Encoding not preserved when creating ROI

[jlandmann]

The salem roi method of the xarray accessor objects is great and to my knowledge and in this user-friendliness nowhere else available.
I'm using it to clip DEMs to shapefiles and get great results. The only point I'm wondering about is if it's intentional that the encoding of the dataset/datarray and its variables is not preserved. In code language:

In[4]: data = xr.open_dataset(gdir.get_filepath('dem_ts'), group=demtype)
In[5]: data.height.encoding
Out[5]: 
{'_FillValue': nan,
 'chunksizes': None,
 'complevel': 0,
 'contiguous': True,
 'dtype': dtype('float32'),
 'fletcher32': False,
 'original_shape': (1, 1201, 3501),
 'shuffle': False,
 'source': 'some_path\\dem_ts.nc',
 'zlib': False}
In[6]: data = data.salem.roi(shape=gdir.get_filepath('outlines'))
In[7]: data.height.encoding
Out[7]: 
{}
In[8]: data.encoding
Out[8]: 
{}

The Fill_Value might interfere with the value that salem fills the mask with if other is not passed here, but in case you agree that this is not too much of a problem I guess catching the encoding before and just inserting it after making the mask could be a solution. Do you see any other problem?

[fmaussion]

Thanks for the report! I'm happy if you send a PR to correct this.

Note that you might open an issue on the xarray repo as well to ask for an option to preserve encoding with where

[jlandmann]

Thanks for the answer! You are right, maybe fixing it upstream would even be a better option...I'll do that as soon as I'm home!

[fmaussion]

This is going to take a while on the xarray side, so feel free to add a hot fix to salem too.