Merge lp://qastaging/~afrantzis/mir/clock-steady-duration-for-timestamp into lp://qastaging/mir

time: Use a steady (monotonic) clock

We use mir::time::Clock to calculate intervals, which is dangerous if we use a non-monotonic clock (std::chrono::high_resolution_clock is not guaranteed to be monotonic). This MP changes the code to use a monotonic clock (std::chrono::steady_clock).

A problem with our current approach is that the clock type leaks into our public interface (through the mir::time::Timestamp type definition). To fix this, this MP also changes the Timestamp definition to be a duration from an unspecified epoch. Since the mir::time::Clock::now() is meaningless (and shouldn't be used) as the wall clock time, having a specific, well-known epoch doesn't matter.

time: Use a steady (monotonic) clock

We use mir::time::Clock to calculate intervals, which is dangerous if we use a non-monotonic clock (std::chrono::high_resolution_clock is not guaranteed to be monotonic). This MP changes the code to use a monotonic clock (std::chrono::steady_clock).

A problem with our current approach is that the clock type leaks into our public interface (through the mir::time::Timestamp type definition). To fix this, this MP also changes the Timestamp definition to be a duration from an unspecified epoch. Since the mir::time::Clock::now() is meaningless (and shouldn't be used) as the wall clock time, having a specific, well-known epoch doesn't matter. The drawback is that time::Timestamp is actually a duration type which can be confusing under some circumstances.

Also see the alternative proposal that introduces the steady clock without the fix for the clock type leak: https://code.launchpad.net/~afrantzis/mir/clock-steady/+merge/240439