TYP: remove mypy ignore from pandas/core/construction.py

[natmokval]

Related to #37715
mypy ignore[assignment] was removed from pandas/core/construction.py

[MarcoGorelli]

thanks for your PR - would it work to just rename data? looks like the issue that it's getting reassigned to a different type

something like

    data_newname = data.astype(dtype, copy=True)
    data_newname.soften_mask()
    data_newname[mask] = fill_value
    return data_newname
else:
    return data.copy()

(the hardest part would arguably be coming up with a good name, i.e. not data_newname but something more descriptive, but I haven't looked at this carefully enough yet to think of what)

[natmokval]

Thank you, @MarcoGorelli for your comment. I will do, as you suggest. What do you think, could be sanitized_data or sanitized_ma a suitable new name for data?

[MarcoGorelli]

looks good to me, leaving open a bit in case others have comments

EDIT: oops, not green

[natmokval]

Looks like mypy raises an error, I will fix it.

[MarcoGorelli]

I think the issue might be that astype isn't meant to be called on numpy.ma.MaskedArray https://numpy.org/doc/stable/reference/routines.ma.html

Calling it, the return type is typed to be ndarray, even though if we print it we see it's numpy.ma.MaskedArray:

data = np.ma.MaskedArray([1,2,3], mask=[False, False, True])
reveal_type(data)  # Revealed type is "numpy.ma.core.MaskedArray[Any, Any]"
reveal_type(data.astype('int64', copy=True))  # Revealed type is "numpy.ndarray[Any, numpy.dtype[Any]]"

Might be worth checking if there's an issue in numpy about it, and if not, then reporting to them? Perhaps this could be made more precise on their end?

The method which says "Convert the input to a masked array of the given data-type" is the one you'd originally used: numpy.ma.asarray, so your original solution was probably the correct one, sorry for the noise 🙈 The docs say it that asarray doesn't copy if the original data is an ndarray anyway

[natmokval]

I think the issue might be that astype isn't meant to be called on numpy.ma.MaskedArray https://numpy.org/doc/stable/reference/routines.ma.html

Calling it, the return type is typed to be ndarray, even though if we print it we see it's numpy.ma.MaskedArray:

data = np.ma.MaskedArray([1,2,3], mask=[False, False, True])
reveal_type(data)  # Revealed type is "numpy.ma.core.MaskedArray[Any, Any]"
reveal_type(data.astype('int64', copy=True))  # Revealed type is "numpy.ndarray[Any, numpy.dtype[Any]]"

Might be worth checking if there's an issue in numpy about it, and if not, then reporting to them? Perhaps this could be made more precise on their end?

Thanks, I'll check if there's an issue on GitHub about it.

The method which says "Convert the input to a masked array of the given data-type" is the one you'd originally used: numpy.ma.asarray, so your original solution was probably the correct one, sorry for the noise 🙈 The docs say it that asarray doesn't copy if the original data is an ndarray anyway

I rolled back the commit to the first one.

[MarcoGorelli]

let's get this in, thanks @natmokval