Pydantic models with nested non-JSON-friendly field types cannot be serialized

What are you really trying to do?

Use this as a param:

class MyClass(BaseModel):
   field_a: list[UUID]

Describe the bug

Today we basically just call dict() on this which then tries to json.dumps that which it can’t because that is not a JSON-friendly dict. Unfortunately Pydantic does not provide a way to turn a model to a JSON-friendly dict. We cannot directly return the string result of json() in our encoder because we need it for nested purposes (i.e. JSONEncoder.default, not JSONEncoder.encode which is only called at the top level, not nested).

The best solution right now I think is to accept anything that has a callable json attribute and invoke it, then turn it back to Python object form via json.loads. This is the only easy way I am aware but has an obvious performance penalty. Was also mentioned at https://stackoverflow.com/questions/65622045/pydantic-convert-to-jsonable-dict-not-full-json-string.

[cretz]

After team discussion what we’ll do is:

• Add support for UUID and maybe a couple of other types at https://pydantic-docs.helpmanual.io/usage/types/, but not ambiguous ones like datetime (yet, even though it can be argued ISO 8601 is close to universal)
• Do not support json() from Pydantic models so we don’t have to parse right back out to dict for nested use. Rather we will wait until Pydantic supports JSON-able dicts.
• Add a sample with easy to take code to allow all Pydantic models to work via the slower JSON string approach

[cretz]

I am still gathering feedback on whether we accept the perf penalty of using json() instead of dict() (only to parse it back again). If Pydantic had json_dict() this would be no problem.

Until we either choose to support stringified JSON in models or Pydantic adds support for JSON-able dicts, a custom converter can be used.