Dict[str, V] round-trips to a field-named model instead of a mapping (additionalProperties ignored)

[ErenAta16]

Describe the issue as clearly as possible:

Dict[str, V] doesn't survive a JSON-Schema round trip. schema_type_to_python treats any {"type": "object"} as a nested model and ignores additionalProperties, so a mapping field comes back as a generated model instead of Dict[str, V]. The generated class is named from the schema title, which for a field like by_user is "By User" — a name with a space in it, not a valid Python identifier.

That means the reconstructed type accepts a fixed (empty) set of keys rather than arbitrary string keys, which is the opposite of what the original annotation meant.

Steps/code to reproduce the bug:

from typing import Dict
from pydantic import BaseModel
from outlines.types.json_schema_utils import (
    json_schema_dict_to_pydantic,
    json_schema_dict_to_dataclass,
)

class Scores(BaseModel):
    by_user: Dict[str, int] = {}

schema = Scores.model_json_schema()
# {"properties": {"by_user": {"additionalProperties": {"type": "integer"},
#                             "default": {}, "title": "By User", "type": "object"}}, ...}

json_schema_dict_to_pydantic(schema).model_fields["by_user"].annotation
json_schema_dict_to_dataclass(schema)

Expected result:

Optional[Dict[str, int]]

Error message:

No error — it degrades quietly:

converter	result
json_schema_dict_to_pydantic	Optional[outlines.types.json_schema_utils.By User]
json_schema_dict_to_dataclass	<class 'outlines.types.json_schema_utils.By User'>
json_schema_dict_to_typeddict	NotRequired[outlines.types.json_schema_utils.By User]

"By User".isidentifier()   # False

The generated model has no fields, so nothing in the original mapping's value type survives either — int is gone.

List[str] in the same model round-trips correctly to Optional[List[str]], so this is specific to the additionalProperties shape.

Outlines/Python version information:

outlines @ 13fe8dd (main), Python 3.10.12, pydantic 2.12.5

Context for the maintainers:

schema_type_to_python's t == "object" branch goes straight to json_schema_dict_to_{pydantic,typeddict,dataclass} without looking at additionalProperties. Reading it first — mapping {"type": "object", "additionalProperties": S} with no properties to Dict[str, schema_type_to_python(S)] — would cover it.

Filing this separately because #1908, #1949, #1951 and #1963 are all open against schema_type_to_python and none of them touch additionalProperties; I checked each of them against this repro.