In a previous blog post, I listed some advantages (IMHO) of using CString instead of std::string class.
There is another subtle (but important!) point which turns the scales in CString‘s favour: in fact, CString is a C++ class that wraps zero-terminated strings (i.e. the good old C strings), instead it seems that the C++ standard library code wants to pretend that zero-terminated strings do not exist…
Let me show this with a sample code (the full C++ source code is attached to this blog post):
// Check with CString
WCHAR * psz = str.GetBuffer(100);
psz = 0; // …or GetWindowText(hwnd, psz, 100)
psz = NULL; // psz is no more valid
assert( wcslen(str) == str.GetLength() ); // OK
// Check with std::string
s = 0; // …or GetWindowText(hwnd, &s, 100)
assert( wcslen(s.c_str()) == s.length() ); // BANG!
So, basically, we have a situation in which (assuming ‘s‘ is an instance of std::wstring) s.length() and wcslen(s.c_str()) do not return the same value.
Considering that Windows APIs (like say GetWindowText) work with zero-terminated strings, CString seems to me a clear winner as a convenient string class for C++ Win32 programming.