DST handling is not inline with pytz dst handling and most probably incorrect

[bernhardboehmer]

Issue Description

The handling of the dst switch does not seem to be right. While changing from dst to non-dst in October, the 2am hour is repeated in local time. However, in terms of time zones the repeated hour should not be utc+2 but utc+1 as pytz is doing as well. Otherwise, e.g., checks for time gaps in timeseries will work incorrectly since we have twice 02:20+2 for instance. Please look at this excerpt from the python console (sorry didn't get formatting too well):

from pytz import timezone
berlin = timezone('Europe/Berlin')
arrow.get(datetime.datetime(2017, 10, 29, 1, 0, 0, tzinfo=utc)).to('Europe/Berlin')
       ->   _<Arrow [2017-10-29T02:00:00+02:00]>_
arrow.get(datetime.datetime(2017, 10, 29, 0, 0, 0, tzinfo=utc)).to('Europe/Berlin')
       ->  _<Arrow [2017-10-29T02:00:00+02:00]>_
datetime.datetime(2017, 10, 29, 1, 0, 0, tzinfo=utc).astimezone(berlin)
      -> _datetime.datetime(2017, 10, 29, 2, 0, tzinfo=<DstTzInfo 'Europe/Berlin' CET+1:00:00 STD>)_
datetime.datetime(2017, 10, 29, 0, 0, 0, tzinfo=utc).astimezone(berlin)
      -> _datetime.datetime(2017, 10, 29, 2, 0, tzinfo=<DstTzInfo 'Europe/Berlin' CEST+2:00:00 DST>)_

System Info

• 🖥 OS name and version: macOS 10.14.6
• 🐍 Python version: python 3.7.6
• 🏹 Arrow version: 0.15.8

[systemcatch]

Hi @bernhardboehmer this type of bug was fixed in version 0.17.0.

(arrow) chris@ThinkPad:~/arrow$ python
Python 3.8.3 (default, Jul  7 2020, 18:57:36) 
[GCC 9.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import arrow
>>> arrow.__version__
'0.17.0'
>>> import datetime
>>> arrow.get(datetime.datetime(2017, 10, 29, 1, 0, 0), tzinfo="UTC").to('Europe/Berlin')
<Arrow [2017-10-29T02:00:00+01:00]>
>>> arrow.get(datetime.datetime(2017, 10, 29, 0, 0, 0), tzinfo="UTC").to('Europe/Berlin')
<Arrow [2017-10-29T02:00:00+02:00]>

We added support for PEP 495 (fold attribute) to disambiguate ambiguous times along with a new property.

>>> before=arrow.Arrow(2017, 10, 29, 2, tzinfo="Europe/Berlin")
>>> before
<Arrow [2017-10-29T02:00:00+02:00]>
>>> before.ambiguous
True
>>> after=arrow.Arrow(2017, 10, 29, 2, tzinfo="Europe/Berlin", fold=1)
>>> after
<Arrow [2017-10-29T02:00:00+01:00]>

[bernhardboehmer]

Excellent! I missed this new version, sorry for this. In terms of comparision this getting more difficult. I use timestamp for this. Any better alternative?

[systemcatch]

I'm not sure what you mean by comparison, maybe a code example would help?

[bernhardboehmer]

Hi systemcatch, thanks for your reply. Here's a code example. The method receives a set of localized arrow-timestamps and it shall check whether there are gaps within. Offset (period) can be set freely. The nodes are the complete list of arrow-timestamps between start and end. The gaps found do not include "end".

The second line (the set statement) is BTW the location triggering my initial bug report.

The final list comprehension is the location where the check failed with plain arrow objects, that's why I switched to ".timestamp".

Regards Bernhard

    def get_missing_timestamps(self, start, end, timestamps):
        """
        tests if the list of timestamps forms an arithmetic series, of the form 
        t_i = t_i-1 + offset, by comparing with a given arithmetic series, obtained by
        UTCTimePeriod.get_offsets()
        Args:
         start: arrow.Arrow
         end: arrow.Arrow
         offset: integer
         timestamps: a list of timestamps with timezones sic [arrow.Arrow(a, b, c, tzinfo='utc'), ...]
        Returns:
         a list of missing timestamps
        """

        timestamps = set(timestamp.timestamp for timestamp in timestamps)
        nodes = self.get_offsets(start, end)
        return [node for node in nodes if node.timestamp not in timestamps]