from unittest.mock import MagicMock, patch

from photools3.llm_session import LlmSession


def test_llm_session_mocked(test_photo):
    # We want to mock OpenAI client's chat.completions.create
    with patch("photools3.llm_session.OpenAI") as mocked_openai:
        # Setup the mock response structure
        mock_client = MagicMock()
        mocked_openai.return_value = mock_client

        # Mock the models list to return a proper object with an ID string
        # This prevents Pydantic ValidationError when LlmSession tries to use model_id
        mock_model = MagicMock()
        mock_model.id = "gpt-4o"
        mock_client.models.list.return_value.data = [mock_model]

        # Create a response object: response.choices[0].message.content
        mock_response = MagicMock()
        mock_choice = MagicMock()
        mock_message = MagicMock()

        mock_message.content = (
            '{"caption": "A beautiful sunset", "tags": ["sunset", "nature"]}'
        )
        mock_choice.message = mock_message
        mock_response.choices = [mock_choice]

        # Configure the client to return our mock response
        mock_client.chat.completions.create.return_value = mock_response

        # Create a new LlmSession instance that will use the mocked OpenAI client
        mocked_session = LlmSession(system_prompt="You are a helper.")

        # Run the test
        result = mocked_session.call(
            prompt="keywords for the attached picture", photos=[test_photo]
        )

        # Verify the results
        assert result["caption"] == "A beautiful sunset"
        assert "sunset" in result["tags"]
        assert "nature" in result["tags"]

        # Verify that the mock was called
        assert mock_client.chat.completions.create.called


def test_llm_session_captions_and_keywords_mocked(test_photo):
    # We want to mock OpenAI client's chat.completions.create
    with patch("photools3.llm_session.OpenAI") as mocked_openai:
        # Setup the mock response structure
        mock_client = MagicMock()
        mocked_openai.return_value = mock_client

        # Mock the models list to return a proper object with an ID string
        # This prevents Pydantic ValidationError when LlmSession tries to use model_id
        mock_model = MagicMock()
        mock_model.id = "gpt-4o"
        mock_client.models.list.return_value.data = [mock_model]

        # Create a response object: response.choices[0].message.content
        mock_response = MagicMock()
        mock_choice = MagicMock()
        mock_message = MagicMock()

        mock_message.content = '{"caption": "A mocked caption", "tags": ["mocked_tag"]}'
        mock_choice.message = mock_message
        mock_response.choices = [mock_choice]

        # Configure the client to return our mock response
        mock_client.chat.completions.create.return_value = mock_response

        # Create a new LlmSessionCaptionsAndKeywords instance that will use the mocked OpenAI client
        from photools3.llm_session import LlmSessionCaptionsAndKeywords

        mocked_session = LlmSessionCaptionsAndKeywords()

        # Run the test
        result = mocked_session.call(photos=[test_photo])

        # Verify the results
        assert result["caption"] == "A mocked caption"
        assert "mocked_tag" in result["tags"]

        # Verify that the mock was called
        assert mock_client.chat.completions.create.called