why is representation type for chrono::nanoseconds a signed integer type?

Question

In gcc 4.84 in file /usr/include/c++/4.8 we have

namespace std  
{
   typedef duration<int64_t, nano>     nanoseconds;

Why is the representation type for nanosecond a signed integer type? Why isn't it unsigned? when can we have a duration object with a negative value ?

Answer 1

when can we have a duration object with a negative value ?

Any time you want to represent a negative duration!

e.g. "Ten seconds earlier" would be std::chrono::seconds(-10) and if you add that to some time_point t then you get a time_point that is ten seconds before t.

The standard says "A duration type measures time between two points in time (time_points)." It doesn't say that it can only measure time between non-decreasing time points. So that means it can be used to measure the time between t1 and t2 even if t2 < t1. And to do that simply, you need a negative value.

If durations couldn't be signed then, to represent an offset that means "earlier" not "later", you'd have to use something like std::pair<bool, duration> where the bool says whether it's a positive or negative offset, and then you'd have to do:

chrono::time_point adjust(chrono::time_point t, pair<bool, duration> offset)
{
  if (offset.first)  // positive
    return t + offset.second;
  else // negative
    return t - offset.second;
}

This is dumb. The language and the hardware already support this far more expressively and efficiently, by using signed integers.

Answer 2

If two objects a and b are defined such that a - b is a duration, then it's desirable to have a - b = -(b - a).

To implement that anti-commutative property, the duration needs to be signed.