BUG/DEPR: combine dtype fixes #13970
Conversation
sinhrks
added
Bug
Reshaping
| self.assertEqual(_find_common_type([dtype, dtype]), 'category') | ||
| self.assertEqual(_find_common_type([np.object, dtype]), np.object) | ||
|
|
||
| dtype = DatetimeTZDtype(unit='ns', tz='US/Eastern') |
There was a problem hiding this comment.
add a test for a datetime64[ns] w/tz against datetime64[ns]. (and maybe with np.object)
|
xref to #4094 (though that's another story) |
|
very minor change. lgtm otherwise. ping when green. |
Current coverage is 85.28% (diff: 100%)@@ master #13970 diff @@
==========================================
Files 139 139
Lines 50216 50206 -10
Methods 0 0
Messages 0 0
Branches 0 0
==========================================
+ Hits 42817 42818 +1
+ Misses 7399 7388 -11
Partials 0 0
|
|
Not really sure we should actually deprecate |
|
I've commented on the original (#10734). |
|
ok, so as discussed in the issue. let's take out the combine deprecation from this PR. ping when ready. |
|
Exclude |
|
thanks! |
DataFrame.combine_first was fixed in pandas-dev#13970. Fix Series using same logic. Remove common._where_compat since it was only referred to by Series.combine_first, name was confusing, and not necessary to the new solution.
combine_firstdtype)_find_common_typeto support extension types)git diff upstream/master | flake8 --diff_find_common_typealso needs to handle numpy datetimelike to be compat with pandas because:comibine_firstrelated issue combine_first not retaining dtypes #7509 is NOT fixed because it is related to filling withalign