issue_comments

In order to maintain a list of currently relevant issues, we mark issues as stale after a period of inactivity

If this issue remains relevant, please comment here or remove the stale label; otherwise it will be marked as closed automatically

@mobigroup copy(deep=True) makes an independent copy but again (as of #1024) does not load data into memory. You need to use .compute() or .load() to load the dask arrays into memory.

And in-memory dataarray couldn't be modified too: import xarray as xr ds = xr.open_mfdataset('./nctiles_monthly/TFLUX/TFLUX*.nc',concat_dim='tile') TFLUX = ds.TFLUX.copy(deep=True) TFLUX.values[1,1,1] = 0*TFLUX.values[1,1,1] TFLUX.values[1,1,1] array([ 60.46262741, 57.78710175, 56.33338547, 56.44807434, ..., 76.84630585, 74.65901184], dtype=float32) But why?

Is it possible to fix the "cache" option for open_mfdataset()? And raise error in other case.

OK, I can reproduce this. The problem is assigning to specific elements of .values when a DataArray's data is not stored as a NumPy array, e.g., if it is a dask array or one of xarray's lazy array classes (used internally for lazy access when opening netCDF files). In practice, this arises with open_mfdataset() or open_dataset(..., cache=False).

It's not entirely obvious how to fix this. Probably the simplest fix is to make .values a readonly numpy array in these cases, so that an error is raised in these cases. This would usually works, but could lead to issues with passing arrays to scikit-learn or pandas due to a Cython bug with read-only memoryview buffers (https://github.com/cython/cython/issues/1605).

Hm, I tested with "cache=True" but it still doesn't work.

I meant open_dataset(cache=False) (it arises with either value for cache for open_mfdataset).

So old code is broken now.

Yes, this is an unfortunate/unintended consequence of https://github.com/pydata/xarray/pull/1024 ("Disable automatic cache with dask").

In practice, this arises with open_mfdataset() or open_dataset() with cache=False.

Hm, I tested with "cache=True" but it still doesn't work.

pip2 install xarray==0.8.2

``` import xarray as xr ds = xr.open_mfdataset('./nctiles_monthly/TFLUX/TFLUX.nc',concat_dim='tile') TFLUX = ds.TFLUX TFLUX.values[1,1,1] = 0TFLUX.values[1,1,1] TFLUX.values[1,1,1]

array([ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., ..., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32) ```

So old code is broken now.

It's works correct for xarray 0.8.2...

I use this dataset: ftp://mit.ecco-group.org/ecco_for_las/version_4/release2/nctiles_monthly/TFLUX My code: ``` import xarray as xr ds = xr.open_mfdataset('./nctiles_monthly/TFLUX/TFLUX.nc',concat_dim='tile') TFLUX = ds.TFLUX TFLUX.values[1,1,1] = 0TFLUX.values[1,1,1] TFLUX.values[1,1,1]

array([ 60.46262741, 57.78710175, 56.33338547, 56.44807434, ..., 76.84630585, 74.65901184], dtype=float32)

Also xarray.show_versions()

INSTALLED VERSIONS

commit: None python: 2.7.14.final.0 python-bits: 64 OS: Darwin OS-release: 17.3.0 machine: x86_64 processor: i386 byteorder: little LC_ALL: en_US.UTF-8 LANG: en_US.UTF-8 LOCALE: None.None

xarray: 0.10.0 pandas: 0.22.0 numpy: 1.13.3 scipy: 0.18.1 netCDF4: 1.2.6 h5netcdf: None Nio: None bottleneck: None cyordereddict: None dask: 0.16.0 matplotlib: 2.1.0 cartopy: None seaborn: None setuptools: 36.5.0 pip: 9.0.1 conda: None pytest: 3.2.3 IPython: 5.1.0 sphinx: None ```

Can you share full information for how you created this dataset? I cannot reproduce this: ``` In [36]: test = xr.DataArray(np.ones(10), ...: coords={ ...: 'x': range(10) ...: }, ...: dims=['x'] ...: ).chunk() ...:

In [37]: test Out[37]: <xarray.DataArray (x: 10)> dask.array<shape=(10,), dtype=float64, chunksize=(10,)> Coordinates: * x (x) int64 0 1 2 3 4 5 6 7 8 9

In [38]: test.values = 0 * test.values

In [39]: test.values Out[39]: array([ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]) ```