strconv.cpp

来自「A*算法 A*算法 A*算法 A*算法A*算法A*算法」· C++ 代码 · 共 2,176 行 · 第 1/5 页

                                     (errno != E2BIG || bufLeft != 0))
#else
#define ICONV_FAILED(cres, bufLeft)  (cres == (size_t)-1)
#endif

#define ICONV_CHAR_CAST(x)  ((ICONV_CONST char **)(x))

// ----------------------------------------------------------------------------
// 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;

    bool IsOk() const
        { return (m2w != (iconv_t)-1) && (w2m != (iconv_t)-1); }

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 const char *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;
};

// 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;
}

const char *wxMBConv_iconv::ms_wcCharsetName = NULL;
bool wxMBConv_iconv::ms_wcNeedsSwap = false;

wxMBConv_iconv::wxMBConv_iconv(const wxChar *name)
{
    // 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 == NULL)
    {
        ms_wcNeedsSwap = false;

        // try charset with explicit bytesex info (e.g. "UCS-4LE"):
        ms_wcCharsetName = WC_NAME_BEST;
        m2w = iconv_open(ms_wcCharsetName, cname);

        if (m2w == (iconv_t)-1)
        {
            // try charset w/o bytesex info (e.g. "UCS4")
            // and check for bytesex ourselves:
            ms_wcCharsetName = WC_NAME;
            m2w = iconv_open(ms_wcCharsetName, cname);

            // last bet, try if it knows WCHAR_T pseudo-charset
            if (m2w == (iconv_t)-1)
            {
                ms_wcCharsetName = "WCHAR_T";
                m2w = iconv_open(ms_wcCharsetName, cname);
            }

            if (m2w != (iconv_t)-1)
            {
                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))
                {
                    ms_wcCharsetName = NULL;
                    wxLogLastError(wxT("iconv"));
                    wxLogError(_("Conversion to charset '%s' doesn't work."), name);
                }
                else
                {
                    ms_wcNeedsSwap = wbuf[0] != (wchar_t)buf[0];
                }
            }
            else
            {
                ms_wcCharsetName = NULL;

                // VS: we must not output an error here, since wxWidgets will safely
                //     fall back to using wxEncodingConverter.
                wxLogTrace(TRACE_STRCONV, wxT("Impossible to convert to/from charset '%s' with iconv, falling back to wxEncodingConverter."), name);
            }
        }
        wxLogTrace(TRACE_STRCONV,
                   wxT("wchar_t charset is '%s', needs swap: %i"),
                   ms_wcCharsetName ? ms_wcCharsetName : "<none>", ms_wcNeedsSwap);
    }
    else // we already have ms_wcCharsetName
    {
        m2w = iconv_open(ms_wcCharsetName, cname);
    }

    // NB: don't ever pass NULL to iconv_open(), it may crash!
    if ( ms_wcCharsetName )
    {
        w2m = iconv_open( cname, ms_wcCharsetName);
    }
    else
    {
        w2m = (iconv_t)-1;
    }
}

wxMBConv_iconv::~wxMBConv_iconv()
{
    if ( m2w != (iconv_t)-1 )
        iconv_close(m2w);
    if ( w2m != (iconv_t)-1 )
        iconv_close(w2m);
}

size_t wxMBConv_iconv::MB2WC(wchar_t *buf, const char *psz, size_t n) const
{
#if wxUSE_THREADS
    // NB: iconv() is MT-safe, but each thread must use it's own iconv_t handle.
    //     Unfortunately there is 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

    size_t inbuf = strlen(psz);
    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
            WC_BSWAP(buf /* _not_ bufPtr */, res)
        }

        // NB: iconv was given only strlen(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[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 (size_t)-1;
    }

    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 inbuf = wxWcslen(psz) * 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
        // 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)
        tmpbuf=(wchar_t*)malloc((inbuf+1)*SIZEOF_WCHAR_T);
        memcpy(tmpbuf,psz,(inbuf+1)*SIZEOF_WCHAR_T);
        WC_BSWAP(tmpbuf, inbuf)
        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))
    {
        //VS: it is ok if iconv fails, hence trace only
        wxLogTrace(TRACE_STRCONV, wxT("iconv failed: %s"), wxSysErrorMsg(wxSysErrorCode()));
        return (size_t)-1;
    }

    return res;
}

#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;
    }

#if wxUSE_FONTMAP
    wxMBConv_win32(const wxChar* name)
    {
        m_CodePage = wxCharsetToCodepage(name);
    }

    wxMBConv_win32(wxFontEncoding encoding)
    {
        m_CodePage = wxEncodingToCodepage(encoding);
    }
#endif

    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
        //
        // note however that using MB_ERR_INVALID_CHARS with CP_UTF7 results in
        // an error (tested under Windows Server 2003) and apparently it is
        // done on purpose, i.e. the function accepts any input in this case
        // and although I'd prefer to return error on ill-formed output, our
        // own wxMBConvUTF7 doesn't detect errors (e.g. lone "+" which is
        // explicitly ill-formed according to RFC 2152) neither so we don't
        // even have any fallback here...
        int flags = m_CodePage == CP_UTF7 ? 0 : 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
                             );

        // 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 ? len - 1 : (size_t)-1;
    }

    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 (size_t)-1;
        }

        // 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 (size_t)-1;
            }
            else // we must resort to double tripping...
            {
                wxWCharBuffer wcBuf(n);
                if ( MB2WC(wcBuf.data(), buf, n) == (size_t)-1 ||
                        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 (size_t)-1;
                }
            }
        }

        // see the comment above for the reason of "len - 1"
        return len - 1;
    }