Reading file into buffer and avoiding splitting lines between reads

Question

I was reading sehe's answer for fast text file reading in C++, which looks like this.

static uintmax_t wc(char const *fname)
{
    static const auto BUFFER_SIZE = 16*1024;
    int fd = open(fname, O_RDONLY);
    if(fd == -1)
        handle_error("open");

    /* Advise the kernel of our access pattern.  */
    posix_fadvise(fd, 0, 0, 1);  // FDADVICE_SEQUENTIAL

    char buf[BUFFER_SIZE + 1];
    uintmax_t lines = 0;

    while(size_t bytes_read = read(fd, buf, BUFFER_SIZE))
    {
        if(bytes_read == (size_t)-1)
            handle_error("read failed");
        if (!bytes_read)
            break;

        for(char *p = buf; (p = (char*) memchr(p, '\n', (buf + bytes_read) - p)); ++p)
            ++lines;
    }

    return lines;
}

This is cool, but I was wondering if a similar approach can be taken when we aren't dealing with a character operation like counting newlines, but want to operate on each line of data. Say for instance I had a file of doubles, and already some function parse_line_to_double to use on each line.

12.44243
4242.910
...

That is, how can I read BUFFER_SIZE bytes into my buffer but avoid splitting the last line read? Effectively, can I ask "Give me BUFFER_SIZE or less bytes while ensuring that the last byte read is a newline character (or EOF)"?

Knowing extremely little about low level IO like this, ideas that came to mind were

• Can I "back up" fd to the most recent newline between iterations?
• Do I have to keep a second buffer holding a copy of the current line being read all the time?

Answer 1

Here is a comparison test. First, lets try the easy way. Just read the file with standard C++ functions:

#include <iostream>
#include <string>  
#include <fstream> //std::ifstream
#include <sstream> //std::stringstream

uintmax_t test1(char const *fname)
{
    std::ifstream fin(fname);
    if(!fin) return 0;
    uintmax_t lines = 0;
    std::string str;
    double value;
    while(fin >> value)
    {
        //std::cout << value << "\n";
        lines++;
    }
    return lines;
}

Next, with std::stringstream this is about 2.5 times faster:

uintmax_t test2(char const *fname)
{
    std::ifstream fin(fname);
    if(!fin) return 0;
    uintmax_t lines = 0;
    std::string str;
    double value;
    std::stringstream ss;
    ss << fin.rdbuf();
    while(ss >> value)
        lines++;
    return lines;
}

Next, lets read the whole file in to memory. This will be fine as long as the file is less than 1 GiB or so. Assuming there is a double value on each line, lets extract that value. test3 is more complicated and less flexible, and it's not any faster than test2:

uintmax_t test3(char const *fname)
{
    std::ifstream fin(fname, std::ios::binary);
    if(!fin) return 0;

    fin.seekg(0, std::ios::end);
    size_t filesize = (size_t)fin.tellg();
    fin.seekg(0);
    std::string str(filesize, 0);
    fin.read(&str[0], filesize);

    double value;
    uintmax_t lines = 0;
    size_t beg = 0;
    size_t i;
    size_t len = str.size();
    for(i = 0; i < len; i++)
    {
        if(str[i] == '\n' || i == len - 1)
        {
            try
            {
                value = std::stod(str.substr(beg, i - beg));
                //std::cout << value << "\n";
                beg = i + 1;
                lines++;
            }
            catch(...)
            {
            }
        }
    }
    return lines;
}

For comparison to the wc function in the question, let's read the whole file in to memory and only count the number of lines. This runs a little faster than wc (as expected), suggesting that there is no need for additional optimizations

uintmax_t test_countlines(char const *fname)
{
    std::ifstream fin(fname, std::ios::binary);
    if(!fin) return 0;
    fin.seekg(0, std::ios::end);
    size_t filesize = (size_t)fin.tellg();
    fin.seekg(0);
    std::string str(filesize, 0);
    fin.read(&str[0], filesize);
    uintmax_t lines = 0;
    for(auto &c : str)
        if(c == '\n')
            lines++;
    return lines;
}