Strange division evaluation for dynamically allocated `cpp_bin_float`

[Michel Morin]

With this code

#include <iostream>
#include <boost/multiprecision/cpp_bin_float.hpp>

using namespace boost::multiprecision;
using Real = number<cpp_bin_float<10000, digit_base_10, std::allocator<void>>>;

int main(int argc, char* argv[])
{
    std::cout << 1 / Real(1.4) << std::endl;
    return 0;
}

assertion fails as

Assertion failed: 
((
    eval_msb(q)
    ==
    cpp_bin_float<Digits, DigitBase, Allocator, Exponent, MinE, MaxE>::bit_count - 1
)), 
function eval_divide, file boost/multiprecision/cpp_bin_float.hpp, line 1086.

or it prints 26699.6 when compiled with NDEBUG. If I change Real(1.4) to Real(1), the program does not terminate.

Tested on g++-5.4.

[John Maddock]

I've tracked this down to a bug in cpp_int, fixed in: ​https://github.com/boostorg/multiprecision/commit/cd2bbcba5bc0f8335fd7d93bf1d8ffe875e722a0

[Michel Morin]

Confirmed that the testcase works fine. Thanks for the quick fix!

Regards, Michel