issues

Summary: I would like to be able to update dataset/dataarray attributes and sync the updates to disk.

Regarding the Zarr backend in Xarray, at the moment once datasets are read from disk all modifications are in memory only. This means that for the particular case of metadata stored as attributes, in order to update an attribute on disk there are two choices:

• Assign the new attribute to the dataset and write the whole dataset again to disk.
• Use Zarr directly to access the dataset attributes and perform an 'attribute sync'.

The former is prohibitive , in particular for large datasets, but in general it is not a good practice that in order to modify one attribute I need to rewrite the whole pixel data.

The second is possible, and certainly I could imagine the need to write a custom function to do this. Is that something that should be available in the xarray library though? Something like ds.sync_attributes() that updates the attributes on disk?

Closes #3517.

I have tried to touch the minimum code as possible while leaving the defaults as they are now. The code works now by redefining to value to save as a list containing the dictionary. However when writing to disk the the dictionary is saved correctly, without the list (I do not understand 100% how it works but it does so someone double check!).

$ cat test.zarr/.zattrs { "foo": { "bar": { "val": 3 } } }

Similar to #2868, but for Zarr. In this case Zarr should support saving nested dictionaries.

import xarray as xr ds = xr.Dataset() ds.attrs['foo'] = {'bar': 1} ds.to_zarr('foo.zarr', mode='w')

Error message:

``` TypeError Traceback (most recent call last) <ipython-input-11-4e7ad4f0a599> in <module> ----> 1 ds.to_zarr('ll.zarr', mode='w')

/srv/conda/envs/notebook/lib/python3.7/site-packages/xarray/core/dataset.py in to_zarr(self, store, mode, synchronizer, group, encoding, compute, consolidated, append_dim) 1614 compute=compute, 1615 consolidated=consolidated, -> 1616 append_dim=append_dim, 1617 ) 1618

/srv/conda/envs/notebook/lib/python3.7/site-packages/xarray/backends/api.py in to_zarr(dataset, store, mode, synchronizer, group, encoding, compute, consolidated, append_dim) 1299 # validate Dataset keys, DataArray names, and attr keys/values 1300 _validate_dataset_names(dataset) -> 1301 _validate_attrs(dataset) 1302 1303 if mode == "a":

/srv/conda/envs/notebook/lib/python3.7/site-packages/xarray/backends/api.py in _validate_attrs(dataset) 209 # Check attrs on the dataset itself 210 for k, v in dataset.attrs.items(): --> 211 check_attr(k, v) 212 213 # Check attrs on each variable within the dataset

/srv/conda/envs/notebook/lib/python3.7/site-packages/xarray/backends/api.py in check_attr(name, value) 204 "a string, an ndarray or a list/tuple of " 205 "numbers/strings for serialization to netCDF " --> 206 "files".format(value) 207 ) 208

TypeError: Invalid value for attr: {'bar': 1} must be a number, a string, an ndarray or a list/tuple of numbers/strings for serialization to netCDF files ```

Additionally the error message mentions netCDF -- even if I am writing a Zarr file.

Workaround is to save as a list, i.e.

ds.attrs['foo'] = [{'bar': 1}]