Migrate VariableArithmetic to NamedArrayArithmetic

[andersy005]

• towards NamedArray tracking issue #8238
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[Illviljan]

Could you do dtypes.py in separate PR @andersy005 ? That one can be merged quite easily I think.

[dcherian]

@andersy005 what are your thoughts on migrating tests for this.

In theory it seems like we'd want to test both Variable and NamedArray, but I also don't like the idea of them potentially diverging once we switch to a new repo. OTOH perhaps there won't be much updating of such core functionality.

cc @pydata/xarray for opinions

[andersy005]

Could you do dtypes.py in separate PR @andersy005 ? That one can be merged quite easily I think.

• See copy the dtypes module to the namedarray package. #8250

[TomNicholas]

In theory it seems like we'd want to test both Variable and NamedArray, but I also don't like the idea of them potentially diverging once we switch to a new repo. OTOH perhaps there won't be much updating of such core functionality.

Maybe we could avoid divergence by importing from the NamedArray test suite and using the same arithmetic tests but on the Variable class?

[andersy005]

In theory it seems like we'd want to test both Variable and NamedArray, but I also don't like the idea of them potentially diverging once we switch to a new repo. OTOH perhaps there won't be much updating of such core functionality.

Maybe we could avoid divergence by importing from the NamedArray test suite and using the same arithmetic tests but on the Variable class?

i'm of the same opinion as @TomNicholas that we can package the shared tests in namedarray, and have the tests for Variable inherit those tests while also including their own specific tests.

[maxrjones]

In theory it seems like we'd want to test both Variable and NamedArray, but I also don't like the idea of them potentially diverging once we switch to a new repo. OTOH perhaps there won't be much updating of such core functionality.

Maybe we could avoid divergence by importing from the NamedArray test suite and using the same arithmetic tests but on the Variable class?

i'm of the same opinion as @TomNicholas that we can package the shared tests in namedarray, and have the tests for Variable inherit those tests while also including their own specific tests.

It should work to use pytest_collection_modifyitems to collect tests from namedarray that should be also tested in xarray without modification. It's slightly more complicated to add external tests that need to override fixtures, but pytest-dev/pytest#421 (comment) lays out a nice solution (see example pseudo-code below for NamedArray/Variable example). I recognize that this type of property testing would be well suited for Hypothesis, but also expect that would slow down progress too much to be an immediate solution.

@andersy005 when do you think would be the best time to test out this method for migrating tests? I could try it out anytime and don't see many downsides to testing while namedarray is still coupled with xarray.

# namedarray/testing.pyimportnumpyasnpclassNamedArrayTestSuite:
@pytest.fixturedeftarget(self):
"""Fixture that needs to be re-declared"""assert0@pytest.fixturedefdata(self):
return0.5*np.arange(10).reshape(2, 5)
deftest_properties(self, target, data):
asserttarget.dims== ("x", "y")
assertnp.array_equal(target.data, data)
asserttarget.attrs== {"key": "value"}
asserttarget.ndim==2asserttarget.sizes== {"x": 2, "y": 5}
asserttarget.size==10asserttarget.nbytes==80assertlen(target) ==2

# namedarray/tests/test_namedarray.pyfromnamedarray.testingimportNamedArrayTestSuitefromnamedarrayimportNamedArrayclassTestNamedArray(NamedArrayTestSuite):
@pytest.fixturedeftarget(self, data):
returnNamedArray(["x", "y"], data, {"key": "value"})

# xarray/tests/test_variable.pyfromnamedarray.testingimportNamedArrayTestSuitefromxarrayimportVariableclassTestVariable(NamedArrayTestSuite):
@pytest.fixturedeftarget(self, data):
data=0.5*np.arange(10).reshape(2, 5)
returnVariable(["x", "y"], data, {"key": "value"})