mpq_rational: static_cast<int64_t>() of certain rationals fails on certain platforms

[mbunkus]

The short version: converting a mpq_rational{10, 3} to an int64_t via static_cast<int64_t>(…) yields std:::numeric_limits<int64_t>::max() on at least the following platforms:

• Debian Buster x86 32-bit (Boost 1.67, gmp 6.1.2)
• A mingw cross compiler from the MXE project (gcc 10.2, Boost 1.74, gmp 6.1.2) in both x86 32-bit (i686-w64-mingw32.static-g++) and x86 64-bit (x86_64-w64-mingw32.static-g++) modes
• Raspberry Pi OS (which is based on Debian Buster; Boost 1.67, gmp 6.1.2) on a Raspberry Pi 4 (armhf architecture)

It works fine and returns 3 on all the other platforms I could test, which are:

• Debian Buster 64-bit (Boost 1.67, gmp 6.1.2); yeah same OS, just 64-bit instead of 32-bit
• Ubuntu Focal 64-bit (Boost 1.71, gmp 6.2.0)
• AlmaLinux 64-bit (Boost 1.66, gmp
• Arch 64-bit (Boost 1.76, gmp 6.2.1)

In summary: it fails on all 32-bit platforms and on Windows in general. It works on 64-bit Linux platforms. Both seem to be independent of the versions of the libraries involved.

However, the conversion works if I convert to boost::multiprecision::int128_t first, then to int64_t. See the following sample code:

#include <iostream>
#include <boost/multiprecision/cpp_int.hpp>
#include <boost/multiprecision/gmp.hpp>

int main(int, char **) {
  boost::multiprecision::mpq_rational value{10, 3};

  std::cout << "fails: " << static_cast<int64_t>(value) << "\n";
  std::cout << "works: " << static_cast<int64_t>(static_cast<boost::multiprecision::int128_t>(value)) << "\n";

  return 0;
}

Compile with e.g.:

g++ -Wall -Wextra -std=c++17 -o test test.cpp -lgmp

Sample output on an affected platform:

$ ./test
fails: 9223372036854775807
works: 3

It should print 3 in both lines.

Things I've tried unsuccessfully:

• Compiling with or without optimizations (-O3)
• Compiling with different C++ standard versions
• Using different gcc versions
• Using clang

Things that do work, even on the aforementioned affected platforms:

• Converting mpq_rational{0, 1}
• Converting mpq_rational{10, 1}
• Converting boost::multiprecision::mpq_rational{boost::multiprecision::mpq_rational{std::numeric_limits<int64_t>::max()}, boost::multiprecision::mpq_rational{1}} - 1

I obviously have no clue why the conversion fails.

Background: I'm trying to convert my program from using boost::rational<int64_t> to boost::multiprecision::mpq_rational. If I understand the documentation correctly, the equivalent of a boost::rational_cast<int64_t>() ("convert this rational to the closest (truncated) int64_t value") is a regular static_cast<int64_t>(). If not, please point me in the right direction. Thanks!

[jzmaddock]

Confirmed, I see the problem.