#pragma hdrstop
#endif
+#ifdef __WXMSW__
+ #include "wx/msw/private.h"
+#endif
+
+#include <errno.h>
#include <ctype.h>
#include <string.h>
#include <stdlib.h>
#include <clib.h>
#endif
+#ifdef HAVE_ICONV_H
+ #include <iconv.h>
+#endif
+#ifdef HAVE_LANGINFO_H
+ #include <langinfo.h>
+#endif
+
+#ifdef __WXMSW__
+ #include <windows.h>
+#endif
+
#include "wx/debug.h"
#include "wx/strconv.h"
+#if defined(WORDS_BIGENDIAN) || defined(__STDC_ISO_10646__)
+#define BSWAP_UCS4(str, len)
+#define BSWAP_UCS2(str, len)
+#else
+#define BSWAP_UCS4(str, len) { unsigned _c; for (_c=0; _c<len; _c++) str[_c]=wxUINT32_SWAP_ALWAYS(str[_c]); }
+#define BSWAP_UCS2(str, len) { unsigned _c; for (_c=0; _c<len; _c++) str[_c]=wxUINT16_SWAP_ALWAYS(str[_c]); }
+#define WC_NEED_BSWAP
+#endif
+#define BSWAP_UTF32(str, len) BSWAP_UCS4(str, len)
+#define BSWAP_UTF16(str, len) BSWAP_UCS2(str, len)
+
+#if SIZEOF_WCHAR_T == 4
+#define WC_NAME "UCS4"
+#define WC_BSWAP BSWAP_UCS4
+#elif SIZEOF_WCHAR_T == 2
+#define WC_NAME "UTF16"
+#define WC_BSWAP BSWAP_UTF16
+#define WC_UTF16
+#endif
+
// ----------------------------------------------------------------------------
// globals
// ----------------------------------------------------------------------------
#if wxUSE_WCHAR_T
+#ifdef WC_UTF16
+
+static size_t encode_utf16(wxUint32 input,wxUint16*output)
+{
+ if (input<=0xffff) {
+ if (output) *output++ = input;
+ return 1;
+ } else
+ if (input>=0x110000) {
+ return (size_t)-1;
+ } else {
+ if (output) {
+ *output++ = (input >> 10)+0xd7c0;
+ *output++ = (input&0x3ff)+0xdc00;
+ }
+ return 2;
+ }
+}
+
+static size_t decode_utf16(wxUint16*input,wxUint32&output)
+{
+ if ((*input<0xd800) || (*input>0xdfff)) {
+ output = *input;
+ return 1;
+ } else
+ if ((input[1]<0xdc00) || (input[1]>=0xdfff)) {
+ output = *input;
+ return (size_t)-1;
+ } else {
+ output = ((input[0] - 0xd7c0) << 10) + (input[1] - 0xdc00);
+ return 2;
+ }
+}
+
+#endif // WC_UTF16
+
// ----------------------------------------------------------------------------
// wxMBConv
// ----------------------------------------------------------------------------
WXDLLEXPORT_DATA(wxMBConvUTF8) wxConvUTF8;
-static unsigned long utf8_max[]={0x7f,0x7ff,0xffff,0x1fffff,0x3ffffff,0x7fffffff,0xffffffff};
+static wxUint32 utf8_max[]={0x7f,0x7ff,0xffff,0x1fffff,0x3ffffff,0x7fffffff,0xffffffff};
size_t wxMBConvUTF8::MB2WC(wchar_t *buf, const char *psz, size_t n) const
{
return (size_t)-1;
} else {
unsigned ocnt=cnt-1;
- unsigned long res=cc&(0x3f>>cnt);
+ wxUint32 res=cc&(0x3f>>cnt);
while (cnt--) {
cc = *psz++;
if ((cc&0xC0)!=0x80) {
// illegal UTF-8 encoding
return (size_t)-1;
}
+#ifdef WC_UTF16
+ size_t pa = encode_utf16(res, buf);
+ if (pa == (size_t)-1)
+ return (size_t)-1;
+ if (buf) buf+=pa;
+ len+=pa;
+#else
if (buf) *buf++=res;
len++;
+#endif
}
}
}
size_t len = 0;
while (*psz && ((!buf) || (len<n))) {
- unsigned long cc=(*psz++)&0x7fffffff;
+ wxUint32 cc;
+#ifdef WC_UTF16
+ size_t pa = decode_utf16(psz,cc);
+ psz += (pa == (size_t)-1) ? 1 : pa;
+#else
+ cc=(*psz++)&0x7fffffff;
+#endif
unsigned cnt;
for (cnt=0; cc>utf8_max[cnt]; cnt++);
if (!cnt) {
#include "wx/encconv.h"
#include "wx/fontmap.h"
+// TODO: add some tables here
+// - perhaps common encodings to common codepages (for Win32)
+// - perhaps common encodings to objects ("UTF8" -> wxConvUTF8)
+// - move wxEncodingConverter meat in here
+
+#ifdef __WIN32__
+#include "wx/msw/registry.h"
+// this should work if M$ Internet Exploiter is installed
+static long CharsetToCodepage(const wxChar *name)
+{
+ if (!name) return GetACP();
+ long CP=-1;
+ wxString cn(name);
+ do {
+ wxString path(wxT("MIME\\Database\\Charset\\"));
+ path += cn;
+ wxRegKey key(wxRegKey::HKCR,path);
+
+ /* two cases: either there's an AliasForCharset string,
+ * or there are Codepage and InternetEncoding dwords.
+ * The InternetEncoding gives us the actual encoding,
+ * the Codepage just says which Windows character set to
+ * use when displaying the data.
+ */
+ if (key.QueryValue(wxT("InternetEncoding"),&CP)) break;
+ // no encoding, see if it's an alias
+ if (!key.QueryValue(wxT("AliasForCharset"),cn)) break;
+ } while (1);
+ return CP;
+}
+#endif
+
class wxCharacterSet
{
+public:
+ const wxChar*cname;
+ wxCharacterSet(const wxChar*name) : cname(name) {}
+ virtual ~wxCharacterSet() {}
+ virtual size_t MB2WC(wchar_t*buf, const char*psz, size_t n) { return (size_t)-1; }
+ virtual size_t WC2MB(char*buf, const wchar_t*psz, size_t n) { return (size_t)-1; }
+ virtual bool usable() { return FALSE; }
+};
+
+class ID_CharSet : public wxCharacterSet
+{
+public:
+ wxMBConv*work;
+ ID_CharSet(const wxChar*name,wxMBConv*cnv) : wxCharacterSet(name), work(cnv) {}
+ size_t MB2WC(wchar_t*buf, const char*psz, size_t n)
+ { return work ? work->MB2WC(buf,psz,n) : (size_t)-1; }
+ size_t WC2MB(char*buf, const wchar_t*psz, size_t n)
+ { return work ? work->WC2MB(buf,psz,n) : (size_t)-1; }
+ bool usable() { return work!=NULL; }
+};
+
+#ifdef HAVE_ICONV_H
+class IC_CharSet : public wxCharacterSet
+{
+public:
+ iconv_t m2w, w2m;
+ IC_CharSet(const wxChar*name) : wxCharacterSet(name), m2w((iconv_t)-1), w2m((iconv_t)-1) {}
+ ~IC_CharSet() {
+ if (m2w!=(iconv_t)-1) iconv_close(m2w);
+ if (w2m!=(iconv_t)-1) iconv_close(w2m);
+ }
+ void LoadM2W() { if (m2w==(iconv_t)-1) m2w=iconv_open(WC_NAME,wxConvLibc.cWX2MB(cname)); }
+ void LoadW2M() { if (w2m==(iconv_t)-1) w2m=iconv_open(wxConvLibc.cWX2MB(cname),WC_NAME); }
+ size_t MB2WC(wchar_t*buf, const char*psz, size_t n) {
+ LoadM2W();
+ size_t inbuf = strlen(psz);
+ size_t outbuf = n*SIZEOF_WCHAR_T;
+ size_t res, cres;
+ fprintf(stderr,"IC Convert to WC using %s\n",(const char*)wxConvLibc.cWX2MB(cname));
+ if (buf) {
+ // have destination buffer, convert there
+ cres = iconv(m2w,&psz,&inbuf,(char**)&buf,&outbuf);
+ res = n-(outbuf/SIZEOF_WCHAR_T);
+ // convert to native endianness
+ WC_BSWAP(buf, res)
+ } else {
+ // no destination buffer... convert using temp buffer
+ // to calculate destination buffer requirement
+ wchar_t tbuf[8];
+ res = 0;
+ do {
+ buf = tbuf; outbuf = 8*SIZEOF_WCHAR_T;
+ cres = iconv(m2w,&psz,&inbuf,(char**)&buf,&outbuf);
+ res += 8-(outbuf/SIZEOF_WCHAR_T);
+ } while ((cres==(size_t)-1) && (errno==E2BIG));
+ }
+ if (cres==(size_t)-1) return (size_t)-1;
+ return res;
+ }
+ size_t WC2MB(char*buf, const wchar_t*psz, size_t n) {
+ LoadW2M();
+#if defined(__BORLANDC__) && (__BORLANDC__ > 0x530)
+ size_t inbuf = std::wcslen(psz);
+#else
+ size_t inbuf = ::wcslen(psz);
+#endif
+ size_t outbuf = n;
+ size_t res, cres;
+ fprintf(stderr,"IC Convert from WC using %s\n",(const char*)wxConvLibc.cWX2MB(cname));
+#ifdef WC_NEED_BSWAP
+ // need to copy to temp buffer to switch endianness
+ // this absolutely doesn't rock!
+ // (no, doing WC_BSWAP twice on the original buffer won't help, as it
+ // could be in read-only memory, or be accessed in some other thread)
+ wchar_t*tmpbuf=(wchar_t*)malloc((inbuf+1)*SIZEOF_WCHAR_T);
+ memcpy(tmpbuf,psz,(inbuf+1)*SIZEOF_WCHAR_T);
+ WC_BSWAP(tmpbuf, inbuf)
+ psz=tmpbuf;
+#endif
+ if (buf) {
+ // have destination buffer, convert there
+ cres = iconv(w2m,(const char**)&psz,&inbuf,&buf,&outbuf);
+ res = n-outbuf;
+ } else {
+ // no destination buffer... convert using temp buffer
+ // to calculate destination buffer requirement
+ char tbuf[16];
+ res = 0;
+ do {
+ buf = tbuf; outbuf = 16;
+ cres = iconv(w2m,(const char**)&psz,&inbuf,&buf,&outbuf);
+ res += 16 - outbuf;
+ } while ((cres==(size_t)-1) && (errno==E2BIG));
+ }
+#ifdef WC_NEED_BSWAP
+ free(tmpbuf);
+#endif
+ if (cres==(size_t)-1) return (size_t)-1;
+ return res;
+ }
+ bool usable() { return TRUE; }
+};
+#endif
+
+#ifdef __WIN32__
+class CP_CharSet : public wxCharacterSet
+{
+public:
+ long CodePage;
+ CP_CharSet(const wxChar*name) : wxCharacterSet(name), CodePage(CharsetToCodepage(name)) {}
+ size_t MB2WC(wchar_t*buf, const char*psz, size_t n) {
+ size_t len = MultiByteToWideChar(CodePage,0,psz,-1,buf,buf?n:0);
+ return len?len:(size_t)-1;
+ }
+ size_t WC2MB(char*buf, const wchar_t*psz, size_t n) {
+ size_t len = WideCharToMultiByte(CodePage,0,psz,-1,buf,buf?n:0,NULL,NULL);
+ return len?len:(size_t)-1;
+ }
+ bool usable() { return CodePage!=-1; }
+};
+#endif
+
+class EC_CharSet : public wxCharacterSet
+{
public:
// temporarily just use wxEncodingConverter stuff,
// so that it works while a better implementation is built
wxFontEncoding enc;
wxEncodingConverter m2w, w2m;
- wxCharacterSet(wxFontEncoding e) : enc(e)
+ EC_CharSet(const wxChar*name) : wxCharacterSet(name), enc(wxFONTENCODING_SYSTEM)
{
+ if (name) enc = wxTheFontMapper->CharsetToEncoding(name, FALSE);
m2w.Init(enc, wxFONTENCODING_UNICODE);
w2m.Init(wxFONTENCODING_UNICODE, enc);
}
+ size_t MB2WC(wchar_t*buf, const char*psz, size_t n) {
+ size_t inbuf = strlen(psz);
+ fprintf(stderr,"EC Convert to WC using %d\n",enc);
+ if (buf) m2w.Convert(psz,buf);
+ return inbuf;
+ }
+ size_t WC2MB(char*buf, const wchar_t*psz, size_t n) {
+#if defined(__BORLANDC__) && (__BORLANDC__ > 0x530)
+ size_t inbuf = std::wcslen(psz);
+#else
+ size_t inbuf = ::wcslen(psz);
+#endif
+ fprintf(stderr,"EC Convert from WC using %d\n",enc);
+ if (buf) w2m.Convert(psz,buf);
+ return inbuf;
+ }
+ bool usable() { return (enc!=wxFONTENCODING_SYSTEM) && (enc!=wxFONTENCODING_DEFAULT); }
};
static wxCharacterSet *wxGetCharacterSet(const wxChar *name)
{
- wxFontEncoding enc = name ? wxTheFontMapper->CharsetToEncoding(name, FALSE)
- : wxFONTENCODING_SYSTEM;
- wxCharacterSet *cset = (enc != wxFONTENCODING_SYSTEM) ? new wxCharacterSet(enc)
- : (wxCharacterSet *)NULL;
- return cset;
+ wxCharacterSet *cset = NULL;
+ if (name) {
+ if (!wxStricmp(name, wxT("UTF8")) || !wxStricmp(name, wxT("UTF-8"))) {
+ cset = new ID_CharSet(name, &wxConvUTF8);
+ } else {
+#ifdef HAVE_ICONV_H
+ cset = new IC_CharSet(name); // may not take NULL
+#endif
+ }
+ }
+ if (cset && cset->usable()) return cset;
+ if (cset) delete cset;
+#ifdef __WIN32__
+ cset = new CP_CharSet(name); // may take NULL
+ if (cset->usable()) return cset;
+#endif
+ if (cset) delete cset;
+ cset = new EC_CharSet(name);
+ if (cset->usable()) return cset;
+ delete cset;
+ return NULL;
}
wxCSConv::wxCSConv(const wxChar *charset)
if (m_deferred) {
if (!m_name) {
#ifdef __UNIX__
- wxChar *lang = wxGetenv(wxT("LC_ALL"));
- if (!lang) lang = wxGetenv(wxT("LANG"));
- wxChar *dot = lang ? wxStrchr(lang, wxT('.')) : (wxChar *)NULL;
- if (dot) SetName(dot+1);
+#if defined(HAVE_LANGINFO_H) && defined(CODESET)
+ // GNU libc provides current character set this way
+ char*alang = nl_langinfo(CODESET);
+ if (alang) SetName(wxConvLibc.cMB2WX(alang));
+ else
+#endif
+ // 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)
+ {
+ wxChar *lang = wxGetenv(wxT("LC_ALL"));
+ if (!lang) lang = wxGetenv(wxT("LC_CTYPE"));
+ if (!lang) lang = wxGetenv(wxT("LANG"));
+ wxChar *dot = lang ? wxStrchr(lang, wxT('.')) : (wxChar *)NULL;
+ if (dot) SetName(dot+1);
+ }
#endif
}
m_cset = wxGetCharacterSet(m_name);
size_t wxCSConv::MB2WC(wchar_t *buf, const char *psz, size_t n) const
{
((wxCSConv *)this)->LoadNow(); // discard constness
+ if (m_cset)
+ return m_cset->MB2WC(buf, psz, n);
+
+ // latin-1 (direct)
+ size_t len=strlen(psz);
if (buf) {
- if (m_cset) {
- m_cset->m2w.Convert(psz, buf);
- } else {
- // latin-1 (direct)
- for (size_t c=0; c<n; c++)
- buf[c] = (unsigned char)(psz[c]);
- }
- return n;
+ for (size_t c=0; c<=len; c++)
+ buf[c] = (unsigned char)(psz[c]);
}
- return strlen(psz);
+ return len;
}
size_t wxCSConv::WC2MB(char *buf, const wchar_t *psz, size_t n) const
{
((wxCSConv *)this)->LoadNow(); // discard constness
- if (buf) {
- if (m_cset) {
- m_cset->w2m.Convert(psz, buf);
- } else {
- // latin-1 (direct)
- for (size_t c=0; c<n; c++)
- buf[c] = (psz[c]>0xff) ? '?' : psz[c];
- }
- return n;
- }
+ if (m_cset)
+ return m_cset->WC2MB(buf, psz, n);
+
+ // latin-1 (direct)
#if defined(__BORLANDC__) && (__BORLANDC__ > 0x530)
- return std::wcslen(psz);
+ size_t len=std::wcslen(psz);
#else
- return ::wcslen(psz);
+ size_t len=::wcslen(psz);
#endif
+ if (buf) {
+ for (size_t c=0; c<=len; c++)
+ buf[c] = (psz[c]>0xff) ? '?' : psz[c];
+ }
+ return len;
}
+#ifdef HAVE_ICONV_H
+class IC_CharSetConverter
+{
+public:
+ iconv_t cnv;
+ IC_CharSetConverter(IC_CharSet*from,IC_CharSet*to) {
+ cnv=iconv_open(wxConvLibc.cWX2MB(to->cname),wxConvLibc.cWX2MB(from->cname));
+ }
+ ~IC_CharSetConverter() {
+ if (cnv!=(iconv_t)-1) iconv_close(cnv);
+ }
+ size_t Convert(char*buf, const char*psz, size_t n) {
+ size_t inbuf = strlen(psz);
+ size_t outbuf = n;
+ size_t res = iconv(cnv,&psz,&inbuf,&buf,&outbuf);
+ if (res==(size_t)-1) return (size_t)-1;
+ return n-outbuf;
+ }
+};
+#endif
+
+class EC_CharSetConverter
+{
+public:
+ wxEncodingConverter cnv;
+ EC_CharSetConverter(EC_CharSet*from,EC_CharSet*to) {
+ cnv.Init(from->enc,to->enc);
+ }
+ size_t Convert(char*buf, const char*psz, size_t n) {
+ size_t inbuf = strlen(psz);
+ if (buf) cnv.Convert(psz,buf);
+ return inbuf;
+ }
+};
+
#else // !wxUSE_WCHAR_T
// ----------------------------------------------------------------------------