Change import to have configurable prefixes via set_options

[aulemahal]

Change Summary

Restructures the import of the default attribute names and prefixes in source.py so that they are actually configurable by patching the variables in utils.py.

Related issue number

No issue sorry.

The problem was the source.py imported the variables directly, this copying the python reference to the string. As strings are non-mutable, someone that would want to customize the attribute names has to re-set the variable in the utils module (I think this is called "monkey-patching" ?). But that doesn't update the version in source.py. Here, the solution is to import the whole utils module object in source, so that the strings are queried as members of the module everytime they are needed, not only a import time.

Checklist

• Unit tests for the changes exist
• Tests pass on CI
• Documentation reflects the changes where applicable

[andersy005]

I didn't realize users would need to override these variables. I'd like to suggest a different approach that exposes these variables and (other future configuration) as options that users can easily manipulate. Using donfig would simplify this process and make it easy to modify these configurations. Let me know what you think.

[aulemahal]

Oh! The way they were declared as constants made me think that was the goal. In our case, we found intake_esm_attrs to be kinda long to type. And, as we are subclassing the catalog class, the user has not explicit knowledge that intake_esm is used, which makes the name a bit weird. We are overriding it with "cat" for now.

I agree that this is not super clean. But there are only 3 variables... If you plan to add more user-controllable stuff, then yes an external package could be useful. For now, I feel it is not needed? I'm a dev at xclim and when we encountered the problem, we decided to go with the set_options function/context solution (like xarray), because it needed no extra package.

[andersy005]

If you have time, let's go with the options dict approach used by xarray for now ( I don't envision other user-configurable settings in the future).

[aulemahal]

@andersy005 Last commit pushed a simplified version of xarray's mechanic for options! I added the class to the API page and mentioned it in the docstring of to_dataset_dict.

[andersy005]

This looks great, @aulemaha! Thank you...