REF: move Block.astype implementation to dtypes/cast.py #40141
Conversation
jorisvandenbossche
added
Refactor
pandas/core/array_algos/cast.py
Outdated
| return values | ||
|
|
||
|
|
||
| def astype_array_safe(values, dtype, copy: bool = False, errors: str = "raise"): |
There was a problem hiding this comment.
values: ArrayLike; dtype: DtypeObj?
There was a problem hiding this comment.
mypy isn't smart enough for that (we also didn't annotate dtype in the code from where I copied this). But will add values: ArrayLike
There was a problem hiding this comment.
make sense. if we move the pandas_dtype call up into the caller we can do it once instead of (block|array)-wise
There was a problem hiding this comment.
if we move the pandas_dtype call up into the caller we can do it once instead of (block|array)-wise
Then also the dtype validation needs to be moved up (requiring some more changes to avoid duplication of that part). At the moment, it's a rather straightfoward cut and paste from blocks.py to cast.py, so I would maybe prefer to keep it that way for this PR.
| dtype = pandas_dtype(dtype) | ||
| values = self.values | ||
| if values.dtype.kind in ["m", "M"]: | ||
| values = self.array_values() |
There was a problem hiding this comment.
could move this into astype_array_safe and use ensure_wrapped_if_datetimelike; would make it robust to AM/BM (though i think both AM and BM now have PRs to make the arrays EAs to begin with)
There was a problem hiding this comment.
Since ArrayManager already stores it as EAs (after this array), I would prefer to leave it here (then your PR changing to store EAs in BlockManager as well can remove those two lines)
pandas/core/dtypes/cast.py
Outdated
| @@ -1225,6 +1227,107 @@ def astype_nansafe( | |||
| return arr.astype(dtype, copy=copy) | |||
|
|
|||
|
|
|||
| def astype_array(values: ArrayLike, dtype: DtypeObj, copy: bool = False): | |||
There was a problem hiding this comment.
can you add a return annotation
pandas/core/dtypes/cast.py
Outdated
|
|
||
| def astype_array_safe( | ||
| values: ArrayLike, dtype, copy: bool = False, errors: str = "raise" | ||
| ): |
jorisvandenbossche
changed the title
This moves the
astypeimplementation (so the current logic of callingEA.astypeorastype_nansafedepending on the array type, handling datetime and string special cases, handling error keyword, etc) from being defined onBlockto thearray_algos/submodule.That's a useful clean-up anyway I think (it's not block specific), and this way it can then be reused in ArrayManager.
I currently put it in
core/array_algos/cast.pysince the other similar functions are also put inarray_algos(shift,putmask,replace,quantileetc). But since this is casting specific and we already have most of the underlying pieces incore/dtypes/cast.py, could also just put it there (either is fine for me).