decode_cf destroys chunks

[rabernat]

Code Sample, a copy-pastable example if possible

import numpy as np
import xarray as xr
xr.DataArray(np.random.rand(1000)).to_dataset(name='random').chunk(100)
ds_cf = xr.decode_cf(ds) 
assert not ds_cf.chunks

Problem description

Calling decode_cf causes variables whose data is dask arrays to be wrapped in two layers of abstractions: DaskIndexingAdapter and LazilyIndexedArray. In the example above

>>> ds.random.variable._data
dask.array<da.random.random_sample, shape=(1000,), dtype=float64, chunksize=(100,)>
>>> ds_cf.random.variable._data
LazilyIndexedArray(array=DaskIndexingAdapter(array=dask.array<da.random.random_sample, shape=(1000,), dtype=float64, chunksize=(100,)>), key=BasicIndexer((slice(None, None, None),))) 

At least part of the problem comes from this line:
https://github.com/pydata/xarray/blob/master/xarray/conventions.py#L1045

This is especially problematic if we want to concatenate several such datasets together with dask. Chunking the decoded dataset creates a nested dask-within-dask array which is sure to cause undesirable behavior down the line

>>> dict(ds_cf.chunk().random.data.dask)
{('xarray-random-bf5298b8790e93c1564b5dca9e04399e',
  0): (<function dask.array.core.getter>, 'xarray-random-bf5298b8790e93c1564b5dca9e04399e', (slice(0, 1000, None),)),
 'xarray-random-bf5298b8790e93c1564b5dca9e04399e': ImplicitToExplicitIndexingAdapter(array=LazilyIndexedArray(array=DaskIndexingAdapter(array=dask.array<da.random.random_sample, shape=(1000,), dtype=float64, chunksize=(100,)>), key=BasicIndexer((slice(None, None, None),))))}

Expected Output

If we call decode_cf on a dataset made of dask arrays, it should preserve the chunks of the original dask arrays. Hopefully this can be addressed by #1752.

Output of xr.show_versions()

commit: 85174cd python: 3.6.2.final.0 python-bits: 64 OS: Darwin OS-release: 16.7.0 machine: x86_64 processor: i386 byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: en_US.UTF-8

xarray: 0.10.0-52-gd8842a6
pandas: 0.20.3
numpy: 1.13.1
scipy: 0.19.1
netCDF4: 1.2.9
h5netcdf: 0.4.1
Nio: None
bottleneck: 1.2.1
cyordereddict: None
dask: 0.16.0
matplotlib: 2.1.0
cartopy: 0.15.1
seaborn: 0.8.1
setuptools: 36.3.0
pip: 9.0.1
conda: None
pytest: 3.2.1
IPython: 6.1.0
sphinx: 1.6.5

[shoyer]

This is basically the same problem as #1372

[rabernat]

Oops! Closing as duplicate.