Why is Pythons I/O so much slower than C++?

Question

I have created some code for testing in Python and C++, where I read two matrices from a file and print something. It seems as if Python needs about twice as much time for I/O:

$ ./test.sh -i Testing/2000.in -p "C++/read-write-only.out" -n 2
Executing: C++/read-write-only.out -i Testing/2000.in > TMPcurrentFileResult.out
It took 8 seconds for 2 executions
MIN: 4 seconds
MAX: 4 seconds

$ ./test.sh -i Testing/2000.in -p "python Python/read-write-only.py" -n 2
Executing: python Python/read-write-only.py -i Testing/2000.in > TMP..Results.out
It took 16 seconds for 2 executions
MIN: 8 seconds
MAX: 8 seconds

This is the code I've used for Python:

#!/usr/bin/python
# -*- coding: utf-8 -*-

from optparse import OptionParser
parser = OptionParser()
parser.add_option("-i", dest="filename", default="bigMatrix.in",
     help="input file with two matrices", metavar="FILE")
(options, args) = parser.parse_args()

def read(filename):
    lines = open(filename, 'r').read().splitlines()
    A = []
    B = []
    matrix = A
    for line in lines:
        if line != "":
            matrix.append(map(int, line.split("\t")))
        else:
            matrix = B
    return A, B

def printMatrix(matrix):
    for line in matrix:
        print "\t".join(map(str,line))

A, B = read(options.filename)
# Do something
printMatrix(B)

This is the C++-Code

#include <sstream>
#include <string>
#include <fstream>
#include <iostream>
#include <vector>
#include <algorithm>

using namespace std;

int getMatrixSize(string filename) {
    string line;
    ifstream infile;
    infile.open (filename.c_str());
    getline(infile, line);
    return count(line.begin(), line.end(), '\t') + 1;
}

void read(string filename, vector< vector<int> > &A, vector< vector<int> > &B){
    string line;
    FILE* matrixfile = freopen(filename.c_str(), "r", stdin);

    int i = 0, j, a;
    while (getline(cin, line) && !line.empty()) {
        istringstream iss(line);
        j = 0;
        while (iss >> a) {
            A[i][j] = a;
            j++;
        }
        i++;
    }

    i = 0;
    while (getline(cin, line)) {
        istringstream iss(line);
        j = 0;
        while (iss >> a) {
            B[i][j] = a;
            j++;
        }
        i++;
    }

    fclose (matrixfile);
}

void printMatrix(vector< vector<int> > matrix, int n) {
    for (int i=0; i < n; i++) {
        for (int j=0; j < n; j++) {
            if (j != 0) {
                cout << "\t";
            }
            cout << matrix[i][j];
        }
        cout << endl;
    }
}

int main (int argc, char* argv[]) {
    string filename;
    if (argc < 3) {
        filename = "bigMatrix.in";
    } else {
        filename = argv[2];
    }

    int n = getMatrixSize(filename);
    vector<int> inner (n);
    vector< vector<int> > A(n, inner), B(n, inner), C(n, inner);
    read (filename, A, B);
    // do something with the matrices
    printMatrix(C, n);
    return 0;
}

Is it possible to get Python as fast as C++ for I/O? How could I improve the I/O for Python / C++?

(I've heard scanf should be faster than cin. Why should it be faster?)

This is the GIT-Repository with all Code.

Answer 1

It takes a while to start the python interpreter. Take this into consideration when running your tests.

Answer 2

When you are not mixing C and C++ file manipulation routines, you should turn off the synchronization with stdio.

http://www.cplusplus.com/reference/iostream/ios_base/sync_with_stdio/

cin and cout should be generally faster then their C counterparts (with synchronization turned off).

As far as the slowness of Python goes, well, why don't you check the implementation of the I/O functions?

Answer 3

You must buffer I/O in any language.

And why are you comparing Python and C++?

Python is an interpreted language, while C++ is compiled.