Ensure encoding["source"] is available for a pathlib.Path object

[ColemanTom]

Closes #5888. At the moment if you pass a Path object, the source will not be encoded in, only if a string is passed in. Given there is already a function to handle this, _normalize_path, I've simplified the logic to just run that over the str/Path to ensure it is always encoded.

• Closes open_[mf]dataset ds.encoding['source'] for pathlib.Path? #5888
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst

I'm not sure if this needs including in the whats-new.rst. I'm happy to add it if desired.

[mathause]

I think it's also possible to pass the whole dataset as bytes & we don't want this to end off in ds.encoding["source"] I hope my suggestion would take care of this.

[ColemanTom]

I think it's also possible to pass the whole dataset as bytes & we don't want this to end off in ds.encoding["source"] I hope my suggestion would take care of this.

Thanks. I had forgotten you could, for example, pass file-like objects as I don't tend to. I've confirmed your suggestion works and have added it to the change.

[mathause]

Let's see if that works now - should auto-merge on green.

Thanks for your contribution and welcome to xarray!

[kwodzicki]

It would be great to further improve this by handling file-like objects. I am running into an issue when trying to read some data from S3 buckets. I can pass in a file-like object and read the data, but there is no .encoding['source'] in the subsequent dataset.

Can simulate this 'issue' using local data:

import xarray as xr
file = '/path/to/file.nc'
fid  = open(file, 'rb')
ds   = xr.open_dataset( fid, engine='h5netcdf' )

This will have no source encoding in the dataset, but it could be grabbed from fid.name.

Could a check be added for file-like objects, and then insert the .path or .name attribute into .encoding['source']?