wxASSERT( nStart <= length() );
//anchor
- const wxChar* p = (const wxChar*)wxMemchr(c_str() + nStart,
- str.c_str()[0],
+ const wxChar* p = (const wxChar*)wxMemchr(c_str() + nStart,
+ str.c_str()[0],
length() - nStart);
-
+
if(!p)
return npos;
wxMemcmp(p, str.c_str(), str.length()) )
{
//anchor again
- p = (const wxChar*)wxMemchr(++p,
- str.c_str()[0],
+ p = (const wxChar*)wxMemchr(++p,
+ str.c_str()[0],
length() - (p - c_str()));
if(!p)
return npos;
}
-
+
return (p - c_str() + str.length() <= length()) ? p - c_str() : npos;
}
}
size_t len = wxStrlen(sz);
-
+
for ( const wxChar *p = c_str() + nStart; p >= c_str(); --p )
{
if ( wxMemchr(sz, *p, len) )
wxStringBase& wxStringBase::replace(size_t nStart, size_t nLen,
size_t nCount, wxChar ch)
{
- return replace(nStart, nLen, wxStringBase(ch, nCount).c_str());
+ return replace(nStart, nLen, wxStringBase(nCount, ch).c_str());
}
wxStringBase& wxStringBase::replace(size_t nStart, size_t nLen,
//Convert a wide character string of a specified length
//to a multi-byte character string, ignoring intermittent null characters
//returns the actual length of the string
-inline size_t wxMbstr(wxCharBuffer& buffer, const wchar_t* szString,
+inline size_t wxMbstr(wxCharBuffer& buffer, const wchar_t* szString,
size_t nStringLen, wxMBConv& conv)
{
const wchar_t* szEnd = szString + nStringLen + 1;
const wchar_t* szPos = szString;
const wchar_t* szStart = szPos;
-
+
size_t nActualLength = 0;
-
+
//Convert the string until the length() is reached, continuing the
//loop every time a null character is reached
while(szPos != szEnd)
//Get the length of the current (sub)string
size_t nLen = conv.WC2MB(NULL, szPos, 0);
-
+
wxASSERT(nLen != (size_t)-1); //should not be true! If it is system wctomb could be bad
-
+
nActualLength += nLen + 1;
-
+
wxASSERT(nActualLength <= (nStringLen<<1) + 1); //If this is true it means buffer overflow
-
+
//Convert the current (sub)string
if ( conv.WC2MB(&buffer.data()[szPos - szStart], szPos, nLen + 1) == (size_t)-1 )
{
wxFAIL_MSG(wxT("Error converting wide-character string to a multi-byte string"));
buffer.data()[0] = '\0';
return 0;
- }
-
+ }
+
//Increment to next (sub)string
//Note that we have to use wxWcslen here instead of nLen
//here because XX2XX gives us the size of the output buffer,
//not neccessarly the length of the string
szPos += wxWcslen(szPos) + 1;
}
-
+
return nActualLength - 1; //success - return actual length
}
//Convert a multi-byte character string of a specified length
//to a wide character string, ignoring intermittent null characters
-//returns the actual length
-inline size_t wxWcstr( wxWCharBuffer& buffer, const char* szString,
- size_t nStringLen, wxMBConv& conv)
+//returns the actual length
+inline size_t wxWcstr( wxWCharBuffer& buffer, const char* szString,
+ size_t nStringLen, wxMBConv& conv)
{
const char* szEnd = szString + nStringLen + 1;
const char* szPos = szString;
const char* szStart = szPos;
size_t nActualLength = 0;
-
+
//Convert the string until the length() is reached, continuing the
//loop every time a null character is reached
while(szPos != szEnd)
//Get the length of the current (sub)string
size_t nLen = conv.MB2WC(NULL, szPos, 0);
-
+
wxASSERT(nLen != (size_t)-1); //should not be true! If it is system mbtowc could be bad
-
+
nActualLength += nLen + 1;
-
+
wxASSERT(nActualLength <= nStringLen + 1); //If this is true it means buffer overflow
//Convert the current (sub)string
wxFAIL_MSG(wxT("Error converting multi-byte string to a wide-character string"));
buffer.data()[0] = '\0';
return 0;
- }
-
+ }
+
//Increment to next (sub)string
//Note that we have to use strlen here instead of nLen
//here because XX2XX gives us the size of the output buffer,
//not neccessarly the length of the string
szPos += strlen(szPos) + 1;
}
-
+
return nActualLength - 1; //success - return actual length
}
//Convert the string
size_t nActualLength = wxWcstr(buffer, psz, nLen, conv);
-
+
if ( !Alloc(nActualLength + 1) )
{
wxFAIL_MSG(wxT("Out of memory in wxString"));
assign(buffer.data(), nActualLength);
}
}
-}
+}
//Convert wxString in Unicode mode to a multi-byte string
const wxCharBuffer wxString::mb_str(wxMBConv& conv) const
//Do the actual conversion (will return a blank string on error)
wxMbstr(buffer, (*this).c_str(), length(), conv);
-
+
return buffer;
}
{
//*2 is the worst case - probably for UTF8
wxCharBuffer buffer((nLen << 1) + 1);
-
+
//do the actual conversion (if it fails we get an empty string)
size_t nActualLength = wxMbstr(buffer, pwz, nLen, conv);
-
+
if ( !Alloc(nActualLength + 1) )
{
wxFAIL_MSG(wxT("Out of memory in wxString"));
}
}
-//Converts this string to a wide character string if unicode
+//Converts this string to a wide character string if unicode
//mode is not enabled and wxUSE_WCHAR_T is enabled
const wxWCharBuffer wxString::wc_str(wxMBConv& conv) const
{
//mb->wc never inflates to more than the length
wxWCharBuffer buffer(length() + 1);
-
+
//Do the actual conversion (will return a blank string on error)
wxWcstr(buffer, (*this).c_str(), length(), conv);
-
+
return buffer;
}
-
+
#endif // wxUSE_WCHAR_T
#endif // Unicode/ANSI