uniform_int_distribution a（），b（），min（）和max（）

Question

http://en.cppreference.com/w/cpp/numeric/random/uniform_int_distribution/params http://en.cppreference.com/w/cpp/numeric/random/uniform_int_distribution/min http://en.cppreference.com/w/cpp/numeric/random/uniform_int_distribution/max

It looks like member function a() is equivalent to member function min() , and member function b() is equivalent to max() .

#include <iostream>
#include <random>

int main() {
    std::uniform_int_distribution<int> dist(5, 10);
    std::cout << "a " << dist.a() << '\n';
    std::cout << "b " << dist.b() << '\n';
    std::cout << "min " << dist.min() << '\n';
    std::cout << "max " << dist.max() << '\n';
}

prints

a 5
b 10
min 5
max 10

When compiled using gcc's standard library. Are these functions really identical, and if so, why are a() and b() defined?

Answer 1

Are these functions really identical

They are identical, because it is reasonable and practical to always assume so. It would be true but pedantic and misleading to say it is not an absolute guarantee.

I only have access to MinGW 4.7.1, which I assume its standard library is the same as GCC one. Inside it, the class template uniform_int_distribution has members:

  /**
   * @brief Returns the inclusive lower bound of the distribution range.
   */
  result_type
  min() const
  { return this->a(); }

  /**
   * @brief Returns the inclusive upper bound of the distribution range.
   */
  result_type
  max() const
  { return this->b(); }

So, after function inlining, ( a and min ) and ( b and max ) should be tanslated into identical code.

Reading the standard section 26.5.8.2.1, you will find that it (only indirectly) says they should return the same value. As a result, a sane library implementer will make them effectively identical or at least not-so-different.

---

why are a() and b() defined?

I can only guess. It may be about consistency, but the consistency is in a less formal sense.

Mathematically, a uniform distribution is like this:

P(i|a,b) = 1/(b-a+1)

And in uniform_int_distribution , we have

uniform_int_distribution::a()
uniform_int_distribution::b()

For Bernoulli distribution and bernoulli_distribution :

P(i|p) = [complicated]

bernoulli_distribution::p()

Poisson:

P(i|mean) = [complicated]

poisson_distribution::mean()

Normal:

P(x|mean, standard-deviation) = [complicated]

normal_distribution::mean()
normal_distribution::stddev()

We can observe that they all tell their parameters. It is not useful to generic code, but that may be helpful in some situations.

Answer 2

a() and b() return distribution parameters while min() and max() return minimum and maximum potentially generated value. For the uniform distribution the value returned by min() and a() are equal, and same for max() and b() . In general, for other distributions there may not be such correspondence, so I guess a() and b() are there for consistency.

Answer 3

Every random number distribution D must have methods

D::result_type D::min();
D::result_type D::max();

which return the "greatest lower bound and the least upper bound on the values potentially returned by d 's operator() , as determined by the current values of d 's parameters" (§25.1.6, [rand.req.dist] ; quote is from paragraph 3(d); the requirement is in Table 117).

Also, there is an expectation that a distribution D and its associated parameter type D::param_type (called P in the following quote) will have corresponding constructors and parameter accessors. Paragraph 9 of the same section:

For each of the constructors of D taking arguments corresponding to parameters of the distribution, P shall have a corresponding constructor subject to the same requirements and taking arguments identical in number, type, and default values. Moreover, for each of the member functions of D that return values corresponding to parameters of the distribution, P shall have a corresponding member function with the identical name, type, and semantics.

So std::uniform_int_distribution::min and std::uniform_int_distribution::max are the bounds of a particular distribution instance's possible return values, while std::uniform_int_distribution::a and std::uniform_int_distribution::b are the parameter values which constructed that particular instance. As it happens, in the particular case of std::uniform_{int,real}_distribution , the parameters precisely correspond to the bounds, but the requirements suggest that both bounds and parameters should be provided.

I suppose it would have been possible to also call the parameter accessors min and max , but the committee chose not to do that.