Talk:cpp/numeric/midpoint

I think "ref std c++20" is missing. How it can be added? Ruslo (talk) 14:16, 1 October 2019 (PDT)

Wouldn't it be clearer to say x[std::midpoint(i,j)] rather than x[i+(j-i)/2]? 82.42.34.14 00:56, 30 April 2020 (PDT)

the proposal actually says pointer midpoint cannot be implemented in portable C++, while integer midpoint can be (and a sample impl is provided), so there must be a difference. --Cubbi (talk) 05:44, 30 April 2020 (PDT)

Guarantees & corner cases?

What are the guarantees and corner cases? Is it guaranteed to return a value in the given range (with the exception of (-inf, inf) -> nan)? Other than inf, is it guaranteed that std::midpoint(-x, x) == 0.0? Here are the ones I tried:

#include <format>
#include <cmath>
#define DO_ONE_LINE(a, b)     std::cout << std::format("midpoint({}, {}) = {} vs {} ({}, {})\n", a, b, std::midpoint(a, b), a + 0.5 * (b - a), #a, #b)
 
int main()
{
    static constexpr auto max = std::numeric_limits<double>::max();
    static constexpr auto inf = std::numeric_limits<double>::infinity();
    static constexpr auto denormMin = std::numeric_limits<double>::denorm_min();
    static constexpr auto lowest = std::numeric_limits<double>::lowest();
    
    DO_ONE_LINE(0.0, inf);
    DO_ONE_LINE(-inf, inf);
    DO_ONE_LINE(-max, max);
    DO_ONE_LINE(-denormMin, lowest);
    DO_ONE_LINE(inf, -inf);
    DO_ONE_LINE(denormMin, lowest);
    DO_ONE_LINE(lowest, denormMin);
    DO_ONE_LINE(std::nextafter(denormMin, 1.0), lowest);
    DO_ONE_LINE(lowest, denormMin);
    DO_ONE_LINE(lowest, std::nextafter(denormMin, 1.0));
    DO_ONE_LINE(std::nextafter(denormMin, 1.0), denormMin);
    DO_ONE_LINE(denormMin, std::nextafter(denormMin, 1.0));
    DO_ONE_LINE(0.0, denormMin);
    DO_ONE_LINE(denormMin, 0.0);
    DO_ONE_LINE(denormMin, denormMin);

which produces

midpoint(0, inf) = inf vs inf (0.0, inf)
midpoint(-inf, inf) = -nan vs -nan (-inf, inf)
midpoint(-1.7976931348623157e+308, 1.7976931348623157e+308) = 0 vs inf (-max, max)
midpoint(-5e-324, -1.7976931348623157e+308) = -8.988465674311579e+307 vs -8.988465674311579e+307 (-denormMin, lowest)
midpoint(inf, -inf) = -nan vs -nan (inf, -inf)
midpoint(5e-324, -1.7976931348623157e+308) = -8.988465674311579e+307 vs -8.988465674311579e+307 (denormMin, lowest)
midpoint(-1.7976931348623157e+308, 5e-324) = -8.988465674311579e+307 vs -8.988465674311579e+307 (lowest, denormMin)
midpoint(1e-323, -1.7976931348623157e+308) = -8.988465674311579e+307 vs -8.988465674311579e+307 (std::nextafter(denormMin, 1.0), lowest)
midpoint(-1.7976931348623157e+308, 5e-324) = -8.988465674311579e+307 vs -8.988465674311579e+307 (lowest, denormMin)
midpoint(-1.7976931348623157e+308, 1e-323) = -8.988465674311579e+307 vs -8.988465674311579e+307 (lowest, std::nextafter(denormMin, 1.0))
midpoint(1e-323, 5e-324) = 1e-323 vs 1e-323 (std::nextafter(denormMin, 1.0), denormMin)
midpoint(5e-324, 1e-323) = 1e-323 vs 5e-324 (denormMin, std::nextafter(denormMin, 1.0))
midpoint(0, 5e-324) = 0 vs 0 (0.0, denormMin)
midpoint(5e-324, 0) = 0 vs 5e-324 (denormMin, 0.0)
midpoint(5e-324, 5e-324) = 5e-324 vs 5e-324 (denormMin, denormMin)

BenFrantzDale (talk) 06:28, 26 April 2024 (PDT)