BUG: Correct Timestamp localization with tz near DST (#11481) #15934
Conversation
mroeschke
changed the title
Codecov Report
@@ Coverage Diff @@
## master #15934 +/- ##
==========================================
+ Coverage 90.99% 90.99% +<.01%
==========================================
Files 145 145
Lines 49570 49570
==========================================
+ Hits 45106 45107 +1
+ Misses 4464 4463 -1
Continue to review full report at Codecov.
|
pandas/_libs/tslib.pyx
Outdated
| @@ -1569,7 +1569,10 @@ cpdef convert_str_to_tsobject(object ts, object tz, object unit, | |||
| ts = obj.value | |||
| if tz is not None: | |||
| # shift for _localize_tso | |||
| ts = tz_convert_single(ts, tz, 'UTC') | |||
| #ts = tz_convert_single(ts, tz, 'UTC') | |||
| ts = tz_localize_to_utc(np.array([ts], dtype='i8'), tz, | |||
There was a problem hiding this comment.
hmm there as to be an error in tz_convert_single. though it might simply be best to just replace that function to a call exactly like this, IOW (and see if anything else breaks)
cdef tz_convert_single(....)
return tz_localize_to_utc(.....)[0]
| @@ -159,6 +159,19 @@ def test_timestamp_constructed_by_date_and_tz_explicit(self): | |||
| self.assertEqual(result.hour, expected.hour) | |||
| self.assertEqual(result, expected) | |||
|
|
|||
| def test_timestamp_constructor_near_dst_boundary(self): | |||
| # GH 11481 & 15777 | |||
There was a problem hiding this comment.
can you give a 1 sentence on what the error was
|
can you scour the xref'ed issues for any more test cases and see if this closes any linked issues? (if you did already then ok). |
mroeschke
force-pushed
the
fix_15777
branch
2 times, most recently
from
c03e06e
to
848c383
Compare
|
Added some additional tests and described the error. Addtionally, I tried dropping in |
TST: more tests, xref pandas-dev#15823, xref pandas-dev#11708
|
thanks @mroeschke added some more tests. will merge on pass. |
git diff upstream/master --name-only -- '*.py' | flake8 --diffSeemed that the
Timestampconstructor with a string incorrectly localized a naive time withtz_convert_singleinstead oftz_localize_to_utcwhich is similar to howDatetimeIndexlocalizes.Now the
Timestampwill raise an error if an ambiguous combination of string and timezone is passed. Should this be the default behavior?I had to slightly change 2 existing tests.
test_timestamp_to_datetime_tzoffsetused a now ambiguous timestamp andtest_setitem_with_tz_dsthad an incorrect expected timestamp.