ENH: allow timestamp subtraction with different timezones (GH15249) #15329
Conversation
pandas/tseries/tests/test_base.py
Outdated
| @@ -816,14 +816,14 @@ def _check(result, expected): | |||
| _check(result, expected) | |||
|
|
|||
| # tz mismatches | |||
| self.assertRaises(TypeError, lambda : dt_tz - ts) | |||
There was a problem hiding this comment.
Instead of commenting out those tests, you will need to adapt the test to check that all those cases actually return the correct value
pandas/tslib.pyx
Outdated
| if self.tzinfo!="UTC": | ||
| self.tz_convert("UTC") | ||
| if other.tzinfo!="UTC": | ||
| other.tz_convert("UTC") |
There was a problem hiding this comment.
I don't think it is needed to convert both to UTC, as the underlying value (the ones that get subtracted), already are stored as UTC. Can you check that?
There was a problem hiding this comment.
If the conversion is actually needed, you can use _is_utc.
jorisvandenbossche
changed the title
jorisvandenbossche
added
Timezones
pandas/tslib.pyx
Outdated
| @@ -1052,8 +1056,11 @@ cdef class _Timestamp(datetime): | |||
| other = Timestamp(other) | |||
|
|
|||
| # validate tz's | |||
| if get_timezone(self.tzinfo) != get_timezone(other.tzinfo): | |||
| raise TypeError("Timestamp subtraction must have the same timezones or no timezones") | |||
| if get_timezone(self.tzinfo) != get_timezone(other.tzinfo): | |||
There was a problem hiding this comment.
this is not enough, you need to simply do the conversion to 'UTC'. so an already UTC tz will be fine, a non-UTC zone will be converted as well, BUT a non-timezone aware will raise, which should be raised to a higher level.
|
If we allow this, we should also fix related |
|
@Vutsuak16 you apparently changed all of the whitespace, so you have a massive diff. Further the most important thing here is to nail down the tests to figure out what is changing. |
|
can you update |
|
closing this as stale, but if you'd like to add more tests / update pls commentt. |
works on #15249
passes the relevant tests but fails on few. Those include numpy dependency libgfortran and some environment issues. The output works just fine.
@jreback