BUG: Timestamp(Timestamp(Ambiguous time)) modifies .value with dateutil tz · Issue #24329 · pandas-dev/pandas (original) (raw)

Pretty obscure bug, but this seems fishy:

In [7]: pd.__version__
Out[7]: '0.24.0.dev0+1300.ge0a68076a.dirty'

# Ambiguous time
In [8]: t = pd.Timestamp(1382835600000000000, tz='dateutil/Europe/London')

# Repr is consistent
In [11]: t
Out[11]: Timestamp('2013-10-27 01:00:00+0100', tz='dateutil//usr/share/zoneinfo/Europe/London')

In [12]: pd.Timestamp(t)
Out[12]: Timestamp('2013-10-27 01:00:00+0100', tz='dateutil//usr/share/zoneinfo/Europe/London')

# .value changes
In [13]: t.value
Out[13]: 1382835600000000000

In [14]: pd.Timestamp(t).value
Out[14]: 1382832000000000000

pytz timezones behave consistently though

In [15]: t = pd.Timestamp(1382835600000000000, tz='Europe/London')

In [16]: t
Out[16]: Timestamp('2013-10-27 01:00:00+0000', tz='Europe/London')

In [17]: pd.Timestamp(t)
Out[17]: Timestamp('2013-10-27 01:00:00+0000', tz='Europe/London')

In [18]: t.value
Out[18]: 1382835600000000000

In [19]: pd.Timestamp(t).value
Out[19]: 1382835600000000000

The fact that the repr between dateutil timezones and pytz timezones don't match can be possible be seen in a change in dateutil somewhere around 2.6? But the main issue that is .value changes.