stateless_rand is a header-only (include/statelessrnd.hpp), stateless (pseudo) C++ random generator / number that implements the linear congruential relation x <- (x * a + c) % m, where x is the current state value, and a, c and m are referred to as multiplier, increment and modulus parameter, respectively. As all linear congruential algorithms, the generated (pseudo) random values suffer from the Marsaglia effect but benefit from the small state size and blazingly fast calculation. Always keep this in mind and use it carefully! (Especially, this library should not be used for security relevant aspects of your project!)
In marked difference to the implementation available within the STL, this implementation is purely functional and thus can be used in constexpr contexts without sacrificing functionality. The most obvious differences to STL's std::linear_congruential_engine are the initialization via the static seed() method and the next() member function, which, instead of altering the internal state, returns a new instance of stateless_rand that represents the next value in the (pseudo) random sequence. Since the internal (constant) state has the size of an integer, this operation is efficient. It is impossible to change the internal state of the generator, it could hence also be considered a random number itself than a generator.
using namespace statelessrnd;
constexpr auto rnd = minstd_rand::seed(42u); // 2027382
constexpr auto rnd2 = rnd.next(); // 1226992407
The state can be accessed via three equivalent methods: * (dereferencing operator), value() or the conversion operator to the underlying type.
constexpr auto a = *rnd; // 2027382
constexpr auto b = rnd.value(); // 2027382
constexpr decltype(rnd)::value_type c{rnd}; // 2027382
static_assert(a == b && a == c);
The sequence is seeded with a seed seed via seed(seed). The first value (seed) is skipped by default, this can be changed though via setting skip_first to false. Linear congruential algorithms produce fully deterministic sequences, seeding can therefore also be used as an entry point to a given sequence.
constexpr auto seed = minstd_rand::seed(42u, false);
static_assert(*seed == 42u);
constexpr auto rnd3 = seed.next().next().next();
static_assert(minstd_rand::seed(*rnd3).value() == rnd3.next().value());
For the sake of compliance with std::linear_congruential_engine, we further provide getters for the minimum and maximum value of the sequence (min() and max(), respectively), as well as a mechanism to skip n values in the sequence via discard(n). The latter is equivalent to n subsequent calls of next(). Note however, that the same might also be achieved by specifying a desired value of the sequence explicitly as a seed, which can be much faster, potentially.
auto stlrnd = std::minstd_rand(*seed);
stlrnd.discard(100);
std::cout << stlrnd() << '\n'; // 544861123
std::cout << seed.discard(101).value() << '\n'; // 544861123
std::cout << minstd_rand::seed(stlrnd()).value() << '\n'; // 1571445537
std::cout << stlrnd() << '\n'; // 1571445537
Seeds lower than min() or greater than max() are considered ill-formed and are clamped onto the interval [min(), max()] during during initialization. In particular, this means for any generator with c=0, that seed = 0 and seed = 1 are equivalent and produce the same sequence!
static_assert(minstd_rand::seed(0u).value() == minstd_rand::seed(1u).value());
The type of the state is accessible via stateless_rand::value_type, a, c and m are template parameters. A decent choice which matches the one of std::minstd_rand, is named minstd_rand in the statlessrnd namespace of this library.
using minstd_rand = stateless_rand<48271u, 0u, 2147483647u>;