+#define ICONV_FAILED(cres, bufLeft) (cres == (size_t)-1)
+#endif
+
+#define ICONV_CHAR_CAST(x) ((ICONV_CONST char **)(x))
+
+#define ICONV_T_INVALID ((iconv_t)-1)
+
+#if SIZEOF_WCHAR_T == 4
+ #define WC_BSWAP wxUINT32_SWAP_ALWAYS
+ #define WC_ENC wxFONTENCODING_UTF32
+#elif SIZEOF_WCHAR_T == 2
+ #define WC_BSWAP wxUINT16_SWAP_ALWAYS
+ #define WC_ENC wxFONTENCODING_UTF16
+#else // sizeof(wchar_t) != 2 nor 4
+ // does this ever happen?
+ #error "Unknown sizeof(wchar_t): please report this to wx-dev@lists.wxwindows.org"
+#endif
+
+// ----------------------------------------------------------------------------
+// wxMBConv_iconv: encapsulates an iconv character set
+// ----------------------------------------------------------------------------
+
+class wxMBConv_iconv : public wxMBConv
+{
+public:
+ wxMBConv_iconv(const wxChar *name);
+ virtual ~wxMBConv_iconv();
+
+ virtual size_t MB2WC(wchar_t *buf, const char *psz, size_t n) const;
+ virtual size_t WC2MB(char *buf, const wchar_t *psz, size_t n) const;
+
+ // classify this encoding as explained in wxMBConv::GetMBNulLen() comment
+ virtual size_t GetMBNulLen() const;
+
+ virtual wxMBConv *Clone() const
+ {
+ wxMBConv_iconv *p = new wxMBConv_iconv(m_name);
+ p->m_minMBCharWidth = m_minMBCharWidth;
+ return p;
+ }
+
+ bool IsOk() const
+ { return (m2w != ICONV_T_INVALID) && (w2m != ICONV_T_INVALID); }
+
+protected:
+ // the iconv handlers used to translate from multibyte
+ // to wide char and in the other direction
+ iconv_t m2w,
+ w2m;
+
+#if wxUSE_THREADS
+ // guards access to m2w and w2m objects
+ wxMutex m_iconvMutex;
+#endif
+
+private:
+ // the name (for iconv_open()) of a wide char charset -- if none is
+ // available on this machine, it will remain NULL
+ static wxString ms_wcCharsetName;
+
+ // true if the wide char encoding we use (i.e. ms_wcCharsetName) has
+ // different endian-ness than the native one
+ static bool ms_wcNeedsSwap;
+
+
+ // name of the encoding handled by this conversion
+ wxString m_name;
+
+ // cached result of GetMBNulLen(); set to 0 meaning "unknown"
+ // initially
+ size_t m_minMBCharWidth;
+};
+
+// make the constructor available for unit testing
+WXDLLIMPEXP_BASE wxMBConv* new_wxMBConv_iconv( const wxChar* name )
+{
+ wxMBConv_iconv* result = new wxMBConv_iconv( name );
+ if ( !result->IsOk() )
+ {
+ delete result;
+ return 0;
+ }
+
+ return result;
+}
+
+wxString wxMBConv_iconv::ms_wcCharsetName;
+bool wxMBConv_iconv::ms_wcNeedsSwap = false;
+
+wxMBConv_iconv::wxMBConv_iconv(const wxChar *name)
+ : m_name(name)
+{
+ m_minMBCharWidth = 0;
+
+ // iconv operates with chars, not wxChars, but luckily it uses only ASCII
+ // names for the charsets
+ const wxCharBuffer cname(wxString(name).ToAscii());
+
+ // check for charset that represents wchar_t:
+ if ( ms_wcCharsetName.empty() )
+ {
+ wxLogTrace(TRACE_STRCONV, _T("Looking for wide char codeset:"));
+
+#if wxUSE_FONTMAP
+ const wxChar **names = wxFontMapperBase::GetAllEncodingNames(WC_ENC);
+#else // !wxUSE_FONTMAP
+ static const wxChar *names_static[] =
+ {
+#if SIZEOF_WCHAR_T == 4
+ _T("UCS-4"),
+#elif SIZEOF_WCHAR_T = 2
+ _T("UCS-2"),
+#endif
+ NULL
+ };
+ const wxChar **names = names_static;
+#endif // wxUSE_FONTMAP/!wxUSE_FONTMAP
+
+ for ( ; *names && ms_wcCharsetName.empty(); ++names )
+ {
+ const wxString nameCS(*names);
+
+ // first try charset with explicit bytesex info (e.g. "UCS-4LE"):
+ wxString nameXE(nameCS);
+
+#ifdef WORDS_BIGENDIAN
+ nameXE += _T("BE");
+#else // little endian
+ nameXE += _T("LE");
+#endif
+
+ wxLogTrace(TRACE_STRCONV, _T(" trying charset \"%s\""),
+ nameXE.c_str());
+
+ m2w = iconv_open(nameXE.ToAscii(), cname);
+ if ( m2w == ICONV_T_INVALID )
+ {
+ // try charset w/o bytesex info (e.g. "UCS4")
+ wxLogTrace(TRACE_STRCONV, _T(" trying charset \"%s\""),
+ nameCS.c_str());
+ m2w = iconv_open(nameCS.ToAscii(), cname);
+
+ // and check for bytesex ourselves:
+ if ( m2w != ICONV_T_INVALID )
+ {
+ char buf[2], *bufPtr;
+ wchar_t wbuf[2], *wbufPtr;
+ size_t insz, outsz;
+ size_t res;
+
+ buf[0] = 'A';
+ buf[1] = 0;
+ wbuf[0] = 0;
+ insz = 2;
+ outsz = SIZEOF_WCHAR_T * 2;
+ wbufPtr = wbuf;
+ bufPtr = buf;
+
+ res = iconv(
+ m2w, ICONV_CHAR_CAST(&bufPtr), &insz,
+ (char**)&wbufPtr, &outsz);
+
+ if (ICONV_FAILED(res, insz))
+ {
+ wxLogLastError(wxT("iconv"));
+ wxLogError(_("Conversion to charset '%s' doesn't work."),
+ nameCS.c_str());
+ }
+ else // ok, can convert to this encoding, remember it
+ {
+ ms_wcCharsetName = nameCS;
+ ms_wcNeedsSwap = wbuf[0] != (wchar_t)buf[0];
+ }
+ }
+ }
+ else // use charset not requiring byte swapping
+ {
+ ms_wcCharsetName = nameXE;
+ }
+ }
+
+ wxLogTrace(TRACE_STRCONV,
+ wxT("iconv wchar_t charset is \"%s\"%s"),
+ ms_wcCharsetName.empty() ? _T("<none>")
+ : ms_wcCharsetName.c_str(),
+ ms_wcNeedsSwap ? _T(" (needs swap)")
+ : _T(""));
+ }
+ else // we already have ms_wcCharsetName
+ {
+ m2w = iconv_open(ms_wcCharsetName.ToAscii(), cname);
+ }
+
+ if ( ms_wcCharsetName.empty() )
+ {
+ w2m = ICONV_T_INVALID;
+ }
+ else
+ {
+ w2m = iconv_open(cname, ms_wcCharsetName.ToAscii());
+ if ( w2m == ICONV_T_INVALID )
+ {
+ wxLogTrace(TRACE_STRCONV,
+ wxT("\"%s\" -> \"%s\" works but not the converse!?"),
+ ms_wcCharsetName.c_str(), cname.data());
+ }
+ }
+}
+
+wxMBConv_iconv::~wxMBConv_iconv()
+{
+ if ( m2w != ICONV_T_INVALID )
+ iconv_close(m2w);
+ if ( w2m != ICONV_T_INVALID )
+ iconv_close(w2m);
+}
+
+size_t wxMBConv_iconv::MB2WC(wchar_t *buf, const char *psz, size_t n) const
+{
+ // find the string length: notice that must be done differently for
+ // NUL-terminated strings and UTF-16/32 which are terminated with 2/4 NULs
+ size_t inbuf;
+ const size_t nulLen = GetMBNulLen();
+ switch ( nulLen )
+ {
+ default:
+ return wxCONV_FAILED;
+
+ case 1:
+ inbuf = strlen(psz); // arguably more optimized than our version
+ break;
+
+ case 2:
+ case 4:
+ // for UTF-16/32 not only we need to have 2/4 consecutive NULs but
+ // they also have to start at character boundary and not span two
+ // adjacent characters
+ const char *p;
+ for ( p = psz; NotAllNULs(p, nulLen); p += nulLen )
+ ;
+ inbuf = p - psz;
+ break;
+ }
+
+#if wxUSE_THREADS
+ // NB: iconv() is MT-safe, but each thread must use its own iconv_t handle.
+ // Unfortunately there are a couple of global wxCSConv objects such as
+ // wxConvLocal that are used all over wx code, so we have to make sure
+ // the handle is used by at most one thread at the time. Otherwise
+ // only a few wx classes would be safe to use from non-main threads
+ // as MB<->WC conversion would fail "randomly".
+ wxMutexLocker lock(wxConstCast(this, wxMBConv_iconv)->m_iconvMutex);
+#endif // wxUSE_THREADS
+
+ size_t outbuf = n * SIZEOF_WCHAR_T;
+ size_t res, cres;
+ // VS: Use these instead of psz, buf because iconv() modifies its arguments:
+ wchar_t *bufPtr = buf;
+ const char *pszPtr = psz;
+
+ if (buf)
+ {
+ // have destination buffer, convert there
+ cres = iconv(m2w,
+ ICONV_CHAR_CAST(&pszPtr), &inbuf,
+ (char**)&bufPtr, &outbuf);
+ res = n - (outbuf / SIZEOF_WCHAR_T);
+
+ if (ms_wcNeedsSwap)
+ {
+ // convert to native endianness
+ for ( unsigned i = 0; i < res; i++ )
+ buf[n] = WC_BSWAP(buf[i]);
+ }
+
+ // NUL-terminate the string if there is any space left
+ if (res < n)
+ buf[res] = 0;
+ }
+ else
+ {
+ // no destination buffer... convert using temp buffer
+ // to calculate destination buffer requirement
+ wchar_t tbuf[8];
+ res = 0;
+
+ do
+ {
+ bufPtr = tbuf;
+ outbuf = 8 * SIZEOF_WCHAR_T;
+
+ cres = iconv(m2w,
+ ICONV_CHAR_CAST(&pszPtr), &inbuf,
+ (char**)&bufPtr, &outbuf );
+
+ res += 8 - (outbuf / SIZEOF_WCHAR_T);
+ }
+ while ((cres == (size_t)-1) && (errno == E2BIG));
+ }
+
+ if (ICONV_FAILED(cres, inbuf))
+ {
+ //VS: it is ok if iconv fails, hence trace only
+ wxLogTrace(TRACE_STRCONV, wxT("iconv failed: %s"), wxSysErrorMsg(wxSysErrorCode()));
+ return wxCONV_FAILED;
+ }
+
+ return res;
+}
+
+size_t wxMBConv_iconv::WC2MB(char *buf, const wchar_t *psz, size_t n) const
+{
+#if wxUSE_THREADS
+ // NB: explained in MB2WC
+ wxMutexLocker lock(wxConstCast(this, wxMBConv_iconv)->m_iconvMutex);
+#endif
+
+ size_t inlen = wxWcslen(psz);
+ size_t inbuf = inlen * SIZEOF_WCHAR_T;
+ size_t outbuf = n;
+ size_t res, cres;
+
+ wchar_t *tmpbuf = 0;
+
+ if (ms_wcNeedsSwap)
+ {
+ // need to copy to temp buffer to switch endianness
+ // (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)
+ tmpbuf = (wchar_t *)malloc(inbuf + SIZEOF_WCHAR_T);
+ for ( size_t i = 0; i < inlen; i++ )
+ tmpbuf[n] = WC_BSWAP(psz[i]);
+
+ tmpbuf[inlen] = L'\0';
+ psz = tmpbuf;
+ }
+
+ if (buf)
+ {
+ // have destination buffer, convert there
+ cres = iconv( w2m, ICONV_CHAR_CAST(&psz), &inbuf, &buf, &outbuf );
+
+ res = n - outbuf;
+
+ // NB: iconv was given only wcslen(psz) characters on input, and so
+ // it couldn't convert the trailing zero. Let's do it ourselves
+ // if there's some room left for it in the output buffer.
+ if (res < n)
+ buf[0] = 0;
+ }
+ 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, ICONV_CHAR_CAST(&psz), &inbuf, &buf, &outbuf );
+
+ res += 16 - outbuf;
+ }
+ while ((cres == (size_t)-1) && (errno == E2BIG));
+ }
+
+ if (ms_wcNeedsSwap)
+ {
+ free(tmpbuf);
+ }
+
+ if (ICONV_FAILED(cres, inbuf))
+ {
+ wxLogTrace(TRACE_STRCONV, wxT("iconv failed: %s"), wxSysErrorMsg(wxSysErrorCode()));
+ return wxCONV_FAILED;
+ }
+
+ return res;
+}
+
+size_t wxMBConv_iconv::GetMBNulLen() const
+{
+ if ( m_minMBCharWidth == 0 )
+ {
+ wxMBConv_iconv * const self = wxConstCast(this, wxMBConv_iconv);
+
+#if wxUSE_THREADS
+ // NB: explained in MB2WC
+ wxMutexLocker lock(self->m_iconvMutex);
+#endif
+
+ wchar_t *wnul = L"";
+ char buf[8]; // should be enough for NUL in any encoding
+ size_t inLen = sizeof(wchar_t),
+ outLen = WXSIZEOF(buf);
+ char *inBuff = (char *)wnul;
+ char *outBuff = buf;
+ if ( iconv(w2m, ICONV_CHAR_CAST(&inBuff), &inLen, &outBuff, &outLen) == (size_t)-1 )
+ {
+ self->m_minMBCharWidth = (size_t)-1;
+ }
+ else // ok
+ {
+ self->m_minMBCharWidth = outBuff - buf;
+ }
+ }
+
+ return m_minMBCharWidth;
+}
+
+#endif // HAVE_ICONV
+
+
+// ============================================================================
+// Win32 conversion classes
+// ============================================================================
+
+#ifdef wxHAVE_WIN32_MB2WC
+
+// from utils.cpp
+#if wxUSE_FONTMAP
+extern WXDLLIMPEXP_BASE long wxCharsetToCodepage(const wxChar *charset);
+extern WXDLLIMPEXP_BASE long wxEncodingToCodepage(wxFontEncoding encoding);
+#endif
+
+class wxMBConv_win32 : public wxMBConv
+{
+public:
+ wxMBConv_win32()
+ {
+ m_CodePage = CP_ACP;
+ m_minMBCharWidth = 0;
+ }
+
+ wxMBConv_win32(const wxMBConv_win32& conv)
+ : wxMBConv()
+ {
+ m_CodePage = conv.m_CodePage;
+ m_minMBCharWidth = conv.m_minMBCharWidth;
+ }
+
+#if wxUSE_FONTMAP
+ wxMBConv_win32(const wxChar* name)
+ {
+ m_CodePage = wxCharsetToCodepage(name);
+ m_minMBCharWidth = 0;
+ }
+
+ wxMBConv_win32(wxFontEncoding encoding)
+ {
+ m_CodePage = wxEncodingToCodepage(encoding);
+ m_minMBCharWidth = 0;
+ }
+#endif // wxUSE_FONTMAP
+
+ virtual size_t MB2WC(wchar_t *buf, const char *psz, size_t n) const
+ {
+ // note that we have to use MB_ERR_INVALID_CHARS flag as it without it
+ // the behaviour is not compatible with the Unix version (using iconv)
+ // and break the library itself, e.g. wxTextInputStream::NextChar()
+ // wouldn't work if reading an incomplete MB char didn't result in an
+ // error
+ //
+ // Moreover, MB_ERR_INVALID_CHARS is only supported on Win 2K SP4 or
+ // Win XP or newer and it is not supported for UTF-[78] so we always
+ // use our own conversions in this case. See
+ // http://blogs.msdn.com/michkap/archive/2005/04/19/409566.aspx
+ // http://msdn.microsoft.com/library/en-us/intl/unicode_17si.asp
+ if ( m_CodePage == CP_UTF8 )
+ {
+ return wxConvUTF8.MB2WC(buf, psz, n);
+ }
+
+ if ( m_CodePage == CP_UTF7 )
+ {
+ return wxConvUTF7.MB2WC(buf, psz, n);
+ }
+
+ int flags = 0;
+ if ( (m_CodePage < 50000 && m_CodePage != CP_SYMBOL) &&
+ IsAtLeastWin2kSP4() )
+ {
+ flags = MB_ERR_INVALID_CHARS;
+ }
+
+ const size_t len = ::MultiByteToWideChar
+ (
+ m_CodePage, // code page
+ flags, // flags: fall on error
+ psz, // input string
+ -1, // its length (NUL-terminated)
+ buf, // output string
+ buf ? n : 0 // size of output buffer
+ );
+ if ( !len )
+ {
+ // function totally failed
+ return wxCONV_FAILED;
+ }
+
+ // if we were really converting and didn't use MB_ERR_INVALID_CHARS,
+ // check if we succeeded, by doing a double trip:
+ if ( !flags && buf )
+ {
+ const size_t mbLen = strlen(psz);
+ wxCharBuffer mbBuf(mbLen);
+ if ( ::WideCharToMultiByte
+ (
+ m_CodePage,
+ 0,
+ buf,
+ -1,
+ mbBuf.data(),
+ mbLen + 1, // size in bytes, not length
+ NULL,
+ NULL
+ ) == 0 ||
+ strcmp(mbBuf, psz) != 0 )
+ {
+ // we didn't obtain the same thing we started from, hence
+ // the conversion was lossy and we consider that it failed
+ return wxCONV_FAILED;
+ }
+ }
+
+ // note that it returns count of written chars for buf != NULL and size
+ // of the needed buffer for buf == NULL so in either case the length of
+ // the string (which never includes the terminating NUL) is one less
+ return len - 1;
+ }
+
+ virtual size_t WC2MB(char *buf, const wchar_t *pwz, size_t n) const
+ {
+ /*
+ we have a problem here: by default, WideCharToMultiByte() may
+ replace characters unrepresentable in the target code page with bad
+ quality approximations such as turning "1/2" symbol (U+00BD) into
+ "1" for the code pages which don't have it and we, obviously, want
+ to avoid this at any price
+
+ the trouble is that this function does it _silently_, i.e. it won't
+ even tell us whether it did or not... Win98/2000 and higher provide
+ WC_NO_BEST_FIT_CHARS but it doesn't work for the older systems and
+ we have to resort to a round trip, i.e. check that converting back
+ results in the same string -- this is, of course, expensive but
+ otherwise we simply can't be sure to not garble the data.
+ */
+
+ // determine if we can rely on WC_NO_BEST_FIT_CHARS: according to MSDN
+ // it doesn't work with CJK encodings (which we test for rather roughly
+ // here...) nor with UTF-7/8 nor, of course, with Windows versions not
+ // supporting it
+ BOOL usedDef wxDUMMY_INITIALIZE(false);
+ BOOL *pUsedDef;
+ int flags;
+ if ( CanUseNoBestFit() && m_CodePage < 50000 )
+ {
+ // it's our lucky day
+ flags = WC_NO_BEST_FIT_CHARS;
+ pUsedDef = &usedDef;
+ }
+ else // old system or unsupported encoding
+ {
+ flags = 0;
+ pUsedDef = NULL;
+ }
+
+ const size_t len = ::WideCharToMultiByte
+ (
+ m_CodePage, // code page
+ flags, // either none or no best fit
+ pwz, // input string
+ -1, // it is (wide) NUL-terminated
+ buf, // output buffer
+ buf ? n : 0, // and its size
+ NULL, // default "replacement" char
+ pUsedDef // [out] was it used?
+ );
+
+ if ( !len )
+ {
+ // function totally failed
+ return wxCONV_FAILED;
+ }
+
+ // if we were really converting, check if we succeeded
+ if ( buf )
+ {
+ if ( flags )
+ {
+ // check if the conversion failed, i.e. if any replacements
+ // were done
+ if ( usedDef )
+ return wxCONV_FAILED;
+ }
+ else // we must resort to double tripping...
+ {
+ wxWCharBuffer wcBuf(n);
+ if ( MB2WC(wcBuf.data(), buf, n) == wxCONV_FAILED ||
+ wcscmp(wcBuf, pwz) != 0 )
+ {
+ // we didn't obtain the same thing we started from, hence
+ // the conversion was lossy and we consider that it failed
+ return wxCONV_FAILED;
+ }
+ }
+ }
+
+ // see the comment above for the reason of "len - 1"
+ return len - 1;
+ }
+
+ virtual size_t GetMBNulLen() const
+ {
+ if ( m_minMBCharWidth == 0 )
+ {
+ int len = ::WideCharToMultiByte
+ (
+ m_CodePage, // code page
+ 0, // no flags
+ L"", // input string
+ 1, // translate just the NUL
+ NULL, // output buffer
+ 0, // and its size
+ NULL, // no replacement char
+ NULL // [out] don't care if it was used
+ );
+
+ wxMBConv_win32 * const self = wxConstCast(this, wxMBConv_win32);
+ switch ( len )
+ {
+ default:
+ wxLogDebug(_T("Unexpected NUL length %d"), len);
+ self->m_minMBCharWidth = (size_t)-1;
+ break;
+
+ case 0:
+ self->m_minMBCharWidth = (size_t)-1;
+ break;
+
+ case 1:
+ case 2:
+ case 4:
+ self->m_minMBCharWidth = len;
+ break;
+ }
+ }
+
+ return m_minMBCharWidth;
+ }
+
+ virtual wxMBConv *Clone() const { return new wxMBConv_win32(*this); }
+
+ bool IsOk() const { return m_CodePage != -1; }
+
+private:
+ static bool CanUseNoBestFit()
+ {
+ static int s_isWin98Or2k = -1;
+
+ if ( s_isWin98Or2k == -1 )
+ {
+ int verMaj, verMin;
+ switch ( wxGetOsVersion(&verMaj, &verMin) )
+ {
+ case wxOS_WINDOWS_9X:
+ s_isWin98Or2k = verMaj >= 4 && verMin >= 10;
+ break;
+
+ case wxOS_WINDOWS_NT:
+ s_isWin98Or2k = verMaj >= 5;
+ break;
+
+ default:
+ // unknown: be conservative by default
+ s_isWin98Or2k = 0;
+ break;
+ }
+
+ wxASSERT_MSG( s_isWin98Or2k != -1, _T("should be set above") );
+ }
+
+ return s_isWin98Or2k == 1;
+ }
+
+ static bool IsAtLeastWin2kSP4()
+ {
+#ifdef __WXWINCE__
+ return false;
+#else
+ static int s_isAtLeastWin2kSP4 = -1;
+
+ if ( s_isAtLeastWin2kSP4 == -1 )
+ {
+ OSVERSIONINFOEX ver;
+
+ memset(&ver, 0, sizeof(ver));
+ ver.dwOSVersionInfoSize = sizeof(ver);
+ GetVersionEx((OSVERSIONINFO*)&ver);
+
+ s_isAtLeastWin2kSP4 =
+ ((ver.dwMajorVersion > 5) || // Vista+
+ (ver.dwMajorVersion == 5 && ver.dwMinorVersion > 0) || // XP/2003
+ (ver.dwMajorVersion == 5 && ver.dwMinorVersion == 0 &&
+ ver.wServicePackMajor >= 4)) // 2000 SP4+
+ ? 1 : 0;
+ }
+
+ return s_isAtLeastWin2kSP4 == 1;
projects / wxWidgets.git / blobdiff