Port from_timestamp_millis() (#818) to 0.4.x

[djc]

Added some changes to deduplicate logic in the code.

[esheppa]

I think we still need to include this part as well: NANOS_IN_SECOND.checked_sub(nsecs)?, (in the millis < 0 case only)

[djc]

Oh yeah, good catch. Added a commit that enables CI to run on the 0.4.x branch, too...

[esheppa]

Looks good! Maybe backporting to 0.4.x is cleaner - it keeps the history of the main branch nicer as well

[djc]

Hmm, let's try a bit more -- we should be a little more forceful about getting authors to target 0.4.x, I think.

[Pscheidl]

This version is cleaner, but should be slightly less efficient - two ifs instead of one and the 'abs' operation is always performed. Not sure if you care about the difference, just explaining the original implementation.

[djc]

Yeah -- unless someone wants to build benchmarks that show a substantial improvement, I'd rather opt for code that's easier to follow.

As for the abs() call -- since abs() will internally call is_negative() anyway, branching in order to prevent an abs() call doesn't actually help -- it will have the same number of branches in the case it's not needed, and an extra branch in the case it is needed.