wxLoadCharacterSets() now appears to work (was a bit slow, but since it's

not called until someone actually uses wxConv_local, it should not affect
too many).


git-svn-id: https://svn.wxwidgets.org/svn/wx/wxWidgets/trunk@2191 c3d73ce0-8a6f-49c7-b76d-6d57e0e08775
This commit is contained in:
Ove Kaaven 1999-04-16 09:58:38 +00:00
parent ba0730de65
commit c71c744a57

View File

@ -1937,27 +1937,25 @@ static void wxLoadCharacterSets(void)
if (already_loaded) return;
#if defined(__UNIX__) && wxUSE_UNICODE
#if defined(__UNIX__)
// search through files in /usr/share/i18n/charmaps
wxString fname;
printf("Commencing load\n");
for (fname = ::wxFindFirstFile(_T("/usr/share/i18n/charmaps/*"));
!fname.IsEmpty();
fname = ::wxFindNextFile()) {
wxPrintf(_("Loading: %s\n"), fname.c_str());
wxTextFile cmap(fname);
if (cmap.Open()) {
wxCharacterSet *cset = new wxCharacterSet;
wxString comchar,escchar;
bool in_charset = FALSE;
wxPrintf(_T("yup, loaded %s\n"),fname.c_str());
// wxFprintf(stderr,_T("Loaded: %s\n"),fname.c_str());
wxString line;
for (line = cmap.GetFirstLine();
!cmap.Eof();
line = cmap.GetNextLine()) {
wxPrintf(_T("line contents: %s\n"),line.c_str());
// wxFprintf(stderr,_T("line contents: %s\n"),line.c_str());
wxStringTokenizer token(line);
wxString cmd = token.GetNextToken();
if (cmd == comchar) {
@ -1974,12 +1972,13 @@ static void wxLoadCharacterSets(void)
comchar = token.GetNextToken();
else if (cmd == _T("<escape_char>"))
escchar = token.GetNextToken();
else if (cmd == _T("<mb_cur_min")) {
else if (cmd == _T("<mb_cur_min>")) {
delete cset;
goto forget_it; // we don't support multibyte charsets ourselves (yet)
cset = (wxCharacterSet *) NULL;
break; // we don't support multibyte charsets ourselves (yet)
}
else if (cmd == _T("CHARMAP")) {
cset->data = (wchar_t *)calloc(256, sizeof(wxChar));
cset->data = (wchar_t *)calloc(256, sizeof(wchar_t));
in_charset = TRUE;
}
else if (cmd == _T("END")) {
@ -1988,19 +1987,31 @@ static void wxLoadCharacterSets(void)
}
else if (in_charset) {
// format: <NUL> /x00 <U0000> NULL (NUL)
// <A> /x41 <U0041> LATIN CAPITAL LETTER A
wxString hex = token.GetNextToken();
// skip whitespace (why doesn't wxStringTokenizer do this?)
while (wxIsEmpty(hex) && token.HasMoreTokens()) hex = token.GetNextToken();
wxString uni = token.GetNextToken();
// just assume that we've got the right format
int pos = ::wxHexToDec(hex.Mid(2,2));
unsigned long uni1 = ::wxHexToDec(uni.Mid(2,2));
unsigned long uni2 = ::wxHexToDec(uni.Mid(4,2));
cset->data[pos] = (uni1 << 16) | uni2;
// skip whitespace again
while (wxIsEmpty(uni) && token.HasMoreTokens()) uni = token.GetNextToken();
if ((hex.GetChar(0) == escchar) && (hex.GetChar(1) == _T('x')) &&
(uni.Left(2) == _T("<U"))) {
hex.MakeUpper(); uni.MakeUpper();
int pos = ::wxHexToDec(hex.Mid(2,2));
if (pos>=0) {
unsigned long uni1 = ::wxHexToDec(uni.Mid(2,2));
unsigned long uni2 = ::wxHexToDec(uni.Mid(4,2));
cset->data[pos] = (uni1 << 16) | uni2;
// wxFprintf(stderr,_T("char %02x mapped to %04x (%c)\n"),pos,cset->data[pos],cset->data[pos]);
}
}
}
}
cset->names.Shrink();
wxCharsets.Add(cset);
forget_it:
continue;
if (cset) {
cset->names.Shrink();
wxCharsets.Add(cset);
}
}
}
#endif