"""Tests for the OpenAI-based extractor with mocked API responses.""" from __future__ import annotations import json from datetime import datetime from unittest.mock import MagicMock, patch import pytest from agent_samochodowy.config import Settings from agent_samochodowy.extraction.openai_extractor import extract_with_openai from agent_samochodowy.models import Post def _post(text: str, group: str = "TIR części zamienne") -> Post: return Post( post_id="test_001", group=group, author_name="Test User", author_profile_url="https://fb.com/test", text=text, timestamp=datetime(2025, 1, 15), post_url="https://fb.com/posts/001", ) def _settings(**overrides) -> Settings: defaults = { "llm_api_key": "sk-test-fake-key", "llm_provider": "openai", "llm_model": "gpt-4o-mini", } defaults.update(overrides) return Settings(**defaults) def _mock_openai_response(data: dict) -> MagicMock: """Build a mock ChatCompletion response with a function call.""" tool_call = MagicMock() tool_call.function.name = "extract_part_query" tool_call.function.arguments = json.dumps(data) message = MagicMock() message.tool_calls = [tool_call] choice = MagicMock() choice.message = message response = MagicMock() response.choices = [choice] return response class TestOpenAIExtractor: @patch("agent_samochodowy.extraction.openai_extractor.openai.OpenAI") def test_extracts_part_request(self, mock_openai_cls): mock_client = MagicMock() mock_openai_cls.return_value = mock_client mock_client.chat.completions.create.return_value = _mock_openai_response({ "is_part_request": True, "brand": "VOLVO", "model": "FH13", "year": None, "category": "pompa wody", "engine_code": None, "oe_numbers": ["20744939"], "reason": "Użytkownik szuka pompy wody do Volvo FH13 z numerem OE", }) result = extract_with_openai( _post("Szukam pompy wody do Volvo FH13, numer 20744939. Pilne!"), _settings(), ) assert result.is_part_request is True assert result.brand == "VOLVO" assert result.model == "FH13" assert result.category == "pompa wody" assert "20744939" in result.oe_numbers assert result.reason @patch("agent_samochodowy.extraction.openai_extractor.openai.OpenAI") def test_rejects_selling_post(self, mock_openai_cls): mock_client = MagicMock() mock_openai_cls.return_value = mock_client mock_client.chat.completions.create.return_value = _mock_openai_response({ "is_part_request": False, "brand": "MAN", "model": "TGA", "year": None, "category": "klocki hamulcowe", "engine_code": None, "oe_numbers": [], "reason": "Post sprzedażowy, nie zapytanie kupujące", }) result = extract_with_openai( _post("Sprzedam klocki hamulcowe MAN TGA, nowe, 200 zł"), _settings(), ) assert result.is_part_request is False @patch("agent_samochodowy.extraction.openai_extractor.openai.OpenAI") def test_handles_engine_code(self, mock_openai_cls): mock_client = MagicMock() mock_openai_cls.return_value = mock_client mock_client.chat.completions.create.return_value = _mock_openai_response({ "is_part_request": True, "brand": "OPEL", "model": "Astra F", "year": None, "category": "alternator", "engine_code": "C18NZ", "oe_numbers": [], "reason": "Zapytanie o alternator z kodem silnika", }) result = extract_with_openai( _post("Kupię alternator do Opla Astry F, silnik C18NZ"), _settings(), ) assert result.is_part_request is True assert result.engine_code == "C18NZ" assert result.brand == "OPEL" @patch("agent_samochodowy.extraction.openai_extractor.openai.OpenAI") def test_oe_normalization(self, mock_openai_cls): """OE numbers should be stripped of spaces/dashes and uppercased.""" mock_client = MagicMock() mock_openai_cls.return_value = mock_client mock_client.chat.completions.create.return_value = _mock_openai_response({ "is_part_request": True, "brand": None, "model": None, "year": None, "category": None, "engine_code": None, "oe_numbers": ["51.06500-6408", "A 000 180 27 01"], "reason": "OE numbers in text", }) result = extract_with_openai( _post("Szukam 51.06500-6408 albo A 000 180 27 01"), _settings(), ) assert "51065006408" in result.oe_numbers assert "A0001802701" in result.oe_numbers @patch("agent_samochodowy.extraction.openai_extractor.openai.OpenAI") def test_no_tool_call_returns_false(self, mock_openai_cls): """When OpenAI returns no function call, treat as not a request.""" mock_client = MagicMock() mock_openai_cls.return_value = mock_client message = MagicMock() message.tool_calls = None choice = MagicMock() choice.message = message response = MagicMock() response.choices = [choice] mock_client.chat.completions.create.return_value = response result = extract_with_openai(_post("Random text"), _settings()) assert result.is_part_request is False @patch("agent_samochodowy.extraction.openai_extractor.openai.OpenAI") def test_passes_correct_model(self, mock_openai_cls): """Verify the model parameter is passed through from settings.""" mock_client = MagicMock() mock_openai_cls.return_value = mock_client mock_client.chat.completions.create.return_value = _mock_openai_response({ "is_part_request": False, "brand": None, "model": None, "year": None, "category": None, "engine_code": None, "oe_numbers": [], "reason": "test", }) extract_with_openai(_post("test"), _settings(llm_model="gpt-4o-mini")) call_kwargs = mock_client.chat.completions.create.call_args.kwargs assert call_kwargs["model"] == "gpt-4o-mini" class TestProviderDispatch: """Test that extraction/__init__.py dispatches correctly.""" @patch("agent_samochodowy.extraction.openai_extractor.extract_with_openai") def test_openai_dispatch(self, mock_extract): from agent_samochodowy.extraction import _extract_with_llm mock_extract.return_value = MagicMock(is_part_request=False) settings = _settings(llm_provider="openai") _extract_with_llm(_post("test"), settings) mock_extract.assert_called_once() @patch("agent_samochodowy.extraction.llm_extractor.extract_with_llm") def test_anthropic_dispatch(self, mock_extract): from agent_samochodowy.extraction import _extract_with_llm mock_extract.return_value = MagicMock(is_part_request=False) settings = _settings(llm_provider="anthropic") _extract_with_llm(_post("test"), settings) mock_extract.assert_called_once()