Printing doubles using string manipulation

Having mostly used Qt and its classes like QString, I wrote a little exercise with plain C++11. I’m looking for two kinds of input: generally improving the code and/or doing it better in an entirely different way (still standard C++11 or C++14, no Boost or anything).

The code is for printing doubles with given number of significant digits, by first producing a string using scientific notation, and then manipulating that string to move the dot and pad it with zeros.

#include <sstream>
#include <iostream>
#include <iomanip>
#include <vector>
#include <algorithm>
#include <stdexcept>

template <typename T>
std::string number_to_sci_string(T number, int precision) {
    std::stringstream ss;
    ss << std::scientific << std::setprecision(precision-1) << number;
    return ss.str();
}

std::string sci_string_to_normal(const std::string &s) {
    auto dot_pos = s.find('.');
    auto e_pos = s.find('e');
    if (dot_pos == std::string::npos || // dot required
        dot_pos < 1 ||  // part1 below must not be empty
        e_pos == std::string::npos || // e required
        e_pos >= s.length()-1 || // must not be last
        e_pos - dot_pos < 2) { // part2 below must not be empty
        throw std::invalid_argument(s);
    }

    auto exponent = std::stoi(s.substr(e_pos + 1), nullptr, 10);
    if (exponent == 0) {
        return s.substr(0, e_pos);
    }
    else {
        auto part1 = s.substr(0, dot_pos);
        auto part2 = s.substr(dot_pos + 1, e_pos - dot_pos - 1);
        auto part2length = static_cast<ssize_t>(part2.length());
        if (exponent >= part2length) {
            auto fill = std::string(exponent - part2length, '0');
            return part1 + part2 + fill;
        }
        else if (exponent < 0) {
            auto fill = "0." + std::string(-1 - exponent, '0');
            // this is the only case where sign matters
            auto sign = (s[0] == '-' || s[0] == '+')
                        ? std::string(1, s[0])
                        : std::string();
            return sign + fill + part1.substr(sign.length()) + part2;
        }
        else { // 0 < exponent < part2.length()
            return part1 + part2.substr(0,  exponent) + "." + part2.substr(exponent);
        }
    }
}

using namespace std;

int main(void)
{
    vector<double> nums{ -2000.0005, -1.2345234, -0.000011345, 0, 0.0000001, 0.1342134, 1.1234, 10000 };
    for_each(nums.begin(), nums.end(), [](double &n) {
        string sci = number_to_sci_string(n, 5);
        string nor = sci_string_to_normal(sci);
        cout << n << ": " << sci << " == " << nor << endl;
    });
}

1. number_to_sci_string looks ok, though you could considerably improve performance by chucking the template, restricting to float, double and long double and using snprintf:

   std::string number_to_sci_string(long double number, int precision) {
       std::string s(std::snprintf(0, 0, "%.*Le", precision-1, number), '');
       std::snprintf(&s[0], s.size()+1, "%.*Le", precision-1, number);
       return s;
   }
   

   Also, I really wonder why you are using a different notion of precision than the standard library…

2. I suggest you be a bit more flexible in what you accept:

   • The exponent e00 should be optional.
   • The fractional part (and decimal point) .00 should be optional.
   • If there’s a fractional part, the integer part should be optional.

3. Don’t use using namespace std;, even in the implementation-file: You don’t control what symbols it contains, and there are no guarantees none are added.

4. The opening brace of main()‘s body is, in contrast to all other functions, on a separate line. Why?

5. Prefer the for-range-loop to std::for_each+lambda. The compiler should compile both to the same result, but the former is simpler and looks better.