I think that a signed value is a more metrologically and physically correct representation of any non-absolute, "non-Kelvin" time system in comparison the current ROS2 unsigned value.

Regardless of metrology correctness signed time value makes computations of durations (always a signed value!) more unerring in the programming sense. MISRA C 2012 rule 10.4 treats signed and unsigned as different incompatible types. It takes very messy coding to account for over- and underflow edge conditions in sign-mixed math.

The current nanoseconds uint64_t gives us ~584.5 years from 1970, int64_t is twice shorter. Correspondingly it's 2554AD and 2262AD.

I think both end dates are remote enough for both unsigned and signed values to be practical in today's applications.

CC: @wjwwood