Support Ramba distributed arrays.

[DrTodd13]

I would like to see Ramba distributed arrays (https://github.com/Python-for-HPC/ramba) supported in xarray as an alternative to Dask arrays.

[DrTodd13]

@TomNicholas Please add an in-progress label for the associated PR I opened.

[DrTodd13]

@TomNicholas @keewis

import ramba
import numpy as np
import xarray

ra1 = ramba.random.randn(10,20)
xa1 = xarray.DataArray(ra1)
np1 = xa1.data.asarray()
print("np1:", np1)
print("xa1.data:", xa1.data, type(xa1.data), xa1.data.dtype, type(xa1.data.dtype))

xa2 = xa1 + 10.0
xa22 = xa2 * 7.1
xa3 = np.sin(xa22)
xa4 = xa3.transpose()
xa5 = xa4.sum()
print("xa5:", xa5, type(xa5))

Using the __array_function__ API and other functions (and not my ill-conceived PR that added ramba to xarray for direct support) , I was able to run the above example. Originally, there were a couple parts of the NumPy API that ramba didn't previously support and I added those. The most difficult of those was 0d arrays. Those are working at least well enough now for the above example so now like I said in the meeting, I'd like even more complicated examples to explore.

Can anybody suggest or provide one?

[dcherian]

Sorry for the late reply @DrTodd13 .

I think #4972 is the real answer here. It looks like @keewis has it mostly working. It hasn't been merged because test_sparse isn't passing yet.