Disambiguation behaviour when converting DateTime skipped by DST transition

[ptomato]

Follow up from #316.

Currently in the polyfill, disambiguating a wall-clock time that is skipped when setting the clock an hour forward for DST will give a time that's either 1 hour earlier or 1 hour later.

@gibson042 >

This seems wrong to me... it means that tz.getAbsoluteFor(skipped_02_45.minus("PT50M"), "earlier") is later than tz.getAbsoluteFor(skipped_02_45, "earlier")! Shouldn't the "two possible absolute times" for a time of day skipped by a positive transition be "immediately before the transition" vs. "the matching time in the pre-transition offset" (e.g., converting a skipped 02:45 to 01:59:59.999999999 vs. 03:45)?

[ptomato]

I don't have a strong opinion one way or the other, but I can think of the following reasons that mildly support keeping it the way it is:

• The current behaviour is how moment-timezone does it.
• The other DST transition (setting the clock back) gives you two possible times that are an hour apart, so there's a sort-of symmetry argument to be made here.

[gibson042]

The current behaviour is how moment-timezone does it.

Can you demonstrate what you mean? I don't see any way to get anything other than a post-transition value for skipped times of day.

> moment("2020-03-08T02:45").tz("America/New_York").format()
'2020-03-08T03:45:00-04:00'

[ptomato]

Oh, you're right, actually; scratch that reason.

[gilmoreorless]

e.g., converting a skipped 02:45 to 01:59:59.999999999

That's not a great approach to take. It would be inconsistent with other languages/libraries that handle disambiguation, and would be a very unintuitive result.

Imagine a task scheduled hourly in America/New_York with the equivalent UTC timestamps listed, using that process above:

America/New_York => UTC
2020-03-08T00:45 => 2020-03-08T05:45
2020-03-08T01:45 => 2020-03-08T06:45
2020-03-08T02:45 => 2020-03-08T06:59:59.999999999  <--- Unexpected
2020-03-08T03:45 => 2020-03-08T07:45

If I saw that result in my application, I'd assume there was a bug somewhere.