BUG: MultiIndex.putmask losing ea dtype #49847
Conversation
| @@ -162,6 +162,34 @@ def test_putmask_multiindex_other(self): | |||
| expected = MultiIndex.from_tuples([right[0], right[1], left[2]]) | |||
| tm.assert_index_equal(result, expected) | |||
|
|
|||
| def test_putmask_keep_dtype(self, any_numeric_ea_dtype): | |||
There was a problem hiding this comment.
Are the two test cases also covering for the all-nan case in #49830 (comment) or do you think it's not necessary?
There was a problem hiding this comment.
It's about losing the dtype, so yes this covers the all nan case. But this does not fix the issue yet
There was a problem hiding this comment.
The variable might be a misnomer then any_numeric_ea_dtype (if it covers all nans, this would also include non EA dtype).
There was a problem hiding this comment.
It covers the case in a sense that the current tests represent the all nan case as well, I’ll add a specific test when actually fixing the issue anyway
pandas/core/indexes/multi.py
Outdated
| zip(subset.levels, self.levels, self.codes) | ||
| ): | ||
| new_elements = value_level.difference(level) | ||
| new_level = level.append(new_elements) |
There was a problem hiding this comment.
Can the two lines above be replaced with new_level = level.union(value_level)?
There was a problem hiding this comment.
Good point, was thinking about duplicates when implementing this, but levels are unique…
|
failure unrelated |
|
Thanks @phofl |
* BUG: MultiIndex.putmask losing ea dtype * Fix typing * Add asv * Simplify and add whatsnew
combine_first()coerces toobjectforMultiIndexand all-nan index columns #49830 (Replace xxxx with the GitHub issue number)doc/source/whatsnew/vX.X.X.rstfile if fixing a bug or adding a new feature.This is something we have to fix before we can address the issue
Its also significantly faster