Dataset constructor with DataArray triggers computation

[eric-czech]

Is it intentional that creating a Dataset with a DataArray and dimension names for a single variable causes computation of that variable? In other words, why does xr.Dataset(dict(a=('d0', xr.DataArray(da.random.random(10))))) cause the dask array to compute?

A longer example:

import dask.array as da
import xarray as xr
x = da.random.randint(1, 10, size=(100, 25))
ds = xr.Dataset(dict(a=xr.DataArray(x, dims=('x', 'y'))))
type(ds.a.data)
dask.array.core.Array

# Recreate the dataset with the same array, but also redefine the dimensions
ds2 = xr.Dataset(dict(a=(('x', 'y'), ds.a))
type(ds2.a.data)
numpy.ndarray

[dcherian]

Looks like a bug to me.

[keewis]

I think this is a duplicate of #4483: xarray.core.variable.as_compatible_data doesn't have a special code path for DataArray and thus falls back to:

xarray/xarray/core/variable.py

Line 219 in 333e8db

data = getattr(data, "values", data)

where DataArray.values will cast using numpy.asarray.

You should be able to work around that by using either xr.Dataset(dict(a=(('x', 'y'), ds.a.variable)) or xr.Dataset(dict(a=(('x', 'y'), ds.a.data))

#4493 tries to fix this, but importing DataArray in as_compatible_data to avoid circular imports feels a bit unclean.

[dcherian]

It seems like we should extract DataArray.variable before we get to anything in variable.py (if we can)

[dcherian]

ds2 = xr.Dataset(dict(a=(('x', 'y'), ds.a))

Thinking about this some more, maybe we should raise an error here. This syntax is a tuple of (dims, data, attrs, encoding); but DataArrays already have dims, attrs and encoding, so we'll need to deal with conflicts. I think it's OK to raise an error and ask the user to pass DataArray.data instead.

[keewis]

I think it's OK to raise an error and ask the user to pass DataArray.data instead.

👍

as_variable is where the tuple is converted to a Variable, so that's where the check (or conversion) should go. I guess we should ask to change #4493?