GH1180 Clean up all/any methods for Series and DataFrame #1188
Conversation
pandas-stubs/core/frame.pyi
Outdated
| @@ -1660,7 +1660,8 @@ class DataFrame(NDFrame, OpsMixin, _GetItemHack): | |||
| bool_only: _bool | None = ..., | |||
| skipna: _bool = ..., | |||
| **kwargs: Any, | |||
| ) -> _bool: ... | |||
| ) -> np.bool: ... | |||
| # FIXME the type below is not correct, should be pd.Series[np.bool] | |||
There was a problem hiding this comment.
The way I look at it is that we are using Series[bool] to correspond to whatever bool is stored inside - being a python one, a numpy one, or even if we use BooleanDtype, so I don't think this comment is necessary.
There was a problem hiding this comment.
Actually the type checker will not accept that, because pd.Series can only be subscribed with S1 which contains bool (the generic boolean from python) but not np.bool.
Yet pd.DataFrame.any will return at runtime pd.Series[np.bool] (but this is not accepted since np.bool is not a subtype of S1).
Let me know if this is not clear,
There was a problem hiding this comment.
The type returned from DataFrame.any() and DataFrame.all() should be Series[_bool] . Then the tests should use that.
There was a problem hiding this comment.
My issue is that _bool does not contain np.bool which is the type we get at runtime. Happy to keep the stubs as is but that would mean that runtime type does not align with static type.
Or we can add np.bool to _bool, open to suggestion.
There was a problem hiding this comment.
I don't think it matters. We have checks like this:
check(assert_type(df.any(), "pd.Series[bool]"), pd.Series, np.bool_)So even though np.bool_ is in the Series, we call the type of the Series[bool]
It's similar to this:
check(assert_type(df.value_counts(), "pd.Series[int]"), pd.Series, np.integer)What's inside the series are numpy integers, but we call that Series[int]
There was a problem hiding this comment.
Okay I see your point, let me fix the comments, thanks for the more detailed vision on this issue!
assert_type()to assert the type of any return valueTechnically we should fix the
pd.DataFrame.allbut that would mean having apd.Series[np.bool], yetnp.boolis not a type ofS1so will leave it there for a bit, it has some FIXME statements there for us to know.