Remove ugly multiple definition hack

[lyx.git] / src / support / unicode.cpp

diff --git a/src/support/unicode.cpp b/src/support/unicode.cpp
index 39090769d2993b261445c3d71adec1417cb9ad2b..95415a5538e45ad11a4ed4695c5b277ae79024c3 100644 (file)
--- a/src/support/unicode.cpp
+++ b/src/support/unicode.cpp
@@ -14,17 +14,19 @@
 
 #include "support/unicode.h"
 #include "support/debug.h"
+#include "support/mutex.h"
 
 #include <iconv.h>
 
 #include <boost/cstdint.hpp>
 
 #include <cerrno>
-#include <iomanip>
 #include <map>
 #include <ostream>
+//Needed in MSVC
 #include <string>
 
+
 using namespace std;
 
 namespace {
@@ -64,6 +66,8 @@ struct IconvProcessor::Impl
        iconv_t cd;
        string tocode_;
        string fromcode_;
+
+       Mutex mutex_; // iconv() is not thread save, see #7240
 };
 
 
@@ -120,6 +124,8 @@ bool IconvProcessor::init()
 int IconvProcessor::convert(char const * buf, size_t buflen,
                char * outbuf, size_t maxoutsize)
 {
+       Mutex::Locker lock(&pimpl_->mutex_);
+
        if (buflen == 0)
                return 0;
 
@@ -197,6 +203,18 @@ int IconvProcessor::convert(char const * buf, size_t buflen,
 }
 
 
+std::string IconvProcessor::from() const
+{
+       return pimpl_->fromcode_;
+}
+
+
+std::string IconvProcessor::to() const
+{
+       return pimpl_->tocode_;
+}
+
+
 namespace {
 
 
@@ -210,17 +228,21 @@ iconv_convert(IconvProcessor & processor, InType const * buf, size_t buflen)
        char const * inbuf = reinterpret_cast<char const *>(buf);
        size_t inbytesleft = buflen * sizeof(InType);
 
-       size_t const outsize = 32768;
-       static char out[outsize];
-       char * outbuf = out;
+       static std::vector<char> outbuf(32768);
+       // The number of UCS4 code points in buf is at most inbytesleft.
+       // The output encoding will use at most
+       // max_encoded_bytes(pimpl_->tocode_) per UCS4 code point.
+       size_t maxoutbufsize = max_encoded_bytes(processor.to()) * inbytesleft;
+       if (outbuf.size() < maxoutbufsize)
+               outbuf.resize(maxoutbufsize);
 
-       int bytes = processor.convert(inbuf, inbytesleft, outbuf, outsize);
+       int bytes = processor.convert(inbuf, inbytesleft, &outbuf[0], outbuf.size());
        if (bytes <= 0)
                // Conversion failed
                // FIXME Maybe throw an exception and handle that in the caller?
                return vector<RetType>();
 
-       RetType const * tmp = reinterpret_cast<RetType const *>(out);
+       RetType const * tmp = reinterpret_cast<RetType const *>(&outbuf[0]);
        return vector<RetType>(tmp, tmp + bytes / sizeof(RetType));
 }
 
@@ -345,4 +367,34 @@ void ucs4_to_multibytes(char_type ucs4, vector<char> & out,
                out.clear();
 }
 
+int max_encoded_bytes(std::string const & encoding)
+{
+       // FIXME: this information should be transferred to lib/encodings
+       // UTF8 uses at most 4 bytes to represent one UCS4 code point
+       // (see RFC 3629). RFC 2279 specifies 6 bytes, but that
+       // information is outdated, and RFC 2279 has been superseded by
+       // RFC 3629.
+       // The CJK encodings use (different) multibyte representation as well.
+       // All other encodings encode one UCS4 code point in one byte
+       // (and can therefore only encode a subset of UCS4)
+       // Furthermore, all encodings that use shifting (like SJIS) do not work with
+       // iconv_codecvt_facet.
+       if (encoding == "UTF-8" ||
+           encoding == "GB" ||
+           encoding == "EUC-TW")
+               return 4;
+       else if (encoding == "EUC-JP")
+               return 3;
+       else if (encoding == "ISO-2022-JP")
+               return 8;
+       else if (encoding == "BIG5" ||
+                encoding == "EUC-KR" ||
+                encoding == "EUC-CN" ||
+                encoding == "SJIS" ||
+                encoding == "GBK")
+               return 2;
+       else
+               return 1;
+}
+
 } // namespace lyx