Getting std :: ifstream to handle LF, CR, and CRLF

Question

Specifically I'm interested in istream& getline ( istream& is, string& str );. Is there an option to the ifstream constructor to tell it to convert all newline encodings to '\n' under the hood? I want to be able to call getline and have it gracefully handle all line endings.

Update: To clarify, I want to be able to write code that compiles almost anywhere, and will take input from almost anywhere. Including the rare files that have '\r' without '\n'. Minimizing inconvenience for any users of the software.

It's easy to workaround the issue, but I'm still curious as to the right way, in the standard, to flexibly handle all text file formats.

getline reads in a full line, up to a '\n', into a string. The '\n' is consumed from the stream, but getline doesn't include it in the string. That's fine so far, but there might be a '\r' just before the '\n' that gets included into the string.

There are three types of line endings seen in text files: '\n' is the conventional ending on Unix machines, '\r' was (I think) used on old Mac operating systems, and Windows uses a pair, '\r' following by '\n'.

The problem is that getline leaves the '\r' on the end of the string.

ifstream f("a_text_file_of_unknown_origin");
string line;
getline(f, line);
if(!f.fail()) { // a non-empty line was read
   // BUT, there might be an '\r' at the end now.
}

Edit Thanks to Neil for pointing out that f.good() isn't what I wanted. !f.fail() is what I want.

I can remove it manually myself (see edit of this question), which is easy for the Windows text files. But I'm worried that somebody will feed in a file containing only '\r'. In that case, I presume getline will consume the whole file, thinking that it is a single line!

.. and that's not even considering Unicode :-)

.. maybe Boost has a nice way to consume one line at a time from any text-file type?

Edit I'm using this, to handle the Windows files, but I still feel I shouldn't have to! And this won't fork for the '\r'-only files.

if(!line.empty() && *line.rbegin() == '\r') {
    line.erase( line.length()-1, 1);
}

Answer 1

The C++ runtime should deal correctly with whatever the endline convention is for your particular platform. Specifically, this code should work on all platforms:

#include <string>
#include <iostream>
using namespace std;

int main() {
    string line;
    while( getline( cin, line ) ) {
        cout << line << endl;
    }
}

Of course, if you are dealing with files from another platform, all bets are off.

As the two most common platforms (Linux and Windows) both terminate lines with a newline character, with Windows preceding it with a carriage return,, you can examine the last character of the line string in the above code to see if it is \r and if so remove it before doing your application-specific processing.

For example, you could provide yourself with a getline style function that looks something like this (not tested, use of indexes, substr etc for pedagogical purposes only):

ostream & safegetline( ostream & os, string & line ) {
    string myline;
    if ( getline( os, myline ) ) {
       if ( myline.size() && myline[myline.size()-1] == '\r' ) {
           line = myline.substr( 0, myline.size() - 1 );
       }
       else {
           line = myline;
       }
    }
    return os;
}

Answer 2

As Neil pointed out, "the C++ runtime should deal correctly with whatever the line ending convention is for your particular platform."

However, people do move text files between different platforms, so that is not good enough. Here is a function that handles all three line endings ("\r", "\n" and "\r\n"):

std::istream& safeGetline(std::istream& is, std::string& t)
{
    t.clear();

    // The characters in the stream are read one-by-one using a std::streambuf.
    // That is faster than reading them one-by-one using the std::istream.
    // Code that uses streambuf this way must be guarded by a sentry object.
    // The sentry object performs various tasks,
    // such as thread synchronization and updating the stream state.

    std::istream::sentry se(is, true);
    std::streambuf* sb = is.rdbuf();

    for(;;) {
        int c = sb->sbumpc();
        switch (c) {
        case '\n':
            return is;
        case '\r':
            if(sb->sgetc() == '\n')
                sb->sbumpc();
            return is;
        case std::streambuf::traits_type::eof():
            // Also handle the case when the last line has no line ending
            if(t.empty())
                is.setstate(std::ios::eofbit);
            return is;
        default:
            t += (char)c;
        }
    }
}

And here is a test program:

int main()
{
    std::string path = ...  // insert path to test file here

    std::ifstream ifs(path.c_str());
    if(!ifs) {
        std::cout << "Failed to open the file." << std::endl;
        return EXIT_FAILURE;
    }

    int n = 0;
    std::string t;
    while(!safeGetline(ifs, t).eof())
        ++n;
    std::cout << "The file contains " << n << " lines." << std::endl;
    return EXIT_SUCCESS;
}

Answer 3

Are you reading the file in BINARY or in TEXT mode? In TEXT mode the pair carriage return/line feed, CRLF, is interpreted as TEXT end of line, or end of line character, but in BINARY you fetch only ONE byte at a time, which means that either character MUST be ignored and left in the buffer to be fetched as another byte! Carriage return means, in the typewriter, that the typewriter car, where the printing arm lies in, has reached the right edge of the paper and is returned to the left edge. This is a very mechanical model, that of the mechanical typewriter. Then the line feed means that the paper roll is rotated a little bit up so the paper is in position to begin another line of typing. As fas as I remember one of the low digits in ASCII means move to the right one character without typing, the dead char, and of course \b means backspace: move the car one character back. That way you can add special effects, like underlying (type underscore), strikethrough (type minus), approximate different accents, cancel out (type X), without needing an extended keyboard, just by adjusting the position of the car along the line before entering the line feed. So you can use byte sized ASCII voltages to automatically control a typewriter without a computer in between. When the automatic typewriter is introduced, AUTOMATIC means that once you reach the farthest edge of the paper, the car is returned to the left AND the line feed applied, that is, the car is assumed to be returned automatically as the roll moves up! So you do not need both control characters, only one, the \n, new line, or line feed.

This has nothing to do with programming but ASCII is older and HEY! looks like some people were not thinking when they begun doing text things! The UNIX platform assumes an electrical automatic typemachine; the Windows model is more complete and allows for control of mechanical machines, though some control characters become less and less useful in computers, like the bell character, 0x07 if I remember well... Some forgotten texts must have been originally captured with control characters for electrically controlled typewriters and it perpetuated the model...

Actually the correct variation would be to just include the \r, line feed, the carriage return being unnecessary, that is, automatic, hence:

char c;
ifstream is;
is.open("",ios::binary);
...
is.getline(buffer, bufsize, '\r');

//ignore following \n or restore the buffer data
if ((c=is.get())!='\n') is.rdbuf()->sputbackc(c);
...

would be the most correct way to handle all types of files. Note however that \n in TEXT mode is actually the byte pair 0x0d 0x0a, but 0x0d IS just \r: \n includes \r in TEXT mode but not in BINARY, so \n and \r\n are equivalent... or should be. This is a very basic industry confusion actually, typical industry inertia, as the convention is to speak of CRLF, in ALL platforms, then fall into different binary interpretations. Strictly speaking, files including ONLY 0x0d (carriage return) as being \n (CRLF or line feed), are malformed in TEXT mode (typewritter machine: just return the car and strikethrough everything...), and are a non-line oriented binary format (either \r or \r\n meaning line oriented) so you are not supposed to read as text! The code ought to fail maybe with some user message. This does not depend on the OS only, but also on the C library implementation, adding to the confusion and possible variations... (particularly for transparent UNICODE translation layers adding another point of articulation for confusing variations).

The problem with the previous code snippet (mechanical typewriter) is that it is very inefficient if there are no \n characters after \r (automatic typewriter text). Then it also assumes BINARY mode where the C library is forced to ignore text interpretations (locale) and give away the sheer bytes. There should be no difference in the actual text characters between both modes, only in the control characters, so generally speaking reading BINARY is better than TEXT mode. This solution is efficient for BINARY mode typical Windows OS text files independently of C library variations, and inefficient for other platform text formats (including web translations into text). If you care about efficiency, the way to go is to use a function pointer, make a test for \r vs \r\n line controls however way you like, then select the best getline user-code into the pointer and invoke it from it.

Incidentally I remember I found some \r\r\n text files too... which translates into double line text just as is still required by some printed text consumers.

Answer 4

One solution would be to first search and replace all line endings to '\n' - just like e.g. Git does by default.

Answer 5

Other than writing your own custom handler or using an external library, you are out of luck. The easiest thing to do is to check to make sure line[line.length() - 1] is not '\r'. On Linux, this is superfluous as most lines will end up with '\n', meaning you'll lose a fair bit of time if this is in a loop. On Windows, this is also superfluous. However, what about classic Mac files which end in '\r'? std::getline would not work for those files on Linux or Windows because '\n' and '\r' '\n' both end with '\n', eliminating the need to check for '\r'. Obviously such a task that works with those files would not work well. Of course, then there exist the numerous EBCDIC systems, something that most libraries won't dare tackle.

Checking for '\r' is probably the best solution to your problem. Reading in binary mode would allow you to check for all three common line endings ('\r', '\r\n' and '\n'). If you only care about Linux and Windows as old-style Mac line endings shouldn't be around for much longer, check for '\n' only and remove the trailing '\r' character.