Test runner for end-to-end query tests.
| 52 | |
| 53 | |
| 54 | class EndToEndTestRunner(BaseTestRunner): |
| 55 | """Test runner for end-to-end query tests.""" |
| 56 | |
| 57 | def __init__(self): |
| 58 | """Initialize end-to-end test runner.""" |
| 59 | super().__init__(TestType.END_TO_END) |
| 60 | |
| 61 | def get_config_defaults(self) -> dict[str, Any]: |
| 62 | """Get default values from config files.""" |
| 63 | defaults = { |
| 64 | 'site': 'all', |
| 65 | 'model': 'gpt-4o-mini', |
| 66 | 'generate_mode': 'list', |
| 67 | 'retrieval_backend': CONFIG.preferred_retrieval_endpoint, |
| 68 | 'prev': [] |
| 69 | } |
| 70 | |
| 71 | # Try to get preferred model from LLM config |
| 72 | if hasattr(CONFIG, 'preferred_llm_provider') and CONFIG.preferred_llm_provider: |
| 73 | llm_provider = CONFIG.get_llm_provider() |
| 74 | if llm_provider and llm_provider.models: |
| 75 | # Use the 'low' model as default for testing |
| 76 | defaults['model'] = llm_provider.models.low or defaults['model'] |
| 77 | |
| 78 | return defaults |
| 79 | |
| 80 | def validate_test_case(self, test_case: dict[str, Any]) -> tuple[bool, str | None]: |
| 81 | """Validate end-to-end test case has required fields.""" |
| 82 | required_fields = ['query'] |
| 83 | missing_fields = [field for field in required_fields if field not in test_case or not test_case[field]] |
| 84 | |
| 85 | if missing_fields: |
| 86 | return False, f"Missing required fields: {missing_fields}" |
| 87 | |
| 88 | # Validate prev field format |
| 89 | if 'prev' in test_case and test_case['prev'] and not isinstance(test_case['prev'], list): |
| 90 | # Try to parse string representation |
| 91 | prev_str = test_case['prev'] |
| 92 | if isinstance(prev_str, str): |
| 93 | if prev_str == '' or prev_str.lower() == 'none': |
| 94 | test_case['prev'] = [] |
| 95 | elif ',' in prev_str: |
| 96 | test_case['prev'] = [q.strip() for q in prev_str.split(',')] |
| 97 | else: |
| 98 | test_case['prev'] = [prev_str] |
| 99 | else: |
| 100 | return False, f"Invalid 'prev' format: expected list or string, got {type(prev_str)}" |
| 101 | |
| 102 | return True, None |
| 103 | |
| 104 | async def run_single_test(self, test_case: dict[str, Any]) -> EndToEndTestResult: |
| 105 | """Run a single end-to-end test.""" |
| 106 | start_time = time.time() |
| 107 | |
| 108 | # Get defaults and merge with test case |
| 109 | defaults = self.get_config_defaults() |
| 110 | |
| 111 | # Create test case object |