Question

I have a large file where each line contains space-separated integers. The task is to sparse this file line-by-line. For the string to int conversion I have three solutions:

static int stringToIntV1(const string& str) {
    return (atoi(str.c_str()));
}

However, if I pass a malformed string, it doesn’t produce any error. For instance the string “123error” is converted to 123.

Second solution:

static int stringToIntV2(const string& str)
{
    int result;
    istringstream myStream(str);

    if (myStream >> result) {
        return result;
    }
    // else
    throw domain_error(str + " is not an int!");
}

I have the same problem here, malformed strings don’t raise an error.

Third solution with Boost (found at Boost Library):

static int stringToIntV3(const string& str)
{
    int iResult = 0;
    try {
        iResult = lexical_cast<int>(str);
    }
    catch(bad_lexical_cast &) {
        throw domain_error(str + " is not an int!");
    }
    return iResult;
}

This one gives correct result.

However, there is a significant difference in the execution time. Testing on a large text file (32 MB), I got the following times:

• (1) with atoi: 4.522s (winner)
• (2) with istringstream: 15.303s (very slow)
• (3) with lexical_cast: 10.958s (between the two)

My question: do you know how to notice malformed strings with atoi? It would give the fastest solution. Or do you know a better solution?

Update: Thanks for the answers. Following the tips, I came up with this solution:

static int stringToIntV4(const string& str)
{
    char * pEnd;
    const char * c_str = str.c_str();
    int result = strtol(c_str, &pEnd, 10);
    if (pEnd == c_str+str.length()) {
        return result;
    }
    // else
    throw domain_error("'" + str + "'" + " is not an int!");
}

The good news is that it yields if there is a problem and as efficient as the atoi version.

Answer 1

I’d use strtol. It takes a parameter that it sets to point at the first character it couldn’t convert, so you can use that to determine whether the entire string was converted.

Edit: as far as speed goes, I’d expect it to be slightly slower than atoi, but faster than the others you tried.