POC goals

1. Model geometry.
2. Model an arbitrary custom constraint. (Geometry type)
3. Model feature type.
4. Override feature type JSON Schema to make it GeoJSON.
5. Generate a typed ID reference.
6. Model common names with language tags as dictionary keys.
7. Model sources with JSON Pointer value.
8. Annotate int type with a "hard range" annotation for Parquet.
9. Figure out how this would propagate through Spark into Parquet file.
10. Generate a Spark schema.

Some observations of base types needed:

• A non-empty string type that has no leading or trailing whitespace.
• A floating-point number [0,1] representing a percentage, used for both confidence and linear referencing.
• A UniqueItems constraint may not be needed because in Pydantic you can just declare the field type to be set/Set and this will generate the right uniqeItems constraint in JSON Schema.

CODE GENERATION:

• Python has a native AST manipulation module, ast. This would be good for parsing the schema source code into a Python AST.
• Either the ast module by itself or that plus astor could be used to traverse the schema source code as a syntax tree and generate other code from it.
• It might be useful to use LibCST to generate the code because you can make a "concrete syntax tree" that contains comments etc.
• It appears that the Pydantic artifact that we REALLY care about in terms of an AST-like structure is the core schema. So we probably don't want to parse the code, we just want to find a way to:
  • Discover all the things that derive from BaseModel (or Feature?), either by parsing the code or by requiring some kind of schema manifest. But note that even with a manifest we still have to parse the code.
  • From all the BaseModel and/or Feature derived classes, get their core schema from __get_pydantic_core_schema__.