Failing tests (Elixir 1.13)

[vbaranov]

Steps to reproduce

mix test

Description of issue

elixir --version
Erlang/OTP 24 [erts-12.3.2.1] [source] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads:1] [jit]

Elixir 1.13.4 (compiled with Erlang/OTP 24)

Constantly failing tests on branch main (e1c4d42):

....................................................................................................................................................

  1) test now now/0 returns the duration since epoch (DurationTests)
     test/duration_test.exs:137
     Expected the difference between -82 and 500000 (500082) to be less than or equal to 500001
     code: assert_in_delta(difference_microseconds, 500_000, 500_001)
     stacktrace:
       test/duration_test.exs:141: (test)

................................................................................................................................................................................................

  2) test format RFC3339 (DateFormatTest.FormatDefault)
     test/format_default_test.exs:338
     Assertion with == failed
     code:  assert format(date, "{RFC3339}") == {:ok, "0010-09-26T07:00:02+00:50"}
     left:  {:ok, "0010-09-26T07:00:02+00:53"}
     right: {:ok, "0010-09-26T07:00:02+00:50"}
     stacktrace:
       test/format_default_test.exs:360: (test)

[gaiabeatrice]

I can't get the first test to fail, but for test/format_default_test.exs:338 I found out about the following:

If (from within this library) I run

naive = ~N[0010-09-26T07:00:02]
timezone = "Europe/Copenhagen"

Tzdata.TimeZoneDatabase.time_zone_periods_from_wall_datetime(
  naive,
  timezone
)

Which is a function called by Timex.Timezone.Database.time_zone_periods_from_wall_datetime/2, I get

{:ok,
 %{
   from_wall: :min,
   std_offset: 0,
   until_wall: ~N[1893-04-01 00:00:00],
   utc_offset: 3208,
   zone_abbr: "LMT"
 }}

If I convert these 3208 seconds to a more readable format (hh:mm:ss), I get 00:53:28, which is the wrong offset that makes the test fail (+00:53). I think something changed in the database, so this value is now different.

I researched timezone changes in Denmark and I found out the following

The first time a common time was used in Denmark, was in 1890, when Copenhagen local time was used as railway time. This was GMT+0:50:20 from Greenwich.

So it does seem like +00:50 is the correct one and the test shouldn't be changed. Unless new research has happened and the database was updated accordingly.

I think this warrants more research on Tzdata.