TST Test failures with OpenML for PyPy

[alfaro96]

The following tests are failing for PyPy:

FAILED datasets/tests/test_openml.py::test_fetch_openml_iris_pandas - TypeErr...
FAILED datasets/tests/test_openml.py::test_fetch_openml_iris_pandas_equal_to_no_frame
FAILED datasets/tests/test_openml.py::test_fetch_openml_iris_multitarget_pandas
FAILED datasets/tests/test_openml.py::test_fetch_openml_anneal_pandas - TypeE...
FAILED datasets/tests/test_openml.py::test_fetch_openml_cpu_pandas - TypeErro...
FAILED datasets/tests/test_openml.py::test_fetch_openml_as_frame_auto - TypeE...
FAILED datasets/tests/test_openml.py::test_convert_arff_data_dataframe_warning_low_memory_pandas
FAILED datasets/tests/test_openml.py::test_fetch_openml_adultcensus_pandas_return_X_y
FAILED datasets/tests/test_openml.py::test_fetch_openml_adultcensus_pandas - ...
FAILED datasets/tests/test_openml.py::test_fetch_openml_miceprotein_pandas - ...
FAILED datasets/tests/test_openml.py::test_fetch_openml_emotions_pandas - Typ...
FAILED datasets/tests/test_openml.py::test_fetch_openml_titanic_pandas - Type...
FAILED datasets/tests/test_openml.py::test_fetch_openml_verify_checksum[True]

with the following error traceback:

2020-11-24T16:19:24.2263007Z self = RangeIndex(start=0, stop=1, step=1)
2020-11-24T16:19:24.2263373Z 
2020-11-24T16:19:24.2263663Z     @cache_readonly
2020-11-24T16:19:24.2264148Z     def nbytes(self) -> int:
2020-11-24T16:19:24.2264461Z         """
2020-11-24T16:19:24.2264877Z         Return the number of bytes in the underlying data.
2020-11-24T16:19:24.2265280Z         """
2020-11-24T16:19:24.2265578Z         rng = self._range
2020-11-24T16:19:24.2265961Z >       return getsizeof(rng) + sum(
2020-11-24T16:19:24.2266404Z             getsizeof(getattr(rng, attr_name))
2020-11-24T16:19:24.2266869Z             for attr_name in ["start", "stop", "step"]
2020-11-24T16:19:24.2267216Z         )
2020-11-24T16:19:24.2267567Z E       TypeError: getsizeof(...)
2020-11-24T16:19:24.2268134Z E           getsizeof(object, default) -> int
2020-11-24T16:19:24.2268491Z E       
2020-11-24T16:19:24.2268855Z E           Return the size of object in bytes.
2020-11-24T16:19:24.2269205Z E       
2020-11-24T16:19:24.2269695Z E       sys.getsizeof(object, default) will always return default on PyPy, and
2020-11-24T16:19:24.2270318Z E       raise a TypeError if default is not provided.
2020-11-24T16:19:24.2270721Z E       
2020-11-24T16:19:24.2271186Z E       First note that the CPython documentation says that this function may
2020-11-24T16:19:24.2271859Z E       raise a TypeError, so if you are seeing it, it means that the program
2020-11-24T16:19:24.2272447Z E       you are using is not correctly handling this case.
2020-11-24T16:19:24.2272840Z E       
2020-11-24T16:19:24.2273307Z E       On PyPy, though, it always raises TypeError.  Before looking for
2020-11-24T16:19:24.2274004Z E       alternatives, please take a moment to read the following explanation as
2020-11-24T16:19:24.2274687Z E       to why it is the case.  What you are looking for may not be possible.
2020-11-24T16:19:24.2275112Z E       
2020-11-24T16:19:24.2275573Z E       A memory profiler using this function is most likely to give results
2020-11-24T16:19:24.2276218Z E       inconsistent with reality on PyPy.  It would be possible to have
2020-11-24T16:19:24.2276874Z E       sys.getsizeof() return a number (with enough work), but that may or
2020-11-24T16:19:24.2277656Z E       may not represent how much memory the object uses.  It doesn't even
2020-11-24T16:19:24.2278343Z E       make really sense to ask how much *one* object uses, in isolation
2020-11-24T16:19:24.2278956Z E       with the rest of the system.  For example, instances have maps,
2020-11-24T16:19:24.2279765Z E       which are often shared across many instances; in this case the maps
2020-11-24T16:19:24.2280459Z E       would probably be ignored by an implementation of sys.getsizeof(),
2020-11-24T16:19:24.2281121Z E       but their overhead is important in some cases if they are many
2020-11-24T16:19:24.2281768Z E       instances with unique maps.  Conversely, equal strings may share
2020-11-24T16:19:24.2282610Z E       their internal string data even if they are different objects---or
2020-11-24T16:19:24.2283287Z E       empty containers may share parts of their internals as long as they
2020-11-24T16:19:24.2284099Z E       are empty.  Even stranger, some lists create objects as you read
2020-11-24T16:19:24.2284800Z E       them; if you try to estimate the size in memory of range(10**6) as
2020-11-24T16:19:24.2285661Z E       the sum of all items' size, that operation will by itself create one
2020-11-24T16:19:24.2286395Z E       million integer objects that never existed in the first place.

I will open a PR marking them as XFAIL to not block #18879 but I think that we should investigate the underlaying issue.

[thomasjpfan]

As a data point, I was not able to reproduce this issue on pypy3.6 on OSX using pypy3.6 from conda-forge.

[lesteve]

Closing after PyPy official support has been dropped #29128