Fix: preserve datetime precision after Timex.shift/2

[tfiedlerdejanze]

Summary of changes

Resolves #731

Unfortunately this broke with elixir-lang/elixir@5a583c7 which was released with Elixir 1.14.3

As Timex.shift/2 calculates the total microseconds of the shift options and then uses DateTime.add/4 to apply, Elixir returns a new datetime with the highest precision it finds (microseconds) opposed to keeping the original precision.

Elixir 1.13.4:

iex(3)> DateTime.utc_now() |> IO.inspect() |> DateTime.truncate(:second) |> IO.inspect() |> Timex.shift(minutes: 1) |> IO.inspect()
~U[2023-04-13 09:56:10.136274Z]
~U[2023-04-13 09:56:10Z]
~U[2023-04-13 09:57:10Z]

Elixir 1.14.4:

iex(1)> DateTime.utc_now() |> IO.inspect() |> DateTime.truncate(:second) |> IO.inspect() |> Timex.shift(minutes: 1) |> IO.inspect()
~U[2023-04-13 09:55:16.405357Z]
~U[2023-04-13 09:55:16Z]
~U[2023-04-13 09:56:16.000000Z]

[tfiedlerdejanze]

@bitwalker is this a change you'd be considering for Timex?

[azizk]

Hope you don't mind if I make further suggestions. How about rewriting the two arrow clauses, like so:

%DateTime{} = original ->
  if shift == 0 do
    original
  else
    original
    |> DateTime.add(shift, :microsecond, Timex.Timezone.Database)
    |> retain_precision(original)
  end

[azizk]

Maybe it's a good idea to create a function for adding microseconds while keeping the precision? Something like:

original |> add_with_precision_preserved(shift)

[azizk]

I would get rid of the case statement below as it seems completely unnecessary, but that may be outside of the scope of this PR. 🤷 😅

[ftes]

@bitwalker Do you have any time plans for reviewing/merging this PR?

"I don't have time" is also a totally fine answer that helps us plan accordingly. Thank you for your efforts!

[petermueller]

Thank you doing this work. I've been unable to complete an upgrade to 1.14 as a huge number of our legacy database tables were created w/ resolution only down to the second.
We thankfully have a shim layer where we can perform this for as well, but would love to see this merged to make our testing a little easier

[edennis]

Any chance of getting this merged? 🙏🏻