issue_comments

yeah, I don't know how to only filter for pint warnings, I tried pytest.mark.filterwarnings("error:::pint[.*]") but that doesn't work

Edit: pytest.mark.filterwarnings("error::pint.UnitStrippedWarning") works so I'm merging.

Looks like an incompatibility with latest pandas

``` ____ TestDataset.test_resample[int-coords] _______

self = <xarray.tests.test_units.TestDataset object at 0x7fa84fb9e490> variant = 'coords', dtype = <class 'int'> return func(all_args, *all_kwargs) xarray/core/common.py:1123: in resample grouper = pd.Grouper(

cls = <class 'pandas.core.groupby.grouper.Grouper'>, args = () kwargs = {'base': 0, 'closed': None, 'freq': '6m', 'label': None, ...} TimeGrouper = <class 'pandas.core.resample.TimeGrouper'>, stacklevel = 2

def __new__(cls, *args, **kwargs):
    if kwargs.get("freq") is not None:
        from pandas.core.resample import TimeGrouper

        # Deprecation warning of `base` and `loffset` since v1.1.0:
        # we are raising the warning here to be able to set the `stacklevel`
        # properly since we need to raise the `base` and `loffset` deprecation
        # warning from three different cases:
        #   core/generic.py::NDFrame.resample
        #   core/groupby/groupby.py::GroupBy.resample
        #   core/groupby/grouper.py::Grouper
        # raising these warnings from TimeGrouper directly would fail the test:
        #   tests/resample/test_deprecated.py::test_deprecating_on_loffset_and_base

        # hacky way to set the stacklevel: if cls is TimeGrouper it means
        # that the call comes from a pandas internal call of resample,
        # otherwise it comes from pd.Grouper
        stacklevel = 4 if cls is TimeGrouper else 2
        if kwargs.get("base", None) is not None:

          warnings.warn(
                "'base' in .resample() and in Grouper() is deprecated.\n"
                "The new arguments that you should use are 'offset' or 'origin'.\n"
                '\n>>> df.resample(freq="3s", base=2)\n'
                "\nbecomes:\n"
                '\n>>> df.resample(freq="3s", offset="2s")\n',
                FutureWarning,
                stacklevel=stacklevel,
            )

E FutureWarning: 'base' in .resample() and in Grouper() is deprecated. E The new arguments that you should use are 'offset' or 'origin'. E
E >>> df.resample(freq="3s", base=2) E
E becomes: E
E >>> df.resample(freq="3s", offset="2s") ```

I just found another issue: pint implements prod (but not yet nanprod) so the prod tests could be un-xfailed. However, we define a custom nanprod function that uses where to replace nan with 1. This won't work on quantities since, unlike nan and 0, a bare 1 cannot be put into quantities with a dimension (i.e. with a unit other than dimensionless).

I don't really understand the purpose of nanprod's min_count parameter (and _maybe_null_out) so I'm not sure how to fix that.

For now, I think we can merge this PR on green and I'll add that issue to the list in #3594.

Thanks for working on this @keewis . Since there are no changes outside test_units.py, I think you should merge whenever you think this is ready.

it seems assert_allclose was one of the sources of UnitStrippedWarnings, but since there's a bug in pint's isclose (fixed on master) the tests fail now.

Edit: pint will be released in the next few days, so most of the failing CI should pass after that.

Also, because it casted to numpy, there were a few bugs that were hidden.