_MSG( "Unknown locale info" );
|
|
cfstr = CFSTR("");
|
|
break;
|
|
}
|
|
|
|
wxCFStringRef str(wxCFRetain(cfstr));
|
|
return str.AsString();
|
|
}
|
|
|
|
#else // !__WXMSW__ && !__DARWIN__
|
|
|
|
/* static */
|
|
wxString wxLocale::GetInfo(wxLocaleInfo index, wxLocaleCategory cat)
|
|
{
|
|
struct lconv *locale_info = localeconv();
|
|
switch (cat)
|
|
{
|
|
case wxLOCALE_CAT_NUMBER:
|
|
switch (index)
|
|
{
|
|
case wxLOCALE_THOUSANDS_SEP:
|
|
return wxString(locale_info->thousands_sep,
|
|
*wxConvCurrent);
|
|
case wxLOCALE_DECIMAL_POINT:
|
|
return wxString(locale_info->decimal_point,
|
|
*wxConvCurrent);
|
|
default:
|
|
return wxEmptyString;
|
|
}
|
|
case wxLOCALE_CAT_MONEY:
|
|
switch (index)
|
|
{
|
|
case wxLOCALE_THOUSANDS_SEP:
|
|
return wxString(locale_info->mon_thousands_sep,
|
|
*wxConvCurrent);
|
|
case wxLOCALE_DECIMAL_POINT:
|
|
return wxString(locale_info->mon_decimal_point,
|
|
*wxConvCurrent);
|
|
default:
|
|
return wxEmptyString;
|
|
}
|
|
default:
|
|
return wxEmptyString;
|
|
}
|
|
}
|
|
|
|
#endif // platform
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// global functions and variables
|
|
// ----------------------------------------------------------------------------
|
|
|
|
// retrieve/change current locale
|
|
// ------------------------------
|
|
|
|
// the current locale object
|
|
static wxLocale *g_pLocale = NULL;
|
|
|
|
wxLocale *wxGetLocale()
|
|
{
|
|
return g_pLocale;
|
|
}
|
|
|
|
wxLocale *wxSetLocale(wxLocale *pLocale)
|
|
{
|
|
wxLocale *pOld = g_pLocale;
|
|
g_pLocale = pLocale;
|
|
return pOld;
|
|
}
|
|
|
|
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// wxLocale module (for lazy destruction of languagesDB)
|
|
// ----------------------------------------------------------------------------
|
|
|
|
class wxLocaleModule: public wxModule
|
|
{
|
|
DECLARE_DYNAMIC_CLASS(wxLocaleModule)
|
|
public:
|
|
wxLocaleModule() {}
|
|
bool OnInit() { return true; }
|
|
void OnExit() { wxLocale::DestroyLanguagesDB(); }
|
|
};
|
|
|
|
IMPLEMENT_DYNAMIC_CLASS(wxLocaleModule, wxModule)
|
|
|
|
|
|
|
|
// ----------------------------------------------------------------------------
|
|
// default languages table & initialization
|
|
// ----------------------------------------------------------------------------
|
|
|
|
|
|
|
|
// --- --- --- generated code begins here --- --- ---
|
|
|
|
// This table is generated by misc/languages/genlang.py
|
|
// When making changes, please put them into misc/languages/langtabl.txt
|
|
|
|
#if !defined(__WIN32__) || defined(__WXMICROWIN__)
|
|
|
|
#define SETWINLANG(info,lang,sublang)
|
|
|
|
#else
|
|
|
|
#define SETWINLANG(info,lang,sublang) \
|
|
info.WinLang = lang, info.WinSublang = sublang;
|
|
|
|
#ifndef LANG_AFRIKAANS
|
|
#define LANG_AFRIKAANS (0)
|
|
#endif
|
|
#ifndef LANG_ALBANIAN
|
|
#define LANG_ALBANIAN (0)
|
|
#endif
|
|
#ifndef LANG_ARABIC
|
|
#define LANG_ARABIC (0)
|
|
#endif
|
|
#ifndef LANG_ARMENIAN
|
|
#define LANG_ARMENIAN (0)
|
|
#endif
|
|
#ifndef LANG_ASSAMESE
|
|
#define LANG_ASSAMESE (0)
|
|
#endif
|
|
#ifndef LANG_AZERI
|
|
#define LANG_AZERI (0)
|
|
#endif
|
|
#ifndef LANG_BASQUE
|
|
#define LANG_BASQUE (0)
|
|
#endif
|
|
#ifndef LANG_BELARUSIAN
|
|
#define LANG_BELARUSIAN (0)
|
|
#endif
|
|
#ifndef LANG_BENGALI
|
|
#define LANG_BENGALI (0)
|
|
#endif
|
|
#ifndef LANG_BULGARIAN
|
|
#define LANG_BULGARIAN (0)
|
|
#endif
|
|
#ifndef LANG_CATALAN
|
|
#define LANG_CATALAN (0)
|
|
#endif
|
|
#ifndef LANG_CHINESE
|
|
#define LANG_CHINESE (0)
|
|
#endif
|
|
#ifndef LANG_CROATIAN
|
|
#define LANG_CROATIAN (0)
|
|
#endif
|
|
#ifndef LANG_CZECH
|
|
#define LANG_CZECH (0)
|
|
#endif
|
|
#ifndef LANG_DANISH
|
|
#define LANG_DANISH (0)
|
|
#endif
|
|
#ifndef LANG_DUTCH
|
|
#define LANG_DUTCH (0)
|
|
#endif
|
|
#ifndef LANG_ENGLISH
|
|
#define LANG_ENGLISH (0)
|
|
#endif
|
|
#ifndef LANG_ESTONIAN
|
|
#define LANG_ESTONIAN (0)
|
|
#endif
|
|
#ifndef LANG_FAEROESE
|
|
#define LANG_FAEROESE (0)
|
|
#endif
|
|
#ifndef LANG_FARSI
|
|
#define LANG_FARSI (0)
|
|
#endif
|
|
#ifndef LANG_FINNISH
|
|
#define LANG_FINNISH (0)
|
|
#endif
|
|
#ifndef LANG_FRENCH
|
|
#define LANG_FRENCH (0)
|
|
#endif
|
|
#ifndef LANG_GEORGIAN
|
|
#define LANG_GEORGIAN (0)
|
|
#endif
|
|
#ifndef LANG_GERMAN
|
|
#define LANG_GERMAN (0)
|
|
#endif
|
|
#ifndef LANG_GREEK
|
|
#define LANG_GREEK (0)
|
|
#endif
|
|
#ifndef LANG_GUJARATI
|
|
#define LANG_GUJARATI (0)
|
|
#endif
|
|
#ifndef LANG_HEBREW
|
|
#define LANG_HEBREW (0)
|
|
#endif
|
|
#ifndef LANG_HINDI
|
|
#define LANG_HINDI (0)
|
|
#endif
|
|
#ifndef LANG_HUNGARIAN
|
|
#define LANG_HUNGARIAN (0)
|
|
#endif
|
|
#ifndef LANG_ICELANDIC
|
|
#define LANG_ICELANDIC (0)
|
|
#endif
|
|
#ifndef LANG_INDONESIAN
|
|
#define LANG_INDONESIAN (0)
|
|
#endif
|
|
#ifndef LANG_ITALIAN
|
|
#define LANG_ITALIAN (0)
|
|
#endif
|
|
#ifndef LANG_JAPANESE
|
|
#define LANG_JAPANESE (0)
|
|
#endif
|
|
#ifndef LANG_KANNADA
|
|
#define LANG_KANNADA (0)
|
|
#endif
|
|
#ifndef LANG_KASHMIRI
|
|
#define LANG_KASHMIRI (0)
|
|
#endif
|
|
#ifndef LANG_KAZAK
|
|
#define LANG_KAZAK (0)
|
|
#endif
|
|
#ifndef LANG_KONKANI
|
|
#define LANG_KONKANI (0)
|
|
#endif
|
|
#ifndef LANG_KOREAN
|
|
#define LANG_KOREAN (0)
|
|
#endif
|
|
#ifndef LANG_LATVIAN
|
|
#define LANG_LATVIAN (0)
|
|
#endif
|
|
#ifndef LANG_LITHUANIAN
|
|
#define LANG_LITHUANIAN (0)
|
|
#endif
|
|
#ifndef LANG_MACEDONIAN
|
|
#define LANG_MACEDONIAN (0)
|
|
#endif
|
|
#ifndef LANG_MALAY
|
|
#define LANG_MALAY (0)
|
|
#endif
|
|
#ifndef LANG_MALAYALAM
|
|
#define LANG_MALAYALAM (0)
|
|
#endif
|
|
#ifndef LANG_MANIPURI
|
|
#define LANG_MANIPURI (0)
|
|
#endif
|
|
#ifndef LANG_MARATHI
|
|
#define LANG_MARATHI (0)
|
|
#endif
|
|
#ifndef LANG_NEPALI
|
|
#define LANG_NEPALI (0)
|
|
#endif
|
|
#ifndef LANG_NORWEGIAN
|
|
#define LANG_NORWEGIAN (0)
|
|
#endif
|
|
#ifndef LANG_ORIYA
|
|
#define LANG_ORIYA (0)
|
|
#endif
|
|
#ifndef LANG_POLISH
|
|
#define LANG_POLISH (0)
|
|
#endif
|
|
#ifndef LANG_PORTUGUESE
|
|
#define LANG_PORTUGUESE (0)
|
|
#endif
|
|
#ifndef LANG_PUNJABI
|
|
#define LANG_PUNJABI (0)
|
|
#endif
|
|
#ifndef LANG_ROMANIAN
|
|
#define LANG_ROMANIAN (0)
|
|
#endif
|
|
#ifndef LANG_RUSSIAN
|
|
#define LANG_RUSSIAN (0)
|
|
#endif
|
|
#ifndef LANG_SAMI
|
|
#define LANG_SAMI (0)
|
|
#endif
|
|
#ifndef LANG_SANSKRIT
|
|
#define LANG_SANSKRIT (0)
|
|
#endif
|
|
#ifndef LANG_SERBIAN
|
|
#define LANG_SERBIAN (0)
|
|
#endif
|
|
#ifndef LANG_SINDHI
|
|
#define LANG_SINDHI (0)
|
|
#endif
|
|
#ifndef LANG_SLOVAK
|
|
#define LANG_SLOVAK (0)
|
|
#endif
|
|
#ifndef LANG_SLOVENIAN
|
|
#define LANG_SLOVENIAN (0)
|
|
#endif
|
|
#ifndef LANG_SPANISH
|
|
#define LANG_SPANISH (0)
|
|
#endif
|
|
#ifndef LANG_SWAHILI
|
|
#define LANG_SWAHILI (0)
|
|
#endif
|
|
#ifndef LANG_SWEDISH
|
|
#define LANG_SWEDISH (0)
|
|
#endif
|
|
#ifndef LANG_TAMIL
|
|
#define LANG_TAMIL (0)
|
|
#endif
|
|
#ifndef LANG_TATAR
|
|
#define LANG_TATAR (0)
|
|
#endif
|
|
#ifndef LANG_TELUGU
|
|
#define LANG_TELUGU (0)
|
|
#endif
|
|
#ifndef LANG_THAI
|
|
#define LANG_THAI (0)
|
|
#endif
|
|
#ifndef LANG_TURKISH
|
|
#define LANG_TURKISH (0)
|
|
#endif
|
|
#ifndef LANG_UKRAINIAN
|
|
#define LANG_UKRAINIAN (0)
|
|
#endif
|
|
#ifndef LANG_URDU
|
|
#define LANG_URDU (0)
|
|
#endif
|
|
#ifndef LANG_UZBEK
|
|
#define LANG_UZBEK (0)
|
|
#endif
|
|
#ifndef LANG_VIETNAMESE
|
|
#define LANG_VIETNAMESE (0)
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_ALGERIA
|
|
#define SUBLANG_ARABIC_ALGERIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_BAHRAIN
|
|
#define SUBLANG_ARABIC_BAHRAIN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_EGYPT
|
|
#define SUBLANG_ARABIC_EGYPT SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_IRAQ
|
|
#define SUBLANG_ARABIC_IRAQ SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_JORDAN
|
|
#define SUBLANG_ARABIC_JORDAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_KUWAIT
|
|
#define SUBLANG_ARABIC_KUWAIT SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_LEBANON
|
|
#define SUBLANG_ARABIC_LEBANON SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_LIBYA
|
|
#define SUBLANG_ARABIC_LIBYA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_MOROCCO
|
|
#define SUBLANG_ARABIC_MOROCCO SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_OMAN
|
|
#define SUBLANG_ARABIC_OMAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_QATAR
|
|
#define SUBLANG_ARABIC_QATAR SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_SAUDI_ARABIA
|
|
#define SUBLANG_ARABIC_SAUDI_ARABIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_SYRIA
|
|
#define SUBLANG_ARABIC_SYRIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_TUNISIA
|
|
#define SUBLANG_ARABIC_TUNISIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_UAE
|
|
#define SUBLANG_ARABIC_UAE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ARABIC_YEMEN
|
|
#define SUBLANG_ARABIC_YEMEN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_AZERI_CYRILLIC
|
|
#define SUBLANG_AZERI_CYRILLIC SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_AZERI_LATIN
|
|
#define SUBLANG_AZERI_LATIN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_CHINESE_SIMPLIFIED
|
|
#define SUBLANG_CHINESE_SIMPLIFIED SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_CHINESE_TRADITIONAL
|
|
#define SUBLANG_CHINESE_TRADITIONAL SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_CHINESE_HONGKONG
|
|
#define SUBLANG_CHINESE_HONGKONG SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_CHINESE_MACAU
|
|
#define SUBLANG_CHINESE_MACAU SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_CHINESE_SINGAPORE
|
|
#define SUBLANG_CHINESE_SINGAPORE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_DUTCH
|
|
#define SUBLANG_DUTCH SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_DUTCH_BELGIAN
|
|
#define SUBLANG_DUTCH_BELGIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_UK
|
|
#define SUBLANG_ENGLISH_UK SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_US
|
|
#define SUBLANG_ENGLISH_US SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_AUS
|
|
#define SUBLANG_ENGLISH_AUS SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_BELIZE
|
|
#define SUBLANG_ENGLISH_BELIZE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_CAN
|
|
#define SUBLANG_ENGLISH_CAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_CARIBBEAN
|
|
#define SUBLANG_ENGLISH_CARIBBEAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_EIRE
|
|
#define SUBLANG_ENGLISH_EIRE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_JAMAICA
|
|
#define SUBLANG_ENGLISH_JAMAICA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_NZ
|
|
#define SUBLANG_ENGLISH_NZ SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_PHILIPPINES
|
|
#define SUBLANG_ENGLISH_PHILIPPINES SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_SOUTH_AFRICA
|
|
#define SUBLANG_ENGLISH_SOUTH_AFRICA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_TRINIDAD
|
|
#define SUBLANG_ENGLISH_TRINIDAD SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ENGLISH_ZIMBABWE
|
|
#define SUBLANG_ENGLISH_ZIMBABWE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_FRENCH
|
|
#define SUBLANG_FRENCH SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_FRENCH_BELGIAN
|
|
#define SUBLANG_FRENCH_BELGIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_FRENCH_CANADIAN
|
|
#define SUBLANG_FRENCH_CANADIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_FRENCH_LUXEMBOURG
|
|
#define SUBLANG_FRENCH_LUXEMBOURG SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_FRENCH_MONACO
|
|
#define SUBLANG_FRENCH_MONACO SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_FRENCH_SWISS
|
|
#define SUBLANG_FRENCH_SWISS SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_GERMAN
|
|
#define SUBLANG_GERMAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_GERMAN_AUSTRIAN
|
|
#define SUBLANG_GERMAN_AUSTRIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_GERMAN_LIECHTENSTEIN
|
|
#define SUBLANG_GERMAN_LIECHTENSTEIN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_GERMAN_LUXEMBOURG
|
|
#define SUBLANG_GERMAN_LUXEMBOURG SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_GERMAN_SWISS
|
|
#define SUBLANG_GERMAN_SWISS SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ITALIAN
|
|
#define SUBLANG_ITALIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_ITALIAN_SWISS
|
|
#define SUBLANG_ITALIAN_SWISS SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_KASHMIRI_INDIA
|
|
#define SUBLANG_KASHMIRI_INDIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_KOREAN
|
|
#define SUBLANG_KOREAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_LITHUANIAN
|
|
#define SUBLANG_LITHUANIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_MALAY_BRUNEI_DARUSSALAM
|
|
#define SUBLANG_MALAY_BRUNEI_DARUSSALAM SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_MALAY_MALAYSIA
|
|
#define SUBLANG_MALAY_MALAYSIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_NEPALI_INDIA
|
|
#define SUBLANG_NEPALI_INDIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_NORWEGIAN_BOKMAL
|
|
#define SUBLANG_NORWEGIAN_BOKMAL SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_NORWEGIAN_NYNORSK
|
|
#define SUBLANG_NORWEGIAN_NYNORSK SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_PORTUGUESE
|
|
#define SUBLANG_PORTUGUESE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_PORTUGUESE_BRAZILIAN
|
|
#define SUBLANG_PORTUGUESE_BRAZILIAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SERBIAN_CYRILLIC
|
|
#define SUBLANG_SERBIAN_CYRILLIC SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SERBIAN_LATIN
|
|
#define SUBLANG_SERBIAN_LATIN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH
|
|
#define SUBLANG_SPANISH SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_ARGENTINA
|
|
#define SUBLANG_SPANISH_ARGENTINA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_BOLIVIA
|
|
#define SUBLANG_SPANISH_BOLIVIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_CHILE
|
|
#define SUBLANG_SPANISH_CHILE SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_COLOMBIA
|
|
#define SUBLANG_SPANISH_COLOMBIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_COSTA_RICA
|
|
#define SUBLANG_SPANISH_COSTA_RICA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_DOMINICAN_REPUBLIC
|
|
#define SUBLANG_SPANISH_DOMINICAN_REPUBLIC SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_ECUADOR
|
|
#define SUBLANG_SPANISH_ECUADOR SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_EL_SALVADOR
|
|
#define SUBLANG_SPANISH_EL_SALVADOR SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_GUATEMALA
|
|
#define SUBLANG_SPANISH_GUATEMALA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_HONDURAS
|
|
#define SUBLANG_SPANISH_HONDURAS SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_MEXICAN
|
|
#define SUBLANG_SPANISH_MEXICAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_MODERN
|
|
#define SUBLANG_SPANISH_MODERN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_NICARAGUA
|
|
#define SUBLANG_SPANISH_NICARAGUA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_PANAMA
|
|
#define SUBLANG_SPANISH_PANAMA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_PARAGUAY
|
|
#define SUBLANG_SPANISH_PARAGUAY SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_PERU
|
|
#define SUBLANG_SPANISH_PERU SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_PUERTO_RICO
|
|
#define SUBLANG_SPANISH_PUERTO_RICO SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_URUGUAY
|
|
#define SUBLANG_SPANISH_URUGUAY SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SPANISH_VENEZUELA
|
|
#define SUBLANG_SPANISH_VENEZUELA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SWEDISH
|
|
#define SUBLANG_SWEDISH SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_SWEDISH_FINLAND
|
|
#define SUBLANG_SWEDISH_FINLAND SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_URDU_INDIA
|
|
#define SUBLANG_URDU_INDIA SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_URDU_PAKISTAN
|
|
#define SUBLANG_URDU_PAKISTAN SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_UZBEK_CYRILLIC
|
|
#define SUBLANG_UZBEK_CYRILLIC SUBLANG_DEFAULT
|
|
#endif
|
|
#ifndef SUBLANG_UZBEK_LATIN
|
|
#define SUBLANG_UZBEK_LATIN SUBLANG_DEFAULT
|
|
#endif
|
|
|
|
|
|
#endif // __WIN32__
|
|
|
|
#define LNG(wxlang, canonical, winlang, winsublang, layout, desc) \
|
|
info.Language = wxlang; \
|
|
info.CanonicalName = wxS(canonical); \
|
|
info.LayoutDirection = layout; \
|
|
info.Description = wxS(desc); \
|
|
SETWINLANG(info, winlang, winsublang) \
|
|
AddLanguage(info);
|
|
|
|
void wxLocale::InitLanguagesDB()
|
|
{
|
|
wxLanguageInfo info;
|
|
wxStringTokenizer tkn;
|
|
|
|
LNG(wxLANGUAGE_ABKHAZIAN, "ab" , 0 , 0 , wxLayout_LeftToRight, "Abkhazian")
|
|
LNG(wxLANGUAGE_AFAR, "aa" , 0 , 0 , wxLayout_LeftToRight, "Afar")
|
|
LNG(wxLANGUAGE_AFRIKAANS, "af_ZA", LANG_AFRIKAANS , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Afrikaans")
|
|
LNG(wxLANGUAGE_ALBANIAN, "sq_AL", LANG_ALBANIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Albanian")
|
|
LNG(wxLANGUAGE_AMHARIC, "am" , 0 , 0 , wxLayout_LeftToRight, "Amharic")
|
|
LNG(wxLANGUAGE_ARABIC, "ar" , LANG_ARABIC , SUBLANG_DEFAULT , wxLayout_RightToLeft, "Arabic")
|
|
LNG(wxLANGUAGE_ARABIC_ALGERIA, "ar_DZ", LANG_ARABIC , SUBLANG_ARABIC_ALGERIA , wxLayout_RightToLeft, "Arabic (Algeria)")
|
|
LNG(wxLANGUAGE_ARABIC_BAHRAIN, "ar_BH", LANG_ARABIC , SUBLANG_ARABIC_BAHRAIN , wxLayout_RightToLeft, "Arabic (Bahrain)")
|
|
LNG(wxLANGUAGE_ARABIC_EGYPT, "ar_EG", LANG_ARABIC , SUBLANG_ARABIC_EGYPT , wxLayout_RightToLeft, "Arabic (Egypt)")
|
|
LNG(wxLANGUAGE_ARABIC_IRAQ, "ar_IQ", LANG_ARABIC , SUBLANG_ARABIC_IRAQ , wxLayout_RightToLeft, "Arabic (Iraq)")
|
|
LNG(wxLANGUAGE_ARABIC_JORDAN, "ar_JO", LANG_ARABIC , SUBLANG_ARABIC_JORDAN , wxLayout_RightToLeft, "Arabic (Jordan)")
|
|
LNG(wxLANGUAGE_ARABIC_KUWAIT, "ar_KW", LANG_ARABIC , SUBLANG_ARABIC_KUWAIT , wxLayout_RightToLeft, "Arabic (Kuwait)")
|
|
LNG(wxLANGUAGE_ARABIC_LEBANON, "ar_LB", LANG_ARABIC , SUBLANG_ARABIC_LEBANON , wxLayout_RightToLeft, "Arabic (Lebanon)")
|
|
LNG(wxLANGUAGE_ARABIC_LIBYA, "ar_LY", LANG_ARABIC , SUBLANG_ARABIC_LIBYA , wxLayout_RightToLeft, "Arabic (Libya)")
|
|
LNG(wxLANGUAGE_ARABIC_MOROCCO, "ar_MA", LANG_ARABIC , SUBLANG_ARABIC_MOROCCO , wxLayout_RightToLeft, "Arabic (Morocco)")
|
|
LNG(wxLANGUAGE_ARABIC_OMAN, "ar_OM", LANG_ARABIC , SUBLANG_ARABIC_OMAN , wxLayout_RightToLeft, "Arabic (Oman)")
|
|
LNG(wxLANGUAGE_ARABIC_QATAR, "ar_QA", LANG_ARABIC , SUBLANG_ARABIC_QATAR , wxLayout_RightToLeft, "Arabic (Qatar)")
|
|
LNG(wxLANGUAGE_ARABIC_SAUDI_ARABIA, "ar_SA", LANG_ARABIC , SUBLANG_ARABIC_SAUDI_ARABIA , wxLayout_RightToLeft, "Arabic (Saudi Arabia)")
|
|
LNG(wxLANGUAGE_ARABIC_SUDAN, "ar_SD", 0 , 0 , wxLayout_RightToLeft, "Arabic (Sudan)")
|
|
LNG(wxLANGUAGE_ARABIC_SYRIA, "ar_SY", LANG_ARABIC , SUBLANG_ARABIC_SYRIA , wxLayout_RightToLeft, "Arabic (Syria)")
|
|
LNG(wxLANGUAGE_ARABIC_TUNISIA, "ar_TN", LANG_ARABIC , SUBLANG_ARABIC_TUNISIA , wxLayout_RightToLeft, "Arabic (Tunisia)")
|
|
LNG(wxLANGUAGE_ARABIC_UAE, "ar_AE", LANG_ARABIC , SUBLANG_ARABIC_UAE , wxLayout_RightToLeft, "Arabic (Uae)")
|
|
LNG(wxLANGUAGE_ARABIC_YEMEN, "ar_YE", LANG_ARABIC , SUBLANG_ARABIC_YEMEN , wxLayout_RightToLeft, "Arabic (Yemen)")
|
|
LNG(wxLANGUAGE_ARMENIAN, "hy" , LANG_ARMENIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Armenian")
|
|
LNG(wxLANGUAGE_ASSAMESE, "as" , LANG_ASSAMESE , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Assamese")
|
|
LNG(wxLANGUAGE_AYMARA, "ay" , 0 , 0 , wxLayout_LeftToRight, "Aymara")
|
|
LNG(wxLANGUAGE_AZERI, "az" , LANG_AZERI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Azeri")
|
|
LNG(wxLANGUAGE_AZERI_CYRILLIC, "az" , LANG_AZERI , SUBLANG_AZERI_CYRILLIC , wxLayout_LeftToRight, "Azeri (Cyrillic)")
|
|
LNG(wxLANGUAGE_AZERI_LATIN, "az" , LANG_AZERI , SUBLANG_AZERI_LATIN , wxLayout_LeftToRight, "Azeri (Latin)")
|
|
LNG(wxLANGUAGE_BASHKIR, "ba" , 0 , 0 , wxLayout_LeftToRight, "Bashkir")
|
|
LNG(wxLANGUAGE_BASQUE, "eu_ES", LANG_BASQUE , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Basque")
|
|
LNG(wxLANGUAGE_BELARUSIAN, "be_BY", LANG_BELARUSIAN, SUBLANG_DEFAULT , wxLayout_LeftToRight, "Belarusian")
|
|
LNG(wxLANGUAGE_BENGALI, "bn" , LANG_BENGALI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Bengali")
|
|
LNG(wxLANGUAGE_BHUTANI, "dz" , 0 , 0 , wxLayout_LeftToRight, "Bhutani")
|
|
LNG(wxLANGUAGE_BIHARI, "bh" , 0 , 0 , wxLayout_LeftToRight, "Bihari")
|
|
LNG(wxLANGUAGE_BISLAMA, "bi" , 0 , 0 , wxLayout_LeftToRight, "Bislama")
|
|
LNG(wxLANGUAGE_BRETON, "br" , 0 , 0 , wxLayout_LeftToRight, "Breton")
|
|
LNG(wxLANGUAGE_BULGARIAN, "bg_BG", LANG_BULGARIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Bulgarian")
|
|
LNG(wxLANGUAGE_BURMESE, "my" , 0 , 0 , wxLayout_LeftToRight, "Burmese")
|
|
LNG(wxLANGUAGE_CAMBODIAN, "km" , 0 , 0 , wxLayout_LeftToRight, "Cambodian")
|
|
LNG(wxLANGUAGE_CATALAN, "ca_ES", LANG_CATALAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Catalan")
|
|
LNG(wxLANGUAGE_CHINESE, "zh_TW", LANG_CHINESE , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Chinese")
|
|
LNG(wxLANGUAGE_CHINESE_SIMPLIFIED, "zh_CN", LANG_CHINESE , SUBLANG_CHINESE_SIMPLIFIED , wxLayout_LeftToRight, "Chinese (Simplified)")
|
|
LNG(wxLANGUAGE_CHINESE_TRADITIONAL, "zh_TW", LANG_CHINESE , SUBLANG_CHINESE_TRADITIONAL , wxLayout_LeftToRight, "Chinese (Traditional)")
|
|
LNG(wxLANGUAGE_CHINESE_HONGKONG, "zh_HK", LANG_CHINESE , SUBLANG_CHINESE_HONGKONG , wxLayout_LeftToRight, "Chinese (Hongkong)")
|
|
LNG(wxLANGUAGE_CHINESE_MACAU, "zh_MO", LANG_CHINESE , SUBLANG_CHINESE_MACAU , wxLayout_LeftToRight, "Chinese (Macau)")
|
|
LNG(wxLANGUAGE_CHINESE_SINGAPORE, "zh_SG", LANG_CHINESE , SUBLANG_CHINESE_SINGAPORE , wxLayout_LeftToRight, "Chinese (Singapore)")
|
|
LNG(wxLANGUAGE_CHINESE_TAIWAN, "zh_TW", LANG_CHINESE , SUBLANG_CHINESE_TRADITIONAL , wxLayout_LeftToRight, "Chinese (Taiwan)")
|
|
LNG(wxLANGUAGE_CORSICAN, "co" , 0 , 0 , wxLayout_LeftToRight, "Corsican")
|
|
LNG(wxLANGUAGE_CROATIAN, "hr_HR", LANG_CROATIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Croatian")
|
|
LNG(wxLANGUAGE_CZECH, "cs_CZ", LANG_CZECH , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Czech")
|
|
LNG(wxLANGUAGE_DANISH, "da_DK", LANG_DANISH , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Danish")
|
|
LNG(wxLANGUAGE_DUTCH, "nl_NL", LANG_DUTCH , SUBLANG_DUTCH , wxLayout_LeftToRight, "Dutch")
|
|
LNG(wxLANGUAGE_DUTCH_BELGIAN, "nl_BE", LANG_DUTCH , SUBLANG_DUTCH_BELGIAN , wxLayout_LeftToRight, "Dutch (Belgian)")
|
|
LNG(wxLANGUAGE_ENGLISH, "en_GB", LANG_ENGLISH , SUBLANG_ENGLISH_UK , wxLayout_LeftToRight, "English")
|
|
LNG(wxLANGUAGE_ENGLISH_UK, "en_GB", LANG_ENGLISH , SUBLANG_ENGLISH_UK , wxLayout_LeftToRight, "English (U.K.)")
|
|
LNG(wxLANGUAGE_ENGLISH_US, "en_US", LANG_ENGLISH , SUBLANG_ENGLISH_US , wxLayout_LeftToRight, "English (U.S.)")
|
|
LNG(wxLANGUAGE_ENGLISH_AUSTRALIA, "en_AU", LANG_ENGLISH , SUBLANG_ENGLISH_AUS , wxLayout_LeftToRight, "English (Australia)")
|
|
LNG(wxLANGUAGE_ENGLISH_BELIZE, "en_BZ", LANG_ENGLISH , SUBLANG_ENGLISH_BELIZE , wxLayout_LeftToRight, "English (Belize)")
|
|
LNG(wxLANGUAGE_ENGLISH_BOTSWANA, "en_BW", 0 , 0 , wxLayout_LeftToRight, "English (Botswana)")
|
|
LNG(wxLANGUAGE_ENGLISH_CANADA, "en_CA", LANG_ENGLISH , SUBLANG_ENGLISH_CAN , wxLayout_LeftToRight, "English (Canada)")
|
|
LNG(wxLANGUAGE_ENGLISH_CARIBBEAN, "en_CB", LANG_ENGLISH , SUBLANG_ENGLISH_CARIBBEAN , wxLayout_LeftToRight, "English (Caribbean)")
|
|
LNG(wxLANGUAGE_ENGLISH_DENMARK, "en_DK", 0 , 0 , wxLayout_LeftToRight, "English (Denmark)")
|
|
LNG(wxLANGUAGE_ENGLISH_EIRE, "en_IE", LANG_ENGLISH , SUBLANG_ENGLISH_EIRE , wxLayout_LeftToRight, "English (Eire)")
|
|
LNG(wxLANGUAGE_ENGLISH_JAMAICA, "en_JM", LANG_ENGLISH , SUBLANG_ENGLISH_JAMAICA , wxLayout_LeftToRight, "English (Jamaica)")
|
|
LNG(wxLANGUAGE_ENGLISH_NEW_ZEALAND, "en_NZ", LANG_ENGLISH , SUBLANG_ENGLISH_NZ , wxLayout_LeftToRight, "English (New Zealand)")
|
|
LNG(wxLANGUAGE_ENGLISH_PHILIPPINES, "en_PH", LANG_ENGLISH , SUBLANG_ENGLISH_PHILIPPINES , wxLayout_LeftToRight, "English (Philippines)")
|
|
LNG(wxLANGUAGE_ENGLISH_SOUTH_AFRICA, "en_ZA", LANG_ENGLISH , SUBLANG_ENGLISH_SOUTH_AFRICA , wxLayout_LeftToRight, "English (South Africa)")
|
|
LNG(wxLANGUAGE_ENGLISH_TRINIDAD, "en_TT", LANG_ENGLISH , SUBLANG_ENGLISH_TRINIDAD , wxLayout_LeftToRight, "English (Trinidad)")
|
|
LNG(wxLANGUAGE_ENGLISH_ZIMBABWE, "en_ZW", LANG_ENGLISH , SUBLANG_ENGLISH_ZIMBABWE , wxLayout_LeftToRight, "English (Zimbabwe)")
|
|
LNG(wxLANGUAGE_ESPERANTO, "eo" , 0 , 0 , wxLayout_LeftToRight, "Esperanto")
|
|
LNG(wxLANGUAGE_ESTONIAN, "et_EE", LANG_ESTONIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Estonian")
|
|
LNG(wxLANGUAGE_FAEROESE, "fo_FO", LANG_FAEROESE , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Faeroese")
|
|
LNG(wxLANGUAGE_FARSI, "fa_IR", LANG_FARSI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Farsi")
|
|
LNG(wxLANGUAGE_FIJI, "fj" , 0 , 0 , wxLayout_LeftToRight, "Fiji")
|
|
LNG(wxLANGUAGE_FINNISH, "fi_FI", LANG_FINNISH , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Finnish")
|
|
LNG(wxLANGUAGE_FRENCH, "fr_FR", LANG_FRENCH , SUBLANG_FRENCH , wxLayout_LeftToRight, "French")
|
|
LNG(wxLANGUAGE_FRENCH_BELGIAN, "fr_BE", LANG_FRENCH , SUBLANG_FRENCH_BELGIAN , wxLayout_LeftToRight, "French (Belgian)")
|
|
LNG(wxLANGUAGE_FRENCH_CANADIAN, "fr_CA", LANG_FRENCH , SUBLANG_FRENCH_CANADIAN , wxLayout_LeftToRight, "French (Canadian)")
|
|
LNG(wxLANGUAGE_FRENCH_LUXEMBOURG, "fr_LU", LANG_FRENCH , SUBLANG_FRENCH_LUXEMBOURG , wxLayout_LeftToRight, "French (Luxembourg)")
|
|
LNG(wxLANGUAGE_FRENCH_MONACO, "fr_MC", LANG_FRENCH , SUBLANG_FRENCH_MONACO , wxLayout_LeftToRight, "French (Monaco)")
|
|
LNG(wxLANGUAGE_FRENCH_SWISS, "fr_CH", LANG_FRENCH , SUBLANG_FRENCH_SWISS , wxLayout_LeftToRight, "French (Swiss)")
|
|
LNG(wxLANGUAGE_FRISIAN, "fy" , 0 , 0 , wxLayout_LeftToRight, "Frisian")
|
|
LNG(wxLANGUAGE_GALICIAN, "gl_ES", 0 , 0 , wxLayout_LeftToRight, "Galician")
|
|
LNG(wxLANGUAGE_GEORGIAN, "ka_GE", LANG_GEORGIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Georgian")
|
|
LNG(wxLANGUAGE_GERMAN, "de_DE", LANG_GERMAN , SUBLANG_GERMAN , wxLayout_LeftToRight, "German")
|
|
LNG(wxLANGUAGE_GERMAN_AUSTRIAN, "de_AT", LANG_GERMAN , SUBLANG_GERMAN_AUSTRIAN , wxLayout_LeftToRight, "German (Austrian)")
|
|
LNG(wxLANGUAGE_GERMAN_BELGIUM, "de_BE", 0 , 0 , wxLayout_LeftToRight, "German (Belgium)")
|
|
LNG(wxLANGUAGE_GERMAN_LIECHTENSTEIN, "de_LI", LANG_GERMAN , SUBLANG_GERMAN_LIECHTENSTEIN , wxLayout_LeftToRight, "German (Liechtenstein)")
|
|
LNG(wxLANGUAGE_GERMAN_LUXEMBOURG, "de_LU", LANG_GERMAN , SUBLANG_GERMAN_LUXEMBOURG , wxLayout_LeftToRight, "German (Luxembourg)")
|
|
LNG(wxLANGUAGE_GERMAN_SWISS, "de_CH", LANG_GERMAN , SUBLANG_GERMAN_SWISS , wxLayout_LeftToRight, "German (Swiss)")
|
|
LNG(wxLANGUAGE_GREEK, "el_GR", LANG_GREEK , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Greek")
|
|
LNG(wxLANGUAGE_GREENLANDIC, "kl_GL", 0 , 0 , wxLayout_LeftToRight, "Greenlandic")
|
|
LNG(wxLANGUAGE_GUARANI, "gn" , 0 , 0 , wxLayout_LeftToRight, "Guarani")
|
|
LNG(wxLANGUAGE_GUJARATI, "gu" , LANG_GUJARATI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Gujarati")
|
|
LNG(wxLANGUAGE_HAUSA, "ha" , 0 , 0 , wxLayout_LeftToRight, "Hausa")
|
|
LNG(wxLANGUAGE_HEBREW, "he_IL", LANG_HEBREW , SUBLANG_DEFAULT , wxLayout_RightToLeft, "Hebrew")
|
|
LNG(wxLANGUAGE_HINDI, "hi_IN", LANG_HINDI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Hindi")
|
|
LNG(wxLANGUAGE_HUNGARIAN, "hu_HU", LANG_HUNGARIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Hungarian")
|
|
LNG(wxLANGUAGE_ICELANDIC, "is_IS", LANG_ICELANDIC , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Icelandic")
|
|
LNG(wxLANGUAGE_INDONESIAN, "id_ID", LANG_INDONESIAN, SUBLANG_DEFAULT , wxLayout_LeftToRight, "Indonesian")
|
|
LNG(wxLANGUAGE_INTERLINGUA, "ia" , 0 , 0 , wxLayout_LeftToRight, "Interlingua")
|
|
LNG(wxLANGUAGE_INTERLINGUE, "ie" , 0 , 0 , wxLayout_LeftToRight, "Interlingue")
|
|
LNG(wxLANGUAGE_INUKTITUT, "iu" , 0 , 0 , wxLayout_LeftToRight, "Inuktitut")
|
|
LNG(wxLANGUAGE_INUPIAK, "ik" , 0 , 0 , wxLayout_LeftToRight, "Inupiak")
|
|
LNG(wxLANGUAGE_IRISH, "ga_IE", 0 , 0 , wxLayout_LeftToRight, "Irish")
|
|
LNG(wxLANGUAGE_ITALIAN, "it_IT", LANG_ITALIAN , SUBLANG_ITALIAN , wxLayout_LeftToRight, "Italian")
|
|
LNG(wxLANGUAGE_ITALIAN_SWISS, "it_CH", LANG_ITALIAN , SUBLANG_ITALIAN_SWISS , wxLayout_LeftToRight, "Italian (Swiss)")
|
|
LNG(wxLANGUAGE_JAPANESE, "ja_JP", LANG_JAPANESE , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Japanese")
|
|
LNG(wxLANGUAGE_JAVANESE, "jw" , 0 , 0 , wxLayout_LeftToRight, "Javanese")
|
|
LNG(wxLANGUAGE_KANNADA, "kn" , LANG_KANNADA , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Kannada")
|
|
LNG(wxLANGUAGE_KASHMIRI, "ks" , LANG_KASHMIRI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Kashmiri")
|
|
LNG(wxLANGUAGE_KASHMIRI_INDIA, "ks_IN", LANG_KASHMIRI , SUBLANG_KASHMIRI_INDIA , wxLayout_LeftToRight, "Kashmiri (India)")
|
|
LNG(wxLANGUAGE_KAZAKH, "kk" , LANG_KAZAK , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Kazakh")
|
|
LNG(wxLANGUAGE_KERNEWEK, "kw_GB", 0 , 0 , wxLayout_LeftToRight, "Kernewek")
|
|
LNG(wxLANGUAGE_KINYARWANDA, "rw" , 0 , 0 , wxLayout_LeftToRight, "Kinyarwanda")
|
|
LNG(wxLANGUAGE_KIRGHIZ, "ky" , 0 , 0 , wxLayout_LeftToRight, "Kirghiz")
|
|
LNG(wxLANGUAGE_KIRUNDI, "rn" , 0 , 0 , wxLayout_LeftToRight, "Kirundi")
|
|
LNG(wxLANGUAGE_KONKANI, "" , LANG_KONKANI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Konkani")
|
|
LNG(wxLANGUAGE_KOREAN, "ko_KR", LANG_KOREAN , SUBLANG_KOREAN , wxLayout_LeftToRight, "Korean")
|
|
LNG(wxLANGUAGE_KURDISH, "ku_TR", 0 , 0 , wxLayout_LeftToRight, "Kurdish")
|
|
LNG(wxLANGUAGE_LAOTHIAN, "lo" , 0 , 0 , wxLayout_LeftToRight, "Laothian")
|
|
LNG(wxLANGUAGE_LATIN, "la" , 0 , 0 , wxLayout_LeftToRight, "Latin")
|
|
LNG(wxLANGUAGE_LATVIAN, "lv_LV", LANG_LATVIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Latvian")
|
|
LNG(wxLANGUAGE_LINGALA, "ln" , 0 , 0 , wxLayout_LeftToRight, "Lingala")
|
|
LNG(wxLANGUAGE_LITHUANIAN, "lt_LT", LANG_LITHUANIAN, SUBLANG_LITHUANIAN , wxLayout_LeftToRight, "Lithuanian")
|
|
LNG(wxLANGUAGE_MACEDONIAN, "mk_MK", LANG_MACEDONIAN, SUBLANG_DEFAULT , wxLayout_LeftToRight, "Macedonian")
|
|
LNG(wxLANGUAGE_MALAGASY, "mg" , 0 , 0 , wxLayout_LeftToRight, "Malagasy")
|
|
LNG(wxLANGUAGE_MALAY, "ms_MY", LANG_MALAY , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Malay")
|
|
LNG(wxLANGUAGE_MALAYALAM, "ml" , LANG_MALAYALAM , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Malayalam")
|
|
LNG(wxLANGUAGE_MALAY_BRUNEI_DARUSSALAM, "ms_BN", LANG_MALAY , SUBLANG_MALAY_BRUNEI_DARUSSALAM , wxLayout_LeftToRight, "Malay (Brunei Darussalam)")
|
|
LNG(wxLANGUAGE_MALAY_MALAYSIA, "ms_MY", LANG_MALAY , SUBLANG_MALAY_MALAYSIA , wxLayout_LeftToRight, "Malay (Malaysia)")
|
|
LNG(wxLANGUAGE_MALTESE, "mt_MT", 0 , 0 , wxLayout_LeftToRight, "Maltese")
|
|
LNG(wxLANGUAGE_MANIPURI, "" , LANG_MANIPURI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Manipuri")
|
|
LNG(wxLANGUAGE_MAORI, "mi" , 0 , 0 , wxLayout_LeftToRight, "Maori")
|
|
LNG(wxLANGUAGE_MARATHI, "mr_IN", LANG_MARATHI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Marathi")
|
|
LNG(wxLANGUAGE_MOLDAVIAN, "mo" , 0 , 0 , wxLayout_LeftToRight, "Moldavian")
|
|
LNG(wxLANGUAGE_MONGOLIAN, "mn" , 0 , 0 , wxLayout_LeftToRight, "Mongolian")
|
|
LNG(wxLANGUAGE_NAURU, "na" , 0 , 0 , wxLayout_LeftToRight, "Nauru")
|
|
LNG(wxLANGUAGE_NEPALI, "ne_NP", LANG_NEPALI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Nepali")
|
|
LNG(wxLANGUAGE_NEPALI_INDIA, "ne_IN", LANG_NEPALI , SUBLANG_NEPALI_INDIA , wxLayout_LeftToRight, "Nepali (India)")
|
|
LNG(wxLANGUAGE_NORWEGIAN_BOKMAL, "nb_NO", LANG_NORWEGIAN , SUBLANG_NORWEGIAN_BOKMAL , wxLayout_LeftToRight, "Norwegian (Bokmal)")
|
|
LNG(wxLANGUAGE_NORWEGIAN_NYNORSK, "nn_NO", LANG_NORWEGIAN , SUBLANG_NORWEGIAN_NYNORSK , wxLayout_LeftToRight, "Norwegian (Nynorsk)")
|
|
LNG(wxLANGUAGE_OCCITAN, "oc" , 0 , 0 , wxLayout_LeftToRight, "Occitan")
|
|
LNG(wxLANGUAGE_ORIYA, "or" , LANG_ORIYA , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Oriya")
|
|
LNG(wxLANGUAGE_OROMO, "om" , 0 , 0 , wxLayout_LeftToRight, "(Afan) Oromo")
|
|
LNG(wxLANGUAGE_PASHTO, "ps" , 0 , 0 , wxLayout_LeftToRight, "Pashto, Pushto")
|
|
LNG(wxLANGUAGE_POLISH, "pl_PL", LANG_POLISH , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Polish")
|
|
LNG(wxLANGUAGE_PORTUGUESE, "pt_PT", LANG_PORTUGUESE, SUBLANG_PORTUGUESE , wxLayout_LeftToRight, "Portuguese")
|
|
LNG(wxLANGUAGE_PORTUGUESE_BRAZILIAN, "pt_BR", LANG_PORTUGUESE, SUBLANG_PORTUGUESE_BRAZILIAN , wxLayout_LeftToRight, "Portuguese (Brazilian)")
|
|
LNG(wxLANGUAGE_PUNJABI, "pa" , LANG_PUNJABI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Punjabi")
|
|
LNG(wxLANGUAGE_QUECHUA, "qu" , 0 , 0 , wxLayout_LeftToRight, "Quechua")
|
|
LNG(wxLANGUAGE_RHAETO_ROMANCE, "rm" , 0 , 0 , wxLayout_LeftToRight, "Rhaeto-Romance")
|
|
LNG(wxLANGUAGE_ROMANIAN, "ro_RO", LANG_ROMANIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Romanian")
|
|
LNG(wxLANGUAGE_RUSSIAN, "ru_RU", LANG_RUSSIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Russian")
|
|
LNG(wxLANGUAGE_RUSSIAN_UKRAINE, "ru_UA", 0 , 0 , wxLayout_LeftToRight, "Russian (Ukraine)")
|
|
LNG(wxLANGUAGE_SAMI, "se_NO", LANG_SAMI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Northern Sami")
|
|
LNG(wxLANGUAGE_SAMOAN, "sm" , 0 , 0 , wxLayout_LeftToRight, "Samoan")
|
|
LNG(wxLANGUAGE_SANGHO, "sg" , 0 , 0 , wxLayout_LeftToRight, "Sangho")
|
|
LNG(wxLANGUAGE_SANSKRIT, "sa" , LANG_SANSKRIT , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Sanskrit")
|
|
LNG(wxLANGUAGE_SCOTS_GAELIC, "gd" , 0 , 0 , wxLayout_LeftToRight, "Scots Gaelic")
|
|
LNG(wxLANGUAGE_SERBIAN, "sr_RS", LANG_SERBIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Serbian")
|
|
LNG(wxLANGUAGE_SERBIAN_CYRILLIC, "sr_RS", LANG_SERBIAN , SUBLANG_SERBIAN_CYRILLIC , wxLayout_LeftToRight, "Serbian (Cyrillic)")
|
|
LNG(wxLANGUAGE_SERBIAN_LATIN, "sr_RS@latin", LANG_SERBIAN , SUBLANG_SERBIAN_LATIN , wxLayout_LeftToRight, "Serbian (Latin)")
|
|
LNG(wxLANGUAGE_SERBIAN_CYRILLIC, "sr_YU", LANG_SERBIAN , SUBLANG_SERBIAN_CYRILLIC , wxLayout_LeftToRight, "Serbian (Cyrillic)")
|
|
LNG(wxLANGUAGE_SERBIAN_LATIN, "sr_YU@latin", LANG_SERBIAN , SUBLANG_SERBIAN_LATIN , wxLayout_LeftToRight, "Serbian (Latin)")
|
|
LNG(wxLANGUAGE_SERBO_CROATIAN, "sh" , 0 , 0 , wxLayout_LeftToRight, "Serbo-Croatian")
|
|
LNG(wxLANGUAGE_SESOTHO, "st" , 0 , 0 , wxLayout_LeftToRight, "Sesotho")
|
|
LNG(wxLANGUAGE_SETSWANA, "tn" , 0 , 0 , wxLayout_LeftToRight, "Setswana")
|
|
LNG(wxLANGUAGE_SHONA, "sn" , 0 , 0 , wxLayout_LeftToRight, "Shona")
|
|
LNG(wxLANGUAGE_SINDHI, "sd" , LANG_SINDHI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Sindhi")
|
|
LNG(wxLANGUAGE_SINHALESE, "si" , 0 , 0 , wxLayout_LeftToRight, "Sinhalese")
|
|
LNG(wxLANGUAGE_SISWATI, "ss" , 0 , 0 , wxLayout_LeftToRight, "Siswati")
|
|
LNG(wxLANGUAGE_SLOVAK, "sk_SK", LANG_SLOVAK , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Slovak")
|
|
LNG(wxLANGUAGE_SLOVENIAN, "sl_SI", LANG_SLOVENIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Slovenian")
|
|
LNG(wxLANGUAGE_SOMALI, "so" , 0 , 0 , wxLayout_LeftToRight, "Somali")
|
|
LNG(wxLANGUAGE_SPANISH, "es_ES", LANG_SPANISH , SUBLANG_SPANISH , wxLayout_LeftToRight, "Spanish")
|
|
LNG(wxLANGUAGE_SPANISH_ARGENTINA, "es_AR", LANG_SPANISH , SUBLANG_SPANISH_ARGENTINA , wxLayout_LeftToRight, "Spanish (Argentina)")
|
|
LNG(wxLANGUAGE_SPANISH_BOLIVIA, "es_BO", LANG_SPANISH , SUBLANG_SPANISH_BOLIVIA , wxLayout_LeftToRight, "Spanish (Bolivia)")
|
|
LNG(wxLANGUAGE_SPANISH_CHILE, "es_CL", LANG_SPANISH , SUBLANG_SPANISH_CHILE , wxLayout_LeftToRight, "Spanish (Chile)")
|
|
LNG(wxLANGUAGE_SPANISH_COLOMBIA, "es_CO", LANG_SPANISH , SUBLANG_SPANISH_COLOMBIA , wxLayout_LeftToRight, "Spanish (Colombia)")
|
|
LNG(wxLANGUAGE_SPANISH_COSTA_RICA, "es_CR", LANG_SPANISH , SUBLANG_SPANISH_COSTA_RICA , wxLayout_LeftToRight, "Spanish (Costa Rica)")
|
|
LNG(wxLANGUAGE_SPANISH_DOMINICAN_REPUBLIC, "es_DO", LANG_SPANISH , SUBLANG_SPANISH_DOMINICAN_REPUBLIC, wxLayout_LeftToRight, "Spanish (Dominican republic)")
|
|
LNG(wxLANGUAGE_SPANISH_ECUADOR, "es_EC", LANG_SPANISH , SUBLANG_SPANISH_ECUADOR , wxLayout_LeftToRight, "Spanish (Ecuador)")
|
|
LNG(wxLANGUAGE_SPANISH_EL_SALVADOR, "es_SV", LANG_SPANISH , SUBLANG_SPANISH_EL_SALVADOR , wxLayout_LeftToRight, "Spanish (El Salvador)")
|
|
LNG(wxLANGUAGE_SPANISH_GUATEMALA, "es_GT", LANG_SPANISH , SUBLANG_SPANISH_GUATEMALA , wxLayout_LeftToRight, "Spanish (Guatemala)")
|
|
LNG(wxLANGUAGE_SPANISH_HONDURAS, "es_HN", LANG_SPANISH , SUBLANG_SPANISH_HONDURAS , wxLayout_LeftToRight, "Spanish (Honduras)")
|
|
LNG(wxLANGUAGE_SPANISH_MEXICAN, "es_MX", LANG_SPANISH , SUBLANG_SPANISH_MEXICAN , wxLayout_LeftToRight, "Spanish (Mexican)")
|
|
LNG(wxLANGUAGE_SPANISH_MODERN, "es_ES", LANG_SPANISH , SUBLANG_SPANISH_MODERN , wxLayout_LeftToRight, "Spanish (Modern)")
|
|
LNG(wxLANGUAGE_SPANISH_NICARAGUA, "es_NI", LANG_SPANISH , SUBLANG_SPANISH_NICARAGUA , wxLayout_LeftToRight, "Spanish (Nicaragua)")
|
|
LNG(wxLANGUAGE_SPANISH_PANAMA, "es_PA", LANG_SPANISH , SUBLANG_SPANISH_PANAMA , wxLayout_LeftToRight, "Spanish (Panama)")
|
|
LNG(wxLANGUAGE_SPANISH_PARAGUAY, "es_PY", LANG_SPANISH , SUBLANG_SPANISH_PARAGUAY , wxLayout_LeftToRight, "Spanish (Paraguay)")
|
|
LNG(wxLANGUAGE_SPANISH_PERU, "es_PE", LANG_SPANISH , SUBLANG_SPANISH_PERU , wxLayout_LeftToRight, "Spanish (Peru)")
|
|
LNG(wxLANGUAGE_SPANISH_PUERTO_RICO, "es_PR", LANG_SPANISH , SUBLANG_SPANISH_PUERTO_RICO , wxLayout_LeftToRight, "Spanish (Puerto Rico)")
|
|
LNG(wxLANGUAGE_SPANISH_URUGUAY, "es_UY", LANG_SPANISH , SUBLANG_SPANISH_URUGUAY , wxLayout_LeftToRight, "Spanish (Uruguay)")
|
|
LNG(wxLANGUAGE_SPANISH_US, "es_US", 0 , 0 , wxLayout_LeftToRight, "Spanish (U.S.)")
|
|
LNG(wxLANGUAGE_SPANISH_VENEZUELA, "es_VE", LANG_SPANISH , SUBLANG_SPANISH_VENEZUELA , wxLayout_LeftToRight, "Spanish (Venezuela)")
|
|
LNG(wxLANGUAGE_SUNDANESE, "su" , 0 , 0 , wxLayout_LeftToRight, "Sundanese")
|
|
LNG(wxLANGUAGE_SWAHILI, "sw_KE", LANG_SWAHILI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Swahili")
|
|
LNG(wxLANGUAGE_SWEDISH, "sv_SE", LANG_SWEDISH , SUBLANG_SWEDISH , wxLayout_LeftToRight, "Swedish")
|
|
LNG(wxLANGUAGE_SWEDISH_FINLAND, "sv_FI", LANG_SWEDISH , SUBLANG_SWEDISH_FINLAND , wxLayout_LeftToRight, "Swedish (Finland)")
|
|
LNG(wxLANGUAGE_TAGALOG, "tl_PH", 0 , 0 , wxLayout_LeftToRight, "Tagalog")
|
|
LNG(wxLANGUAGE_TAJIK, "tg" , 0 , 0 , wxLayout_LeftToRight, "Tajik")
|
|
LNG(wxLANGUAGE_TAMIL, "ta" , LANG_TAMIL , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Tamil")
|
|
LNG(wxLANGUAGE_TATAR, "tt" , LANG_TATAR , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Tatar")
|
|
LNG(wxLANGUAGE_TELUGU, "te" , LANG_TELUGU , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Telugu")
|
|
LNG(wxLANGUAGE_THAI, "th_TH", LANG_THAI , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Thai")
|
|
LNG(wxLANGUAGE_TIBETAN, "bo" , 0 , 0 , wxLayout_LeftToRight, "Tibetan")
|
|
LNG(wxLANGUAGE_TIGRINYA, "ti" , 0 , 0 , wxLayout_LeftToRight, "Tigrinya")
|
|
LNG(wxLANGUAGE_TONGA, "to" , 0 , 0 , wxLayout_LeftToRight, "Tonga")
|
|
LNG(wxLANGUAGE_TSONGA, "ts" , 0 , 0 , wxLayout_LeftToRight, "Tsonga")
|
|
LNG(wxLANGUAGE_TURKISH, "tr_TR", LANG_TURKISH , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Turkish")
|
|
LNG(wxLANGUAGE_TURKMEN, "tk" , 0 , 0 , wxLayout_LeftToRight, "Turkmen")
|
|
LNG(wxLANGUAGE_TWI, "tw" , 0 , 0 , wxLayout_LeftToRight, "Twi")
|
|
LNG(wxLANGUAGE_UIGHUR, "ug" , 0 , 0 , wxLayout_LeftToRight, "Uighur")
|
|
LNG(wxLANGUAGE_UKRAINIAN, "uk_UA", LANG_UKRAINIAN , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Ukrainian")
|
|
LNG(wxLANGUAGE_URDU, "ur" , LANG_URDU , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Urdu")
|
|
LNG(wxLANGUAGE_URDU_INDIA, "ur_IN", LANG_URDU , SUBLANG_URDU_INDIA , wxLayout_LeftToRight, "Urdu (India)")
|
|
LNG(wxLANGUAGE_URDU_PAKISTAN, "ur_PK", LANG_URDU , SUBLANG_URDU_PAKISTAN , wxLayout_LeftToRight, "Urdu (Pakistan)")
|
|
LNG(wxLANGUAGE_UZBEK, "uz" , LANG_UZBEK , SUBLANG_DEFAULT , wxLayout_LeftToRight, "Uzbek")
|
|
LNG(wxLANGUAGE_UZBEK_CYRILLIC, "uz" , LANG_UZBEK , SUBLANG_UZBEK_CYRILLIC , wxLayout_LeftToRight, "Uzbek (Cyrillic)")
|
|
LNG(wxLANGUAGE_UZBEK_LATIN, "uz" , LANG_UZBEK , SUBLANG_UZBEK_LATIN , wxLayout_LeftToRight, "Uzbek (Latin)")
|
|
LNG(wxLANGUAGE_VALENCIAN, "ca_ES@valencia", 0 , 0 , wxLayout_LeftToRight, "Valencian (Souternhern Catalan)")
|
|
LNG(wxLANGUAGE_VIETNAMESE, "vi_VN", LANG_VIETNAMESE, SUBLANG_DEFAULT , wxLayout_LeftToRight, "Vietnamese")
|
|
LNG(wxLANGUAGE_VOLAPUK, "vo" , 0 , 0 , wxLayout_LeftToRight, "Volapuk")
|
|
LNG(wxLANGUAGE_WELSH, "cy" , 0 , 0 , wxLayout_LeftToRight, "Welsh")
|
|
LNG(wxLANGUAGE_WOLOF, "wo" , 0 , 0 , wxLayout_LeftToRight, "Wolof")
|
|
LNG(wxLANGUAGE_XHOSA, "xh" , 0 , 0 , wxLayout_LeftToRight, "Xhosa")
|
|
LNG(wxLANGUAGE_YIDDISH, "yi" , 0 , 0 , wxLayout_LeftToRight, "Yiddish")
|
|
LNG(wxLANGUAGE_YORUBA, "yo" , 0 , 0 , wxLayout_LeftToRight, "Yoruba")
|
|
LNG(wxLANGUAGE_ZHUANG, "za" , 0 , 0 , wxLayout_LeftToRight, "Zhuang")
|
|
LNG(wxLANGUAGE_ZULU, "zu" , 0 , 0 , wxLayout_LeftToRight, "Zulu")
|
|
}
|
|
#undef LNG
|
|
|
|
// --- --- --- generated code ends here --- --- ---
|
|
|
|
#endif // wxUSE_INTL
|