[BUGS] Resolve reference in JSON schema for OCI Generative AI compatibility

[paxiaatucsdedu]

Problem

with_structured_output(..., method="json_schema") fails when using Pydantic models containing Enums or nested models because OCI Generative AI API doesn't support JSON Schema $ref and $defs keywords.

Error:

ServiceError: Invalid JSON payload received. Unknown name "$defs" at 'generation_config.response_schema'
Solution

See issue #65

Solutions

• Added OCIUtils.resolve_schema_refs() to recursively inline all $ref references and remove $defs before sending schemas to OCI API
• Integrated schema resolution into with_structured_output() for json_schema method
• Added unit test: test_with_structured_output_json_schema_nested_refs() to verify fix with nested models and enums

Testing

For Literal Pydantic data structure:

Input:

class Status(BaseModel):
    status: Literal["pending", "running", "done"]

class Formatted(BaseModel):
    answer: str
    user_request: str
    status: Status


structured_chat = chat.with_structured_output(Formatted, method="json_schema")
structured_response = structured_chat.invoke(
    [HumanMessage("Tell me something about the statue of liberty")]
)

print(structured_response)

Output:
answer='The Statue of Liberty is a colossal neoclassical sculpture on Liberty Island in New York Harbor in New York City. The copper statue, a gift from the people of France to the people of the United States, was designed by French sculptor Frédéric Auguste Bartholdi and built by Gustave Eiffel. It was dedicated on October 28, 1886.' user_request='Tell me something about the statue of liberty' status=Status(status='done')

For Enum Pydantic data structure:

Input:

class Domain(Enum):
    NATURE = "NATURE"
    TECHNICAL = "TECHNICAL"
    HISTORICAL = "HISTORICAL"


class Formatted(BaseModel):
    answer: str
    user_request: str
    domain: Domain


structured_chat = chat.with_structured_output(Formatted, method="json_schema")
structured_response = structured_chat.invoke(
    [HumanMessage("Tell me something about the statue of liberty")]
)

print(structured_response)

Output:
answer='The Statue of Liberty was a gift from the people of France to the United States and was dedicated on October 28, 1886. It is a symbol of freedom and democracy.' user_request='Tell me something about the statue of liberty' domain=<Domain.HISTORICAL: 'HISTORICAL'>

For nested reference Pydantic data structure:

Input:

class Color(Enum):
    RED = "RED"
    BLUE = "BLUE"
    GREEN = "GREEN"

class Item(BaseModel):
    name: str
    color: Color  # ← Creates $ref to Color

class Response(BaseModel):
    message: str
    items: List[Item]  # ← Array with $ref inside

# Test
structured_chat = chat.with_structured_output(Response, method="json_schema")
structured_response = structured_chat.invoke([
    HumanMessage("Tell me about 3 colored objects")
])

print(structured_response)

Output:
message='Here are 3 colored objects.' items=[Item(name='Apple', color=<Color.RED: 'RED'>), Item(name='Sky', color=<Color.BLUE: 'BLUE'>), Item(name='Grass', color=<Color.GREEN: 'GREEN'>)]