Improve error handling and robustness in turbine model loads

[rafmudaf]

Redesign the turbine model loading routine to handle errors and be more robust

An issue was noted were the !include turbine.yaml method of including a turbine model was failing. This was due to how the Farm class was supporting the various methods of loading a turbine model. Turbine definitions could be loaded in three ways:

1. Pass a string that matches a turbine included in Floris's turbine library
2. Pass a dictionary representation of a turbine definition

• Note that there's an option to use the yaml keyword "!include" which results in the yaml library preprocessing the inputs and loading the specified file directly into main input file. The result is that floris sees the turbine definition as a dict.

3. A string selecting an turbine that exists in an external turbine library specified in turbine_library_path

Currently, if the second option is used, the Farm.turbine_type is a list of dictionaries. However, dictionaries cannot be hashed (i.e. compared) so the subsequent np.unique command in Farm's turbine loading fails. While it may have been possible to find a workaround, I decided to split the work to load the turbine definitions into distinct steps. This pull request does the following to Farm:

• Check turbine_types with an attrs validator to raise an error if it is not a valid length
• Enforce type checking with an attrs field-validator for turbine_type and turbine_definitions
• Creates a turbine definition cache to avoid doing file I/o repeatedly if a definition has already been loaded
• Treats turbine_types as immutable so that exporting Farm preserves the input data as it was given
• Adds a unit test for the turbine definition handling

Bonuses:

• Remove all mention of JSON except from the legacy input reader
• Expand unit tests for FromDictMixin
• Let load_yaml function fail if the data given is invalid
• Consolidate floating and land-based test turbine definitions

Related issue

I don't see an open issue for this.

Impacted areas of the software

Primarily farm.py. Other changes are cosmetic, related to removing JSON, or testing

Test results, if applicable

All tests pass + new tests are added

[paulf81]

This is so impressive @rafmudaf , thank you!! I pulled this version down and confirm it resolves the issue which I was seeing

[misi9170]

@rafmudaf I've put in one suggestion to think about, but it isn't critical to functionality. Everything runs as expected now.

[RHammond2]

This looks good to me, especially the additional commentary on what's going on in each step. I was originally unsure of the change in control flow, but agree that with the expanded commentary it's worthwhile.

I would just add that one keyword argument, and then this is good to go from my perspective.

[rafmudaf]

Thanks for the reviews @RHammond2 @misi9170 @paulf81