This header provides an equivalent to C++20's std::in_range<integer>(value) for floating-point values.

It defines the function

namespace in_range_ext {
  template<integer I, std::floating_point F> constexpr bool in_range(F f);
}

which returns true iff the floating-point value f is in range for integer type I.

integer is a concept matching std::integral types other than character types and bool (as for std::in_range).

For this one function, most of the work is in finding, at compile time, the highest and lowest floating-point values in the range of the integer type. This cannot be done by direct conversion in the general case in C++ because of rounding. For example on a typical 32-/64-bit system with

std::numeric_limits<int>::digits   31
std::numeric_limits<int>::max()    0x7fffffff (2^31 - 1)
std::numeric_limits<float>::radix  2
std::numeric_limits<float>::digts  24

float(0x7fffffff) rounded to nearest is 0x80000000 (2^31), just outside int range.

Instead, the code establishes these boundaries "manually" using a decomposed floating-point representation.

It is designed to work for arbitrary integer width, floating-point radix, and precision, including the common cases where the floating-point range is much larger, and less common ones where it may be smaller, for example IEEE half-precision aka binary16.