[wleeaf]

Packing a std::chrono::time_point whose period is longer than one second crashes with SIGFPE:

std::chrono::time_point<std::chrono::system_clock, std::chrono::minutes> tp(std::chrono::minutes(2));
msgpack::sbuffer sbuf;
msgpack::pack(sbuf, tp); // Floating point exception

The same applies to std::chrono::hours and C++20 std::chrono::sys_days, and to msgpack::object(tp, zone).

The pack and object_with_zone adaptors in v1/adaptor/cpp11/chrono.hpp compute the nanosecond part as count % (1000000000 / nano_num) * nano_num, where nano_num = num * (1000000000 / den). For a period longer than one second, 1000000000 / nano_num is 0. For a period shorter than a nanosecond (std::pico), nano_num itself is 0. Either way the code divides by zero.

This change splits the time since the epoch into whole seconds and the nanosecond remainder with duration_cast, keeping the existing adjustment for negative remainders. The encoded bytes for the periods that already worked are unchanged (checked for seconds, milliseconds, and system_clock::time_point min/max).

Tests: new system_clock_minutes, system_clock_minutes_negative and system_clock_picoseconds cases in msgpack_cpp11.cpp, plus system_clock_minutes and system_clock_minutes_negative in object_with_zone.cpp. Each new case hits the division by zero (UBSan "division by zero" at chrono.hpp:254/305) without the fix and passes with it.

[redboltz]

@wleeaf , thank you for sending the PR. It looks good to me.