Prototype a means to detect non-strict types in Pydantic models at lint time

[DMRobertson]

I don't like not-strict-by-default either. FWIW there is the intention to have that feature At Some Point (pydantic/pydantic#1098).

For enforcement, we could write a linter which parses the AST and complains if a type inheriting from BaseModel has an annotation which involves one of the non-strict types, or involves a constrained type without strict=True. I don't like the idea of writing something home-grown for that, but

• we don't want a repeat of stringy power levels
• maybe it wouldn't be too bad to write? Reminds be a bit of the mypy plugin...

Originally posted by @DMRobertson in #13188 (comment)

[DMRobertson]

Rather than parsing syntax or writing plugins, I think we could do something simpler. I'm imagining a new linting script which does the following.

1. Patch pydantic's constrained types to complain if strict is False. I'd suggest wrapping ConstrainedStr.__get_validators__ with something that asserts cls.strict is Truthy.
2. Patch something to inspect the list of fields at model definition time. If some field has a non-strict type, complain. Note that this involves recursively checking type parameters, e.g. dictionary values, list elements. Might be able to define an __init_subclass__ on BaseModel, or else patch something on ModelMetaclass.
3. Import everything from Synapse, recursively so that we define all the models.