How to generate a random number in C++?

Question

Using modulo may introduce bias into the random numbers, depending on the random number generator. See this question for more info. Of course, it's perfectly possible to get repeating numbers in a random sequence.

Try some C++11 features for better distribution:

#include <random>
#include <iostream>

int main()
{
    std::random_device dev;
    std::mt19937 rng(dev());
    std::uniform_int_distribution<std::mt19937::result_type> dist6(1,6); // distribution in range [1, 6]

    std::cout << dist6(rng) << std::endl;
}

See this question/answer for more info on C++11 random numbers. The above isn't the only way to do this, but is one way.

The most fundamental problem of your test application is that you call srand once and then call rand one time and exit.

The whole point of srand function is to initialize the sequence of pseudo-random numbers with a random seed.

It means that if you pass the same value to srand in two different applications (with the same srand/rand implementation) then you will get exactly the same sequence of rand() values read after that in both applications.

However in your example application pseudo-random sequence consists only of one element - the first element of a pseudo-random sequence generated from seed equal to current time of 1 sec precision. What do you expect to see on output then?

Obviously when you happen to run application on the same second - you use the same seed value - thus your result is the same of course (as Martin York already mentioned in a comment to the question).

Actually you should call srand(seed) one time and then call rand() many times and analyze that sequence - it should look random.

AMENDMENT 1 - example code:

OK I get it. Apparently verbal description is not enough (maybe language barrier or something... :) ).

Old-fashioned C code example based on the same srand()/rand()/time() functions that was used in the question:

#include <stdlib.h>
#include <time.h>
#include <stdio.h>

int main(void)
{
    unsigned long j;
    srand( (unsigned)time(NULL) );

    for( j = 0; j < 100500; ++j )
    {
        int n;

        /* skip rand() readings that would make n%6 non-uniformly distributed
          (assuming rand() itself is uniformly distributed from 0 to RAND_MAX) */
        while( ( n = rand() ) > RAND_MAX - (RAND_MAX-5)%6 )
        { /* bad value retrieved so get next one */ }

        printf( "%d,\t%d\n", n, n % 6 + 1 );
    }

    return 0;
}

^^^ THAT sequence from a single run of the program is supposed to look random.

Please NOTE that I don't recommend to use rand/srand functions in production for the reasons explained below and I absolutely don't recommend to use function time as a random seed for the reasons that IMO already should be quite obvious. Those are fine for educational purposes and to illustrate the point sometimes but for any serious use they are mostly useless.

AMENDMENT 2 - detailed explanation:

It is important to understand that as of now there is none C or C++ standard features (library functions or classes) producing actually random data definitively (i.e. guaranteed by the standard to be actually random). The only standard feature that approaches this problem is std::random_device that unfortunately still does not provide guarantees of actual randomness.

Depending on the nature of application you should first decide if you really need truly random (unpredictable) data. Notable case when you do most certainly need true randomness is information security - e.g. generating symmetric keys, asymmetric private keys, salt values, security tokens, etc.

However security-grade random numbers is a separate industry worth a separate article. I'm briefly discussing them in this answer of mine.

In most cases Pseudo-Random Number Generator is sufficient - e.g. for scientific simulations or games. In some cases consistently defined pseudo-random sequence is even required - e.g. in games you may choose to generate exactly same maps in runtime to avoid storing lots of data in your distribution.

The original question and reoccurring multitude of identical/similar questions (and even many misguided "answers" to them) indicate that first and foremost it is important to distinguish random numbers from pseudo-random numbers AND to understand what is pseudo-random number sequence in the first place AND to realize that pseudo-random number generators are NOT used the same way you could use true random number generators.

Intuitively when you request random number - the result returned shouldn't depend on previously returned values and shouldn't depend if anyone requested anything before and shouldn't depend in what moment and by what process and on what computer and from what generator and in what galaxy it was requested. That is what word "random" means after all - being unpredictable and independent of anything - otherwise it is not random anymore, right? With this intuition it is only natural to search the web for some magic spells to cast to get such random number in any possible context.

^^^ THAT kind of intuitive expectations IS VERY WRONG and harmful in all cases involving Pseudo-Random Number Generators - despite being reasonable for true random numbers.

While the meaningful notion of "random number" exists (kind of) - there is no such thing as "pseudo-random number". A Pseudo-Random Number Generator actually produces pseudo-random number sequence.

Pseudo-random sequence is in fact always deterministic (predetermined by its algorithm and initial parameters) - i.e. there is actually nothing random about it.

When experts talk about quality of PRNG they actually talk about statistical properties of the generated sequence (and its notable sub-sequences). For example if you combine two high quality PRNGs by using them both in turns - you may produce bad resulting sequence - despite them generating good sequences each separately (those two good sequences may simply correlate to each other and thus combine badly).

Specifically rand()/srand(s) pair of functions provide a singular per-process non-thread-safe(!) pseudo-random number sequence generated with implementation-defined algorithm. Function rand() produces values in range [0, RAND_MAX].

Quote from C11 standard (ISO/IEC 9899:2011):

The srand function uses the argument as a seed for a new sequence of pseudo-random numbers to be returned by subsequent calls to rand. If srand is then called with the same seed value, the sequence of pseudo-random numbers shall be repeated. If rand is called before any calls to srand have been made, the same sequence shall be generated as when srand is first called with a seed value of 1.

Many people reasonably expect that rand() would produce a sequence of semi-independent uniformly distributed numbers in range 0 to RAND_MAX. Well it most certainly should (otherwise it's useless) but unfortunately not only standard doesn't require that - there is even explicit disclaimer that states "there is no guarantees as to the quality of the random sequence produced". In some historical cases rand/srand implementation was of very bad quality indeed. Even though in modern implementations it is most likely good enough - but the trust is broken and not easy to recover. Besides its non-thread-safe nature makes its safe usage in multi-threaded applications tricky and limited (still possible - you may just use them from one dedicated thread).

New class template std::mersenne_twister_engine<> (and its convenience typedefs - std::mt19937/std::mt19937_64 with good template parameters combination) provides per-object pseudo-random number generator defined in C++11 standard. With the same template parameters and the same initialization parameters different objects will generate exactly the same per-object output sequence on any computer in any application built with C++11 compliant standard library. The advantage of this class is its predictably high quality output sequence and full consistency across implementations.

Also there are more PRNG engines defined in C++11 standard - std::linear_congruential_engine<> (historically used as fair quality srand/rand algorithm in some C standard library implementations) and std::subtract_with_carry_engine<>. They also generate fully defined parameter-dependent per-object output sequences.

Modern day C++11 example replacement for the obsolete C code above:

#include <iostream>
#include <chrono>
#include <random>

int main()
{
    std::random_device rd;
    // seed value is designed specifically to make initialization
    // parameters of std::mt19937 (instance of std::mersenne_twister_engine<>)
    // different across executions of application
    std::mt19937::result_type seed = rd() ^ (
            (std::mt19937::result_type)
            std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()
                ).count() +
            (std::mt19937::result_type)
            std::chrono::duration_cast<std::chrono::microseconds>(
                std::chrono::high_resolution_clock::now().time_since_epoch()
                ).count() );

    std::mt19937 gen(seed);

    for( unsigned long j = 0; j < 100500; ++j )
    /* ^^^Yes. Generating single pseudo-random number makes no sense
       even if you use std::mersenne_twister_engine instead of rand()
       and even when your seed quality is much better than time(NULL) */    
    {
        std::mt19937::result_type n;
        // reject readings that would make n%6 non-uniformly distributed
        while( ( n = gen() ) > std::mt19937::max() -
                                    ( std::mt19937::max() - 5 )%6 )
        { /* bad value retrieved so get next one */ }

        std::cout << n << '\t' << n % 6 + 1 << '\n';
    }

    return 0;
}

The version of previous code that uses std::uniform_int_distribution<>

#include <iostream>
#include <chrono>
#include <random>

int main()
{
    std::random_device rd;
    std::mt19937::result_type seed = rd() ^ (
            (std::mt19937::result_type)
            std::chrono::duration_cast<std::chrono::seconds>(
                std::chrono::system_clock::now().time_since_epoch()
                ).count() +
            (std::mt19937::result_type)
            std::chrono::duration_cast<std::chrono::microseconds>(
                std::chrono::high_resolution_clock::now().time_since_epoch()
                ).count() );

    std::mt19937 gen(seed);
    std::uniform_int_distribution<unsigned> distrib(1, 6);

    for( unsigned long j = 0; j < 100500; ++j )
    {
        std::cout << distrib(gen) << ' ';
    }

    std::cout << '\n';
    return 0;
}

Whenever you do a basic web search for random number generation in the C++ programming language this question is usually the first to pop up! I want to throw my hat into the ring to hopefully better clarify the concept of pseudo-random number generation in C++ for future coders that will inevitably search this same question on the web!

The Basics

Pseudo-random number generation involves the process of utilizing a deterministic algorithm that produces a sequence of numbers whose properties approximately resemble random numbers. I say approximately resemble, because true randomness is a rather elusive mystery in mathematics and computer science. Hence, why the term pseudo-random is utilized to be more pedantically correct!

Before you can actually use a PRNG, i.e., pseudo-random number generator, you must provide the algorithm with an initial value often referred too as the seed. However, the seed must only be set once before using the algorithm itself!

/// Proper way!
seed( 1234 ) /// Seed set only once...
for( x in range( 0, 10) ):
  PRNG( seed ) /// Will work as expected

/// Wrong way!
for( x in rang( 0, 10 ) ):
  seed( 1234 ) /// Seed reset for ten iterations!
  PRNG( seed ) /// Output will be the same...

Thus, if you want a good sequence of numbers, then you must provide an ample seed to the PRNG!

The Old C Way

The backwards compatible standard library of C that C++ has, uses what is called a linear congruential generator found in the cstdlib header file! This PRNG functions through a discontinuous piecewise function that utilizes modular arithmetic, i.e., a quick algorithm that likes to use the modulo operator '%'. The following is common usage of this PRNG, with regards to the original question asked by @Predictability:

#include <iostream>
#include <cstdlib>
#include <ctime>

int main( void )
{
  int low_dist  = 1;
  int high_dist = 6;
  std::srand( ( unsigned int )std::time( nullptr ) );
  for( int repetition = 0; repetition < 10; ++repetition )
    std::cout << low_dist + std::rand() % ( high_dist - low_dist ) << std::endl;
  return 0;
}

The common usage of C's PRNG houses a whole host of issues such as:

1. The overall interface of std::rand() isn't very intuitive for the proper generation of pseudo-random numbers between a given range, e.g., producing numbers between [1, 6] the way @Predictability wanted.
2. The common usage of std::rand() eliminates the possibility of a uniform distribution of pseudo-random numbers, because of the Pigeonhole Principle.
3. The common way std::rand() gets seeded through std::srand( ( unsigned int )std::time( nullptr ) ) technically isn't correct, because time_t is considered to be a restricted type. Therefore, the conversion from time_t to unsigned int is not guaranteed!

For more detailed information about the overall issues of using C's PRNG, and how to possibly circumvent them, please refer to Using rand() (C/C++): Advice for the C standard library’s rand() function!

The Standard C++ Way

Since the ISO/IEC 14882:2011 standard was published, i.e., C++11, the random library has been apart of the C++ programming language for a while now. This library comes equipped with multiple PRNGs, and different distribution types such as: uniform distribution, normal distribution, binomial distribution, etc. The following source code example demonstrates a very basic usage of the random library, with regards to @Predictability's original question:

#include <iostream>
#include <cctype>
#include <random>

using u32    = uint_least32_t; 
using engine = std::mt19937;

int main( void )
{
  std::random_device os_seed;
  const u32 seed = os_seed();

  engine generator( seed );
  std::uniform_int_distribution< u32 > distribute( 1, 6 );

  for( int repetition = 0; repetition < 10; ++repetition )
    std::cout << distribute( generator ) << std::endl;
  return 0;
}

The 32-bit Mersenne Twister engine, with a uniform distribution of integer values was utilized in the above example. (The name of the engine in source code sounds weird, because its name comes from its period of 2^19937-1 ). The example also uses std::random_device to seed the engine, which obtains its value from the operating system (If you are using a Linux system, then std::random_device returns a value from /dev/urandom).

Take note, that you do not have to use std::random_device to seed any engine. You can use constants or even the chrono library! You also don't have to use the 32-bit version of the std::mt19937 engine, there are other options! For more information about the capabilities of the random library, please refer to cplusplus.com

All in all, C++ programmers should not use std::rand() anymore, not because its bad, but because the current standard provides better alternatives that are more straight forward and reliable. Hopefully, many of you find this helpful, especially those of you who recently web searched generating random numbers in c++!

If you are using boost libs you can obtain a random generator in this way:

#include <iostream>
#include <string>

// Used in randomization
#include <ctime>
#include <boost/random/mersenne_twister.hpp>
#include <boost/random/uniform_int_distribution.hpp>
#include <boost/random/variate_generator.hpp>

using namespace std;
using namespace boost;

int current_time_nanoseconds(){
    struct timespec tm;
    clock_gettime(CLOCK_REALTIME, &tm);
    return tm.tv_nsec;
}

int main (int argc, char* argv[]) {
    unsigned int dice_rolls = 12;
    random::mt19937 rng(current_time_nanoseconds());
    random::uniform_int_distribution<> six(1,6);

    for(unsigned int i=0; i<dice_rolls; i++){
        cout << six(rng) << endl;
    }
}

Where the function current_time_nanoseconds() gives the current time in nanoseconds which is used as a seed.

---

Here is a more general class to get random integers and dates in a range:

#include <iostream>
#include <ctime>
#include <boost/random/mersenne_twister.hpp>
#include <boost/random/uniform_int_distribution.hpp>
#include <boost/random/variate_generator.hpp>
#include "boost/date_time/posix_time/posix_time.hpp"
#include "boost/date_time/gregorian/gregorian.hpp"


using namespace std;
using namespace boost;
using namespace boost::posix_time;
using namespace boost::gregorian;


class Randomizer {
private:
    static const bool debug_mode = false;
    random::mt19937 rng_;

    // The private constructor so that the user can not directly instantiate
    Randomizer() {
        if(debug_mode==true){
            this->rng_ = random::mt19937();
        }else{
            this->rng_ = random::mt19937(current_time_nanoseconds());
        }
    };

    int current_time_nanoseconds(){
        struct timespec tm;
        clock_gettime(CLOCK_REALTIME, &tm);
        return tm.tv_nsec;
    }

    // C++ 03
    // ========
    // Dont forget to declare these two. You want to make sure they
    // are unacceptable otherwise you may accidentally get copies of
    // your singleton appearing.
    Randomizer(Randomizer const&);     // Don't Implement
    void operator=(Randomizer const&); // Don't implement

public:
    static Randomizer& get_instance(){
        // The only instance of the class is created at the first call get_instance ()
        // and will be destroyed only when the program exits
        static Randomizer instance;
        return instance;
    }
    bool method() { return true; };

    int rand(unsigned int floor, unsigned int ceil){
        random::uniform_int_distribution<> rand_ = random::uniform_int_distribution<> (floor,ceil);
        return (rand_(rng_));
    }

    // Is not considering the millisecons
    time_duration rand_time_duration(){
        boost::posix_time::time_duration floor(0, 0, 0, 0);
        boost::posix_time::time_duration ceil(23, 59, 59, 0);
        unsigned int rand_seconds = rand(floor.total_seconds(), ceil.total_seconds());
        return seconds(rand_seconds);
    }


    date rand_date_from_epoch_to_now(){
        date now = second_clock::local_time().date();
        return rand_date_from_epoch_to_ceil(now);
    }

    date rand_date_from_epoch_to_ceil(date ceil_date){
        date epoch = ptime(date(1970,1,1)).date();
        return rand_date_in_interval(epoch, ceil_date);
    }

    date rand_date_in_interval(date floor_date, date ceil_date){
        return rand_ptime_in_interval(ptime(floor_date), ptime(ceil_date)).date();
    }

    ptime rand_ptime_from_epoch_to_now(){
        ptime now = second_clock::local_time();
        return rand_ptime_from_epoch_to_ceil(now);
    }

    ptime rand_ptime_from_epoch_to_ceil(ptime ceil_date){
        ptime epoch = ptime(date(1970,1,1));
        return rand_ptime_in_interval(epoch, ceil_date);
    }

    ptime rand_ptime_in_interval(ptime floor_date, ptime ceil_date){
        time_duration const diff = ceil_date - floor_date;
        long long gap_seconds = diff.total_seconds();
        long long step_seconds = Randomizer::get_instance().rand(0, gap_seconds);
        return floor_date + seconds(step_seconds);
    }
};