Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

6.0 KB
Raw Blame
from decimal import Decimal

from fields.message_fields import ExploreMessageListItem, MessageListItem, RetrieverResource, WebMessageListItem


def _retrieved_video_resource() -> dict[str, object]:
    # Shaped like a real stored retriever_resources entry (core/rag/entities/citation_metadata.py's
    # RetrievalSourceMetadata, the write-side model) for a video citation with a start/end offset.
    return {
        "position": 1,
        "dataset_id": "af57b949-2259-4fe4-b58a-0117fd3a6b92",
        "document_id": "9196b0c1-898e-4e8e-9d78-18cb07896b89",
        "document_name": "training.mp4",
        "retriever_from": "workflow",
        "score": 0.86,
        "content": "transcript excerpt",
        "page": None,
        "title": "training.mp4",
        "files": None,
        "doc_metadata": {"start_seconds": 1036.0, "end_seconds": 1054.6, "document_name": "training.mp4"},
    }


def _base_kwargs() -> dict[str, object]:
    inputs: dict[str, object] = {}
    retriever_resources: list[object] = []
    agent_thoughts: list[object] = []
    message_files: list[object] = []
    extra_contents: list[object] = []
    return {
        "id": "m1",
        "conversation_id": "c1",
        "inputs": inputs,
        "query": "hi",
        "answer": "answer",
        "retriever_resources": retriever_resources,
        "agent_thoughts": agent_thoughts,
        "message_files": message_files,
        "status": "normal",
        "extra_contents": extra_contents,
    }


class TestExploreMessageListItem:
    def test_retriever_resource_preserves_knowledge_fs_citation_identity(self) -> None:
        kwargs = _base_kwargs()
        payload = MessageListItem.model_validate(
            {
                **kwargs,
                "retriever_resources": [
                    {
                        "position": 1,
                        "document_id": "11111111-1111-4111-8111-111111111111",
                        "document_asset_id": "22222222-2222-4222-8222-222222222222",
                        "document_revision": 4,
                        "document_version": 7,
                    }
                ],
            }
        ).model_dump(mode="json")["retriever_resources"][0]

        assert payload["document_id"] == "11111111-1111-4111-8111-111111111111"
        assert payload["document_asset_id"] == "22222222-2222-4222-8222-222222222222"
        assert payload["document_revision"] == 4
        assert payload["document_version"] == 7

    def test_exposes_metadata_for_history_rehydration(self) -> None:
        # The Explore/installed-app surface must surface message_metadata (incl. reasoning)
        # so the chat-with-history client can rehydrate the thinking panel on reload.
        item = ExploreMessageListItem.model_validate(
            {**_base_kwargs(), "metadata": {"reasoning": {"llm": "thinking..."}}}
        )

        payload = item.model_dump(mode="json")

        assert payload["metadata"] == {"reasoning": {"llm": "thinking..."}}

    def test_metadata_defaults_to_none(self) -> None:
        item = ExploreMessageListItem.model_validate(_base_kwargs())
        assert item.model_dump(mode="json")["metadata"] is None

    def test_base_message_list_item_has_no_metadata(self) -> None:
        # Guard the public service-API contract: the base item must not leak metadata.
        payload = MessageListItem.model_validate(_base_kwargs()).model_dump(mode="json")
        assert "metadata" not in payload

    def test_message_list_item_exposes_usage_fields(self) -> None:
        payload = MessageListItem.model_validate(
            {
                **_base_kwargs(),
                "message_tokens": 7,
                "answer_tokens": 11,
                "provider_response_latency": 1.25,
                "total_price": Decimal("0.0001234"),
                "currency": "USD",
            }
        ).model_dump(mode="json")

        assert payload["message_tokens"] == 7
        assert payload["answer_tokens"] == 11
        assert payload["total_tokens"] == 18
        assert payload["provider_response_latency"] == 1.25
        assert payload["total_price"] == "0.0001234"
        assert payload["currency"] == "USD"

    def test_web_message_list_item_exposes_usage_and_metadata(self) -> None:
        payload = WebMessageListItem.model_validate(
            {
                **_base_kwargs(),
                "metadata": {"usage": {"total_tokens": 18}},
                "message_tokens": 7,
                "answer_tokens": 11,
            }
        ).model_dump(mode="json")

        assert payload["metadata"] == {"usage": {"total_tokens": 18}}
        assert payload["message_tokens"] == 7
        assert payload["answer_tokens"] == 11
        assert payload["total_tokens"] == 18


class TestRetrieverResourceDocMetadata:
    def test_doc_metadata_survives_validation_from_a_stored_dict(self) -> None:
        # GET /messages re-validates the raw dicts stored in Message.retriever_resources
        # (JSON-backed, written from RetrievalSourceMetadata) against RetrieverResource.
        # Before this field existed, ResponseModel's extra="ignore" silently dropped
        # doc_metadata here -- the video player on a reloaded conversation had nothing
        # to read start_seconds/end_seconds from, even though the data was never lost
        # in storage.
        resource = RetrieverResource.model_validate(_retrieved_video_resource())

        payload = resource.model_dump(mode="json")

        assert payload["doc_metadata"] == {
            "start_seconds": 1036.0,
            "end_seconds": 1054.6,
            "document_name": "training.mp4",
        }
        assert payload["title"] == "training.mp4"
        assert payload["retriever_from"] == "workflow"

    def test_message_list_item_propagates_doc_metadata_through_retriever_resources(self) -> None:
        payload = MessageListItem.model_validate(
            {**_base_kwargs(), "retriever_resources": [_retrieved_video_resource()]}
        ).model_dump(mode="json")

        assert payload["retriever_resources"][0]["doc_metadata"]["start_seconds"] == 1036.0