Fixtures making IntNA tests difficult to run interactively

[jbrockmendel]

Trying to run tests.arrays.test_integer TestArithmeticOps.test_error in the REPL, need to produce the data arg that gets passed to test_error. data is defined in a fixture:

@pytest.fixture(params=[Int8Dtype, Int16Dtype, Int32Dtype, Int64Dtype,
                        UInt8Dtype, UInt16Dtype, UInt32Dtype, UInt64Dtype])
def dtype(request):
    return request.param()


@pytest.fixture
def data(dtype):
    return integer_array(make_data(), dtype=dtype)

Trying to do this manually:

>>> from pandas.tests.arrays.test_integer import *
>>> dtype = Int64Dtype
>>>
>>> d = make_data()
>>> dat = integer_array(d, dtype=dtype)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "pandas/core/arrays/integer.py", line 104, in integer_array
    values, mask = coerce_to_array(values, dtype=dtype, copy=copy)
  File "pandas/core/arrays/integer.py", line 159, in coerce_to_array
    raise ValueError("invalid dtype specified {}".format(dtype))
ValueError: invalid dtype specified <class 'pandas.core.arrays.integer.Int64Dtype'>

What am I missing here? Figuring out exactly what is tested should be really easy.

[TomAugspurger]

You need to call the dtype.

dat = integer_array(d, dtype=dtype())

I think having an actual repr for instances of that dtype may help. with the confusion. We could also check in coerce_to_array whether the dtype is a class, rather than an instance.

[jbrockmendel]

thanks

[TomAugspurger]

Might leave this open to see how things can be improved.

I think part of the confusion is that NumPy's top-level np.<dtype> functions / classes are kinda weird. You're expected to pass the class in places where we expect the dtypes.

pd.Series([1, 2], dtype=np.int32)

rather than np.int32() which returns... something.

pyarrow's pa.array() expects instances of its top-level types, rather than the class, so we're consistent with pyarrow here.