Propagation breaks down when provided times are not in UTC
parzivail opened this issue · 4 comments
From #8:
davidtgillard: Would the Kind of output DateTimes be kept consistent with input DateTimes? Or would the Kind of output DateTimes always be UTC?
I believe outputs should always be in UTC for consistency. The caller can always convert it to whatever timezone they'd like.
The documentation could indicate as such.
That's reasonable. It would reduce overhead on our part and would leave the decision up to the consumer to convert to the time zone that they prefer, should they need something other than UTC.
On the face of it, this approach seems like surprising behavior to me - if a developer calls a pure function, e.g. Observe, with Datetimes in, say, NST, then it seems inconsistent that they would get back UTC DateTimes. However, the alternative with DateTime - trying to keep the DateTimes in a single well-defined and consistent zone - seems like a bad idea, because the API exposes time-related state for the caller to consume as they wish (e.g. EciCoordinate owns a DateTime). So I'd use DateTimeOffset to make it explicit in the datatype that it's all in UTC.
But the internet is strewn with DateTime vs DateTimeOffset discussions. Speaking practically, I'm happy with the proposed DateTime approach above.