What is the most efficient way to generate random strings in C++? - c++

Question

I need to generate random strings efficiently. In the following, you will see my first try. I compiled the code with gcc and -O3 optimization level. It takes 18.5 seconds to generate 10^7 random strings of length 64:
#include <iostream>
#include <random>
#include <algorithm>
std::string chars {"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890!##$%^&*()`~-_=+[{]{|;:'\",<.>/?"};
std::random_device rd;
std::mt19937 generator(rd());
std::string rand_str (int length) {
std::string output (chars);
std::shuffle(output.begin(), output.end(), generator);
return output.substr(0, length);
}
int main() {
std::string str;
for (long i=0; i<10000000; ++i)
str = rand_str (64);
}
I checked std::sample in c++17 and it is not faster than the above method. In addition, it will not change the order of characters and so it is not really random.
Edit: The std::shuffle is not a good choice, since, it will not allow duplicates. Based on comments I modified the code. This time it takes more than 9 minutes for 10^7 random numbers.
std::string rand_str (size_t length) {
const size_t char_size = chars.size();
std::uniform_int_distribution<> random_int (0, char_size - 1);
std::string output;
for (size_t i=0; i<length; ++i)
output.push_back(chars[random_int(generator)]);
return output;
}
Question
Are there more efficient ways to do this in modern C++?
I appreciate any suggestions to improve the code.

Answer 1

#include <iostream>
#include <random>
#include <algorithm>
#include <chrono>
std::string chars {"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890!##$%^&*()`~-_=+[{]{|;:'\",<.>/?"};
std::random_device rd;
std::mt19937 generator(rd());
std::string rand_str(int length) {
std::string output;
output.reserve(length);
while(length>0)
{
auto randNumb = generator();
while(randNumb > 93 && length--)
{
output.push_back(chars[randNumb%93]);
randNumb/=93;
}
}
return output;
}
int main() {
auto startTP = std::chrono::system_clock::now();
std::string rand_bytes;
for (long i=0; i<10000000; ++i)
rand_bytes = std::move(rand_str(64));
auto endTP = std::chrono::system_clock::now();
std::cout << "This took: " << std::chrono::duration_cast<std::chrono::microseconds>(endTP-startTP).count() << std::endl;
}
This takes around 3 seconds on my machine. The trick is to call the random number generator as little as possible and to allocate the memory only once.
What I'm doing is converting randNumber from base 10 to base 93(the length of chars). After that im using every base 93 digit as a different random number. This provides around 5 numbers per generated random number.