How to translate scanf exact matching into modern c++ stringstream reading

Question

I am currenlty working on a project and I'd like to use modern cpp instead of relying on old c for reading files. For context I'm trying to read wavefront obj files.

I have this old code snippet :

const char *line;
float x, y, z;
if(sscanf(line, "vn %f %f %f", &x, &y, &z) != 3)
    break; // quitting loop because couldn't scan line correctly

which I've translated into :

string line;
string word;
stringstream ss(line);
float x, y, z;
if (!(ss >> word >> x >> y >> z)) // "vn x y z"
    break; // quitting loop because couldn't scan line correctly

The difference is that I use a string to skip the first word but I'd like it to match "vn" the same as sscanf does. Is this possible with stringstream or should I keep relying on sscanf for exact pattern matching ?

Also I'm trying to translate

sscanf(line, " %d/%d/%d", &i1, &i2, &i3);

but I'm having a hard time which again orients me towards not using modern cpp for my file reader.

Answer 1

I'd argue that stringstream is not "modern C++"¹. (I'll also admit we don't have something as a good replacement for scanf; we do have std::format that is very nice, and replaces std::cout << shenanigans with a syntax reminescent of Python's format strings, and is much faster. Sadly, it hasn't hit standard libraries yet.)

I'll actually say that I think your sscanf code is cleaner, and saner than the stringstream code: sscanf leaves things in an easier-to-understand state when parsing the line fails.

There's libraries you can use to build a line parser; Boost::spirit::qi is probably the most well-known. You can do things like

auto success = (ss >> qi::phrase_match("vn " >> qi::double >> ' ' >> qi::double >> ' ' >> qi::double, x, y, z)); 

If this option does not fill you with joy, love for the world and faith in all-reaching beauty, you're exactly like me and find it a poor replacement for the compactness and ease of understanding the original scanf offers.

There's the std::regex library these days (since C++11), and that might very well be what you're looking for! However, you'd need to write expressions for "here comes a floating point number" yourself, and that is not cool at all, either.

---

¹ without wanting to hurt anyone, I think the iostreams library approach was the one single standard library feature that made C++ awkwarder to use than most contemporary languages for anything IO-related. And that has a lasting effect on code quality.

Answer 2

For anyone interested here is how I parsed my obj file with stringstream style instead of sscanf style

Old code:

for(;;)
{
    if(fgets(line_buffer, sizeof(line_buffer), in) == NULL)
    {
        error= false; // eof
        break;
    }
        
    // force endl
    line_buffer[sizeof(line_buffer) -1]= 0;
        
    // skip spaces
    char *line= line_buffer;
    while(*line && isspace(*line))
        line++;
        
    if(line[0] == 'v')
    {
        float x, y, z;
        if(line[1] == ' ')          // position x y z
        {
            if(sscanf(line, "v %f %f %f", &x, &y, &z) != 3)
                break;
            positions.push_back( vec3(x, y, z) );
        }
        else if(line[1] == 'n')     // normal x y z
        {
            if(sscanf(line, "vn %f %f %f", &x, &y, &z) != 3)
                break;
            normals.push_back( vec3(x, y, z) );
        }
        else if(line[1] == 't')     // texcoord x y
        {
            if(sscanf(line, "vt %f %f", &x, &y) != 2)
                break;
            texcoords.push_back( vec2(x, y) );
        }
    }
        
    else if(line[0] == 'f')
    {
        idp.clear();
        idt.clear();
        idn.clear();
            
        int next;
        for(line= line +1; ; line= line + next)
        {
            idp.push_back(0); 
            idt.push_back(0); 
            idn.push_back(0);         // 0: invalid index
                
            next= 0;
            if(sscanf(line, " %d/%d/%d %n", &idp.back(), &idt.back(), &idn.back(), &next) == 3) 
                continue;
            else if(sscanf(line, " %d/%d %n", &idp.back(), &idt.back(), &next) == 2)
                continue;
            else if(sscanf(line, " %d//%d %n", &idp.back(), &idn.back(), &next) == 2)
                continue;
            else if(sscanf(line, " %d %n", &idp.back(), &next) == 1)
                continue;
            else if(next == 0)      //endl
                break;
        }
}

New code: (with the @JerryCoffin code for std::istream& operator>>(std::istream& is, char const* s);)

std::string line;
while (getline(file, line))
{
    std::stringstream ss(line);
    std::string tag;

    // istream& operator>> skips whitespaces unless std::skipws is disable
    // ignore empty lines
    if (!(ss >> tag))
        continue;

    // ignore comments
    if (tag[0] == '#')
        continue;

    float x, y, z;
    if (tag == "v")
    {
        if (!(ss >> x >> y >> z))
            break;
        positions_tmp.emplace_back(x, y, z);
    }
    else if (tag == "vt")
    {
        if (!(ss >> x >> y))
            break;
        texcoords_tmp.emplace_back(x, y);
    }
    else if (tag == "vn")
    {
        if (!(ss >> x >> y >> z))
            break;
        normals_tmp.emplace_back(x, y, z);
    }
    else if (tag == "f")
    {
        for (int i = 0; i < 3; ++i)
        {
            unsigned int idp(0), idt(0), idn(0);

            // reset stringstream after each read so that it keeps current position when fail
            auto pos = ss.tellg();
            auto state = ss.rdstate() && ~std::ios_base::failbit;
            auto reset = [state, pos](std::stringstream& ss)
            { ss.clear(state); ss.seekg(pos); return !ss.fail(); };

            // will try to match either i//i or i/i/i or i/i or i (all .obj "f" configurations)
            if (reset(ss) && ss >> idp >> "//" >> idn) {}
            else if (reset(ss) && ss >> idp >> "/" >> idt >> "/" >> idn) {}
            else if (reset(ss) && ss >> idp >> "/" >> idt) {}
            else if (reset(ss) && ss >> idp) {}

            // do stuff with indexes
            // ...
        }
    }
}

I changed a bit of logic here and there but the trickiest part was on the f i/i/i i/i/i i/i/i lines where you have to check if it matches right. If it doesn't you have to reset the stringstream to it's original state and position (at least remove std::ios_base::failbit and and replace position).

I wanted to keep the aspect of one line = one case, so it might be a little harder to understand with the lambda function, but I could also reverse the logic and handle faces edge by edge (not sure if I'll use it though).