#include <config.h>
#include "support/docstream.h"
+#include "support/lstrings.h"
#include "support/unicode.h"
+#include <algorithm>
#include <cerrno>
#include <cstdio>
#include <cstring>
using lyx::ucs4_codeset;
+
+#if defined(_MSC_VER) && (_MSC_VER >= 1600)
+std::locale::id numpunct<lyx::char_type>::id;
+
+namespace std {
+// Implementation of numpunct<lyx::char_type> defined in numpunct_lyx_char_type.h
+typedef basic_string<lyx::char_type> string_type;
+
+string_type numpunct<lyx::char_type>::truename() const
+{
+ return lyx::from_ascii(numpunct<char>::truename());
+}
+
+string_type numpunct<lyx::char_type>::falsename() const
+{
+ return lyx::from_ascii(numpunct<char>::falsename());
+}
+
+} // namespace std
+
+#endif // _MSC_VER >= 1600
+
+
namespace {
// We use C IO throughout this file, because the facets might be used with
} else
out_cd_ = (iconv_t)(-1);
}
+ string const & encoding() const { return encoding_; }
protected:
virtual ~iconv_codecvt_facet()
{
fflush(stderr);
}
}
- virtual result do_out(state_type &, intern_type const * from,
+ result do_out(state_type &, intern_type const * from,
intern_type const * from_end, intern_type const *& from_next,
extern_type * to, extern_type * to_end,
- extern_type *& to_next) const
+ extern_type *& to_next) const override
{
+#define WORKAROUND_ICONV_BUG 1
+#if WORKAROUND_ICONV_BUG
+ // Due to a bug in some iconv versions, when the last char
+ // in the buffer is a wide char and the output encoding is
+ // ISO-2022-JP and we are going to switch to another encoding,
+ // the appropriate escape sequence for changing the character
+ // set is not output (see bugs 5216, 5280, and also 5489).
+ // As a workaround, we append a nul char in order to force
+ // a switch to ASCII, and then remove it from output after
+ // the conversion.
+ intern_type * from_new = nullptr;
+ intern_type const * from_old = from;
+ size_t extra = 0;
+ if (*(from_end - 1) >= 0x80 && encoding_ == "ISO-2022-JP") {
+ size_t len = from_end - from;
+ from_new = new intern_type[len + 1];
+ memcpy(from_new, from, len * sizeof(intern_type));
+ from_new[len] = 0;
+ from_end = from_new + len + 1;
+ from = from_new;
+ extra = 1;
+ }
+#endif
size_t inbytesleft = (from_end - from) * sizeof(intern_type);
size_t outbytesleft = (to_end - to) * sizeof(extern_type);
+#if WORKAROUND_ICONV_BUG
+ outbytesleft += extra * sizeof(extern_type);
+#endif
from_next = from;
to_next = to;
result const retval = do_iconv(out_cd_,
reinterpret_cast<char const **>(&from_next),
&inbytesleft, &to_next, &outbytesleft);
+#if WORKAROUND_ICONV_BUG
+ // Remove from output the nul char that we inserted at the end
+ // of the input buffer in order to circumvent an iconv bug.
+ if (from_new) {
+ --to_next;
+ --from_next;
+ from_next = from_old + (from_next - from);
+ from = from_old;
+ delete[] from_new;
+ }
+#endif
if (retval == base::error) {
fprintf(stderr,
"Error %d returned from iconv when converting from %s to %s: %s\n",
fprintf(stderr, "\nStopped at: 0x%04x\n", c);
fputs("Unconverted input:", stderr);
for (intern_type const * i = from_next + 1; i < from_end; ++i) {
- unsigned int const c = *i;
- fprintf(stderr, " 0x%04x", c);
+ unsigned int const cc = *i;
+ fprintf(stderr, " 0x%04x", cc);
}
fputs("\nConverted output:", stderr);
for (extern_type const * i = to; i < to_next; ++i) {
// extern_type may be signed, avoid output of
// something like 0xffffffc2
- unsigned int const c =
+ unsigned int const cc =
*reinterpret_cast<unsigned char const *>(i);
- fprintf(stderr, " 0x%02x", c);
+ fprintf(stderr, " 0x%02x", cc);
}
fputc('\n', stderr);
fflush(stderr);
}
return retval;
}
- virtual result do_unshift(state_type &, extern_type * to,
- extern_type *, extern_type *& to_next) const
+ result do_unshift(state_type &, extern_type * to,
+ extern_type *, extern_type *& to_next) const override
{
// utf8 does not use shifting
to_next = to;
return base::noconv;
}
- virtual result do_in(state_type &,
+ result do_in(state_type &,
extern_type const * from, extern_type const * from_end,
extern_type const *& from_next,
intern_type * to, intern_type * to_end,
- intern_type *& to_next) const
+ intern_type *& to_next) const override
{
size_t inbytesleft = (from_end - from) * sizeof(extern_type);
size_t outbytesleft = (to_end - to) * sizeof(intern_type);
fprintf(stderr, "\nStopped at: 0x%02x\n", c);
fputs("Unconverted input:", stderr);
for (extern_type const * i = from_next + 1; i < from_end; ++i) {
- unsigned int const c =
+ unsigned int const cc =
*reinterpret_cast<unsigned char const *>(i);
- fprintf(stderr, " 0x%02x", c);
+ fprintf(stderr, " 0x%02x", cc);
}
fputs("\nConverted output:", stderr);
for (intern_type const * i = to; i < to_next; ++i) {
- unsigned int const c = *i;
- fprintf(stderr, " 0x%02x", c);
+ unsigned int const cc = *i;
+ fprintf(stderr, " 0x%02x", cc);
}
fputc('\n', stderr);
fflush(stderr);
}
return retval;
}
- virtual int do_encoding() const throw()
+ int do_encoding() const noexcept override
{
return 0;
}
- virtual bool do_always_noconv() const throw()
+ bool do_always_noconv() const noexcept override
{
return false;
}
- virtual int do_length(state_type & /*state*/, extern_type const * from,
- extern_type const * end, size_t max) const
+ int do_length(state_type & /*state*/, extern_type const * from,
+ extern_type const * end, size_t max) const override
{
// The docs are a bit unclear about this method.
// It seems that we should calculate the actual length of the
return min(length, max);
#endif
}
- virtual int do_max_length() const throw()
+ int do_max_length() const noexcept override
{
- // 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)
- // Note that BIG5 and SJIS do not work with LaTeX (see lib/encodings).
- // 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" ||
- encoding_ == "ISO-2022-JP")
- return 3;
- else if (encoding_ == "BIG5" ||
- encoding_ == "EUC-KR" ||
- encoding_ == "EUC-CN" ||
- encoding_ == "SJIS" ||
- encoding_ == "GBK")
- return 2;
- else
- return 1;
+ return lyx::max_encoded_bytes(encoding_);
}
private:
/// Do the actual conversion. The interface is equivalent to that of
inbytesleft, to, outbytesleft);
if (converted == (size_t)(-1)) {
switch(errno) {
+ case 0:
+ // As strange as it may seem, this
+ // does happen on windows when parsing
+ // comments with accented chars in
+ // tex2lyx. See the following thread
+ // for details
+ // http://thread.gmane.org/gmane.editors.lyx.devel/117636
+ break;
case EINVAL:
case E2BIG:
return base::partial;
string encoding_;
};
-} // namespace anon
+} // namespace
namespace lyx {
template<class Ios>
void setEncoding(Ios & ios, string const & encoding, ios_base::openmode mode)
{
- // We must imbue the stream before openening the file
+ // We must imbue the stream before opening the file
locale global;
locale locale(global, new iconv_codecvt_facet(encoding, mode));
ios.imbue(locale);
}
-const char * iconv_codecvt_facet_exception::what() const throw()
+const char * iconv_codecvt_facet_exception::what() const noexcept
{
return "iconv problem in iconv_codecvt_facet initialization";
}
-idocfstream::idocfstream(string const & encoding) : base()
+ifdocstream::ifdocstream() : base()
{
- setEncoding(*this, encoding, in);
+ setEncoding(*this, "UTF-8", in);
+}
+
+
+ifdocstream::ifdocstream(SetEnc const & enc) : base()
+{
+ setEncoding(*this, enc.encoding, in);
}
-idocfstream::idocfstream(const char* s, ios_base::openmode mode,
+ifdocstream::ifdocstream(const char* s, ios_base::openmode mode,
string const & encoding)
: base()
{
}
-odocfstream::odocfstream(): base()
+ofdocstream::ofdocstream(): base()
{
setEncoding(*this, "UTF-8", out);
}
-odocfstream::odocfstream(const char* s, ios_base::openmode mode,
+ofdocstream::ofdocstream(SetEnc const & enc) : base()
+{
+ setEncoding(*this, enc.encoding, out);
+}
+
+
+ofdocstream::ofdocstream(const char* s, ios_base::openmode mode,
string const & encoding)
: base()
{
}
-void odocfstream::reset(string const & encoding)
+void ofdocstream::reset(string const & encoding)
{
setEncoding(*this, encoding, out);
}
if (has_facet<iconv_codecvt_facet>(os.rdbuf()->getloc())) {
// This stream must be a file stream, since we never imbue
// any other stream with a locale having a iconv_codecvt_facet.
+ iconv_codecvt_facet const & facet =
+ use_facet<iconv_codecvt_facet>(os.rdbuf()->getloc());
+
+ // FIXME Changing the codecvt facet of an open file is allowed,
+ // but unsafe for facets that use internal state (see the thread
+ // "iostreams: Does imbue() need to be called before open()?"
+ // in comp.std.c++.
+ // Currently it seems to work with gcc and MSVC, but not with
+ // clang on OS X.
+ // Avoid imbueing with the same encoding again if possible.
+ if (facet.encoding() == e.encoding)
+ return os;
+
// Flush the stream so that all pending output is written
// with the old encoding.
os.flush();
+
locale locale(os.rdbuf()->getloc(),
new iconv_codecvt_facet(e.encoding, ios_base::out));
- // FIXME Does changing the codecvt facet of an open file
- // stream always work? It does with gcc 4.1, but I have read
- // somewhere that it does not with MSVC.
- // What does the standard say?
os.imbue(locale);
}
return os;
}
+idocstream & operator<<(idocstream & is, SetEnc e)
+{
+ if (has_facet<iconv_codecvt_facet>(is.rdbuf()->getloc())) {
+ // This stream must be a file stream, since we never imbue
+ // any other stream with a locale having a iconv_codecvt_facet.
+ iconv_codecvt_facet const & facet =
+ use_facet<iconv_codecvt_facet>(is.rdbuf()->getloc());
+
+ // FIXME Changing the codecvt facet of an open file is allowed,
+ // but unsafe for facets that use internal state (see the thread
+ // "iostreams: Does imbue() need to be called before open()?"
+ // in comp.std.c++.
+ // Currently it seems to work with gcc and MSVC, but not with
+ // clang on OS X.
+ // Avoid imbueing with the same encoding again if possible.
+ if (facet.encoding() == e.encoding)
+ return is;
+
+ locale locale(is.rdbuf()->getloc(),
+ new iconv_codecvt_facet(e.encoding, ios_base::in));
+ is.imbue(locale);
+ }
+ return is;
+}
+
+
#if ! defined(USE_WCHAR_T)
odocstream & operator<<(odocstream & os, char c)
{
}
#endif
-}
+} // namespace lyx
+
#if ! defined(USE_WCHAR_T) && defined(__GNUC__)
// We get undefined references to these virtual methods. This looks like
// a bug in gcc. The implementation here does not do anything useful, since
-// it is overriden in iconv_codecvt_facet.
+// it is overridden in iconv_codecvt_facet.
namespace std {
template<> codecvt<lyx::char_type, char, mbstate_t>::result
template<>
-int codecvt<lyx::char_type, char, mbstate_t>::do_encoding() const throw()
+int codecvt<lyx::char_type, char, mbstate_t>::do_encoding() const noexcept
{
return 0;
}
template<>
-bool codecvt<lyx::char_type, char, mbstate_t>::do_always_noconv() const throw()
+bool codecvt<lyx::char_type, char, mbstate_t>::do_always_noconv() const noexcept
{
return true;
}
-#if __GNUC__ == 3 && __GNUC_MINOR__ < 4
-
-template<>
-int codecvt<lyx::char_type, char, mbstate_t>::do_length(
- mbstate_t const &, const char *, const char *, size_t) const
-{
- return 1;
-}
-
-#else
-
template<>
int codecvt<lyx::char_type, char, mbstate_t>::do_length(
mbstate_t &, const char *, const char *, size_t) const
return 1;
}
-#endif
-
template<>
-int codecvt<lyx::char_type, char, mbstate_t>::do_max_length() const throw()
+int codecvt<lyx::char_type, char, mbstate_t>::do_max_length() const noexcept
{
return 4;
}