test_scalar_cube_load

[marqh]

new test to check that a scalar data variable in a netcdf file will be loaded into a cube

addressing long standing bug reports, including #1427

[cpelley]

Thanks @marqh

[marqh]

A zero dimensional scalar variable can never be a netcdf coordinate variable. The definitions within the CF Conventions and the netCDF user guide are explicit on this.
A 'coordinate variable' is one dimensional, with the same name as the dimension it is defining.
A scalar variable is not defined with respect to a dimension, it is a scalar.

[marqh]

I am getting test failures unrelated to my change, e.g.
https://travis-ci.org/SciTools/iris/jobs/292568227#L5125

 File "/home/travis/miniconda/envs/test-environment/lib/python2.7/site-packages/dask/utils.py", line 688, in __call__

return getattr(obj, self.method)(*args, **kwargs)

AttributeError: 'NetCDFDataProxy' object has no attribute 'reshape'

[djkirkham]

Looks like an issue around the new masked array functionality in Dask. dask.array.from_array now has a keyword argument asarray which, when false, prevents Dask converting each chunk to an ndarray. This is intended to allow masked arrays to be passed through. For non-scalar arrays the chunks are the results of calls to __getitem__ on the underlying object, but for scalar arrays there is no __getitem__ call, so in this particular case the NetCDFDataProxy object is being returned by dask at the lowest level, and a reshape() call is being made on it (by Dask), which fails. I've raised this issue on the Dask repo (dask/dask#2823).

[pelson]

@djkirkham: would you mind helping to add a unit test that covers the issue identified with dask?

[djkirkham]

I've added a workaround and unit tests in #2878

[pelson]

@marqh, would you mind rebasing this and adding a what's new? I think the tests should go green what that done, and I'd be happy to merge for the 2.0.0rc1.

[marqh]

would you mind rebasing this and adding a what's new? I think the tests should go green what that done, and I'd be happy to merge for the 2.0.0rc1.

@pelson I have rebased and added a what's new sentence to the 2.0.rst directly

awaiting test results ... ...

[pelson]

Thanks @marqh.