Can I overload CArchive << operator to work with std::string?

Question

I am using std::string in my MFC application and I want to store it in doc's Serialize() function. I don't want to store them as CString because it writes its own stuff in there and my goal is to create a file that I know the format of and can be read by other application without needing CString. So I would like to store my std::strings as 4 bytes (int) string length followed by buffer of that size containing the string.

void CMyDoc::Serialize(CArchive& ar)
{
    std::string theString;

    if (ar.IsStoring())
    {
        // TODO: add storing code here
        int size = theString.size();
        ar << size;
        ar.Write( theString.c_str(), size );

    }
    else
    {
        // TODO: add loading code here
        int size = 0;
        ar >> size;
        char * bfr = new char[ size ];
        ar.Read( bfr, size);
        theString = bfr;
        delete [] bfr;
    }
}

The above code is not great and I have to allocate a temp bfr to read the string. First can I read the string directly into std::string without the temp buffer? Secondly can I overload the << buffer for std::string / CArchive so I can simply use ar << theString? Overall is there a better way to read/write std::string using CArchive object?

Conciliar

Answer 1

You could build an inplace CString from your stl string and serialize that. Something like:

CString c_string(my_stl_string.c_str();
ar << c_string;

You could put this in a global operater overload so it can you can just

ar << my_c_string;

from anywhere eg:

CArchive& operator<<(CArchive rhs, string lhs) {
    CString c_string(lhs.c_str());
    rhs << c_string;
}

Fassold

Answer 2

Try:

theString.resize(size);
ar.Read(&theString[0], size);

Technically &theString[0] is not guaranteed to point to a contiguous character buffer, but the C++ committee did a survey and found that all existing implementations work this way.

Alfaro

Answer 3

Its probably better to write the data as a CString for various reasons, but if you have to convert your String (m_sString) into an ASCII character string, maybe something like this will work for you...

void myclass::Serialize(CArchive & ar)
{
    CHAR* buf;
    DWORD len;
    if (ar.IsStoring()) // Writing
    {
        len = m_sString.GetLength(); // Instead of null terminated string, store size.
        ar << len;
        buf = (CHAR*)malloc(len);
        WideCharToMultiByte(CP_UTF8, 0, m_sString, len, buf, len, NULL, NULL); // Convert wide to single bytes
        ar.Write(buf, len); // Write ascii chars
        free(buf);
    }
    else // Reading
    {
        ar >> len;
        buf = (CHAR*)malloc(len);
        ar.Read(buf, len); // Read ascii string
        MultiByteToWideChar(CP_UTF8, 0, buf, len, m_sString.GetBufferSetLength(len), len); // Convert ascii bytes to CString wide bytes
        free(buf);
    }
}

Barrister

Answer 4

Just to add a fully functional and correct example (hopefully) for both std::string and std::wstring:

#include <string>
#include <gsl/gsl>

template <typename Char>
CArchive& operator<<(CArchive& ar, const std::basic_string<Char>& rhs)
{
    const auto size = rhs.size();
    ar << size;
    ar.Write(rhs.data(), gsl::narrow_cast<UINT>(size) * sizeof(Char));
    return ar;
}

template <typename Char>
CArchive& operator>>(CArchive& ar, std::basic_string<Char>& rhs)
{
    size_t size{};
    ar >> size;
    rhs.resize(size);
    ar.Read(rhs.data(), gsl::narrow_cast<UINT>(size) * sizeof(Char));
    return ar;
}

Writing ...

std::wstring ws{ L"wide string" };
ar << ws;

std::string ns{ L"narrow string" };
ar << ns;

Reading ...

std::wstring ws;
ar >> ws;

std::string ns;
ar >> ns;

Candlewood

Answer 5

If you are working with a library that only works with c-style strings, there is no way to safely write directly to the std::string. That issue is fixed in C++0x. So something like

// NOT PORTABLE, don't do this
theString.resize(size);
ar.Read( const_cast<char *>(theString.c_str(), size);

Would probably work, but it could create some subtle, hard-to-track bugs later on. Of course your question implies that you have profiled your code and figured out that creating the buffer and copying the data twice is actually a bottleneck in your code. If you haven't, then you shouldn't be fretting about inefficiencies yet.

Olimpia

I have tried this but c_str() returns a 'const char *' and that's a problem. I probably could typecast it to simply 'char *' but that would somewhat violating the c_str() function. – Conciliar

Thus the "don't do this" comment. You can use a std::vector<char> or since you're in MFC land use CString – Libenson

Answer 6

I suppose you could violate STL guidelines and inherit std::string and add your own buffer getter/setter. Then override the copy constructor for std::string and transfer ownership of the buffer.

Monger