[ArrowStringArray] BUG: fix test_astype_string for Float32Dtype #40998
Conversation
pandas/core/arrays/string_arrow.py
Outdated
| # numerical issues with Float32Dtype | ||
| na_values = scalars._mask | ||
| result = scalars._data | ||
| result = lib.ensure_string_array(result, copy=False, convert_na_value=False) |
There was a problem hiding this comment.
umm this doesn't copy? (its fine but then you are mutating)
There was a problem hiding this comment.
that's not a change in this PR. will need to look back why we do this.
There was a problem hiding this comment.
updated to be consistent with StringArray. I guess that the assumption was that a copy would probably have been created in most cases (unless an object array was passed) and explicitly avoid the additional copy made in ensure_string_array since that does not track if a copy/new array has been already been created before making a copy. The default for copy is False.
pandas/core/arrays/string_arrow.py
Outdated
| na_values = scalars._mask | ||
| result = scalars._data | ||
| result = lib.ensure_string_array(result, copy=False, convert_na_value=False) | ||
| result[na_values] = ArrowStringDtype.na_value |
There was a problem hiding this comment.
Alternatively, might also be able to do pa.array(result, mask=mask, type=pa.string()), then pyarrow will use the mask for missing values, and we don't need the setitem operation.
jorisvandenbossche
changed the title
|
thanks @simonjayhawkins |
this also fixes Interval, but that test doesn't fail in #39908 for a single StringDtype parameterized by storage.