CI/COMPAT: read_stata failing in numpy dev pipeline

[AlexKirko]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Problem description

Looks like a change in numpy broke our read_stata code, so now we have a bunch of read_stata test errors popping up during CI in the numpy_dev pipeline. Maybe this Numpy PR somehow broke things? I don't think any other PR merged during the last 24 hours could have possibly done so, although I might be missing something.

Error example:

pandas/tests/io/test_stata.py:52: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pandas/io/stata.py:1928: in read_stata
    data = reader.read()
pandas/io/stata.py:1646: in read
    cols_ = np.where(self.dtyplist)[0]
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = ([<class 'numpy.int8'>, <class 'numpy.int16'>, <class 'numpy.int32'>, <class 'numpy.float32'>, <class 'numpy.float64'>, <class 'numpy.float32'>, ...],)
kwargs = {}
relevant_args = ([<class 'numpy.int8'>, <class 'numpy.int16'>, <class 'numpy.int32'>, <class 'numpy.float32'>, <class 'numpy.float64'>, <class 'numpy.float32'>, ...], None, None)

>   ???
E   ValueError: invalid __array_struct__

[jreback]

cc @bashtage

[bashtage]

That is a NumPy bug. numpy/numpy#16939

[bashtage]

Fixed it anyway. No reason to create an array of dtypes.

[AlexKirko]

@bashtage , thanks!
Looks like there are also a bunch of other places where we create an array of dtypes and then assign them to a numpy array with something like result[:] = values.

[AlexKirko]

should we alter this issue to incorporate other test fails (basically anywhere we create a list of dtypes and assign it to a numpy array) or open a new issue?
Nevermind, I somehow missed that it's basically an upstream bug and not intended numpy behavior. numpy/numpy#16941 should fix this. Thanks, @TomAugspurger !

[TomAugspurger]

We've already opened an issue upstream: numpy/numpy#16939

[simonjayhawkins]

closing in favour of #35481 as Stata tests no longer failing