added wxEncodingConverter::Convert(char*,wxChar*) and variants ; filled equivalence tables ; GetAllEquivalents now returns semi-sorted array (platfrom equivs, then rest of world)
git-svn-id: https://svn.wxwidgets.org/svn/wx/wxWidgets/trunk@5177 c3d73ce0-8a6f-49c7-b76d-6d57e0e08775
This commit is contained in:
@@ -66,7 +66,7 @@ static CharsetItem* BuildReverseTable(wxUint16 *tbl)
|
||||
wxEncodingConverter::wxEncodingConverter()
|
||||
{
|
||||
m_Table = NULL;
|
||||
m_UnicodeInput = FALSE;
|
||||
m_UnicodeInput = m_UnicodeOutput = FALSE;
|
||||
m_JustCopy = FALSE;
|
||||
}
|
||||
|
||||
@@ -85,6 +85,7 @@ bool wxEncodingConverter::Init(wxFontEncoding input_enc, wxFontEncoding output_e
|
||||
|
||||
if (input_enc == output_enc) {m_JustCopy = TRUE; return TRUE;}
|
||||
|
||||
m_UnicodeOutput = (output_enc == wxFONTENCODING_UNICODE);
|
||||
m_JustCopy = FALSE;
|
||||
|
||||
if (input_enc == wxFONTENCODING_UNICODE)
|
||||
@@ -150,6 +151,7 @@ bool wxEncodingConverter::Init(wxFontEncoding input_enc, wxFontEncoding output_e
|
||||
}
|
||||
|
||||
|
||||
|
||||
void wxEncodingConverter::Convert(const wxChar* input, wxChar* output)
|
||||
{
|
||||
if (m_JustCopy)
|
||||
@@ -173,6 +175,83 @@ void wxEncodingConverter::Convert(const wxChar* input, wxChar* output)
|
||||
}
|
||||
|
||||
|
||||
#if wxUSE_UNICODE // otherwise wxChar === char
|
||||
|
||||
void wxEncodingConverter::Convert(const char* input, wxChar* output)
|
||||
{
|
||||
wxASSERT_MSG(!m_UnicodeInput, wxT("You cannot convert from unicode if input is const char*!"));
|
||||
|
||||
const char *i;
|
||||
wxChar *o;
|
||||
|
||||
if (m_JustCopy)
|
||||
{
|
||||
for (i = input, o = output; *i != 0;)
|
||||
*(o++) = (wxChar)(*(i++));
|
||||
*o = 0;
|
||||
return;
|
||||
}
|
||||
|
||||
wxASSERT_MSG(m_Table != NULL, wxT("You must call wxEncodingConverter::Init() before actually converting!"));
|
||||
|
||||
for (i = input, o = output; *i != 0;)
|
||||
*(o++) = (wxChar)(m_Table[(wxUint8)*(i++)]);
|
||||
*o = 0;
|
||||
}
|
||||
|
||||
|
||||
|
||||
void wxEncodingConverter::Convert(const wxChar* input, char* output)
|
||||
{
|
||||
wxASSERT_MSG(!m_UnicodeOutput, wxT("You cannot convert to unicode if output is const char*!"));
|
||||
|
||||
const wxChar *i;
|
||||
char *o;
|
||||
|
||||
if (m_JustCopy)
|
||||
{
|
||||
for (i = input, o = output; *i != 0;)
|
||||
*(o++) = (char)(*(i++));
|
||||
*o = 0;
|
||||
return;
|
||||
}
|
||||
|
||||
wxASSERT_MSG(m_Table != NULL, wxT("You must call wxEncodingConverter::Init() before actually converting!"));
|
||||
|
||||
if (m_UnicodeInput)
|
||||
for (i = input, o = output; *i != 0; i++, o++)
|
||||
*o = (char)(m_Table[(wxUint16)*i]);
|
||||
else
|
||||
for (i = input, o = output; *i != 0; i++, o++)
|
||||
*o = (char)(m_Table[(wxUint8)*i]);
|
||||
*o = 0;
|
||||
}
|
||||
|
||||
|
||||
|
||||
void wxEncodingConverter::Convert(const char* input, char* output)
|
||||
{
|
||||
wxASSERT_MSG(!m_UnicodeOutput, wxT("You cannot convert to unicode if output is const char*!"));
|
||||
wxASSERT_MSG(!m_UnicodeInput, wxT("You cannot convert from unicode if input is const char*!"));
|
||||
|
||||
const char *i;
|
||||
char *o;
|
||||
|
||||
if (m_JustCopy)
|
||||
{
|
||||
strcpy(output, input);
|
||||
return;
|
||||
}
|
||||
|
||||
wxASSERT_MSG(m_Table != NULL, wxT("You must call wxEncodingConverter::Init() before actually converting!"));
|
||||
|
||||
for (i = input, o = output; *i != 0;)
|
||||
*(o++) = (char)(m_Table[(wxUint8)*(i++)]);
|
||||
*o = 0;
|
||||
}
|
||||
|
||||
#endif // wxUSE_UNICODE
|
||||
|
||||
|
||||
wxString wxEncodingConverter::Convert(const wxString& input)
|
||||
{
|
||||
@@ -195,6 +274,9 @@ wxString wxEncodingConverter::Convert(const wxString& input)
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
// Following tables describe classes of encoding equivalence.
|
||||
//
|
||||
|
||||
@@ -208,7 +290,9 @@ wxString wxEncodingConverter::Convert(const wxString& input)
|
||||
static wxFontEncoding
|
||||
EquivalentEncodings[][NUM_OF_PLATFORMS][ENC_PER_PLATFORM+1] = {
|
||||
|
||||
// West European (Latin1)
|
||||
// *** Please put more common encodings as first! ***
|
||||
|
||||
// West European
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_1, wxFONTENCODING_ISO8859_15, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1252, STOP},
|
||||
@@ -216,13 +300,61 @@ static wxFontEncoding
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Central European (Latin2)
|
||||
// Central European
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_2, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1250, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Baltic
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_13, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1257, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Hebrew
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_8, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1255, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Greek
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_7, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1253, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Arabic
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_6, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1256, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Turkish
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_9, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1254, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
// Cyrillic
|
||||
{
|
||||
/* unix */ {wxFONTENCODING_ISO8859_5, STOP},
|
||||
/* windows */ {wxFONTENCODING_CP1251, STOP},
|
||||
/* os2 */ {STOP},
|
||||
/* mac */ {STOP}
|
||||
},
|
||||
|
||||
{{STOP},{STOP},{STOP},{STOP}} /* Terminator */
|
||||
/* no, _not_ Arnold! */
|
||||
@@ -257,7 +389,9 @@ wxFontEncodingArray wxEncodingConverter::GetPlatformEquivalents(wxFontEncoding e
|
||||
for (e = 0; EquivalentEncodings[clas][i][e] != STOP; e++)
|
||||
if (EquivalentEncodings[clas][i][e] == enc)
|
||||
{
|
||||
for (f = EquivalentEncodings[clas][platform]; *f != STOP; f++)
|
||||
for (f = EquivalentEncodings[clas][platform]; *f != STOP; f++)
|
||||
if (*f == enc) arr.Add(enc);
|
||||
for (f = EquivalentEncodings[clas][platform]; *f != STOP; f++)
|
||||
if (arr.Index(*f) == wxNOT_FOUND) arr.Add(*f);
|
||||
i = NUM_OF_PLATFORMS/*hack*/; break;
|
||||
}
|
||||
@@ -274,6 +408,8 @@ wxFontEncodingArray wxEncodingConverter::GetAllEquivalents(wxFontEncoding enc)
|
||||
int i, clas, e, j ;
|
||||
wxFontEncoding *f;
|
||||
wxFontEncodingArray arr;
|
||||
|
||||
arr = GetPlatformEquivalents(enc); // we want them to be first items in array
|
||||
|
||||
clas = 0;
|
||||
while (EquivalentEncodings[clas][0][0] != STOP)
|
||||
|
Reference in New Issue
Block a user