Allow dicts to have both patternProperties and additionalProperties

[jparise]

Support for patternProperties was introduced in #332, but that logic unfortunately made patternProperties and additionalProperties mutually exclusive. JSON Schema supports their combination:

https://json-schema.org/understanding-json-schema/reference/object.html#additional-properties

This prevented well-typed code generation for dictionaries that use regex-constrained string keys.

[jparise]

This patch is for the v1.10.x branch. The main (v2) patch is #4642.

[jparise]

The failing FastAPI test in this branch appears to be unrelated to this change. It's dying in an SQLAlchemy code path (sqlalchemy/engine/cursor.py:164: ValueError).

[samuelcolvin]

thanks so much.

[matejetz]

i think this introduced inconsistency regarding the json schema vs pydantic validation behavior.

in the following, pydantic rightfully raises a ValidationError

class Model(BaseModel):
    a: Dict[constr(regex="^text$"), str]

Model(a={'foo': 'bar'})  # correctly raises ValidationError

however, the json schema produced by Model.schema() will now validate the following, as 'additionalProperties': { 'type': 'string'} just allows any string properties.

{
  "a": {
    "foo": "bar"
  }
}

i also doubt this is the expected behavior when using a regex-constrained dict key. instead, imho, additionalProperties should be false by default and might possibly be changed by passing some param to Field.

[samuelcolvin]

What version are you using? Might be worth trying on main.

We can't really make a change to this in v1, but we can get it right in V2.

[matejetz]

sorry, i forgot to include that my comment concerns v1, which this PR was also targeting.
why do you think you can't change this, i'd almost go as far and classify this as a bug or at least unwanted/ unexpected behavior.

i did not have a chance to take a look at v2 yet but will definitely do, looks very promising. love the lib! 👍

[ElectronicRU]

Yes, this seems like a wrong change to me - it makes JSON schema more permissive than the actual schema is. Can @samuelcolvin elaborate why that's not changeable in v1?