did it pass ?

Author: mohammedahmed18

codeflash/optimization/function_optimizer.py

@@ -1823,14 +1823,14 @@ def run_optimized_candidate(
                     # if the test unmatched percentage is greater than 50%, we can't fix it
                     return self.get_results_not_matched_error()
 
+                print(f"should try to fix it, diffs: {diffs}")
                 # with the parsed test results diff ask the llm to fix the candidate to match the test results of the original code, and run again
                 # self.run_optimized_candidate(
                 #     optimization_candidate_index=optimization_candidate_index,
                 #     baseline_results=baseline_results,
                 #     original_helper_code=original_helper_code,
                 #     file_path_to_helper_classes=file_path_to_helper_classes,
                 # )
-                print(f"should try to fix it, diffs: {diffs}")
                 return self.get_results_not_matched_error()
 
             logger.info(f"loading|Running performance tests for candidate {optimization_candidate_index}...")

codeflash/verification/equivalence.py

@@ -18,8 +18,8 @@
 class TestDiffScope(Enum):
     RETURN_VALUE = "return_value"
     STDOUT = "stdout"
-    TIMED_OUT = "timed_out"
     DID_PASS = "did_pass"  # noqa: S105
+    TIMED_OUT = "timed_out"
 
 
 @dataclass
@@ -28,6 +28,8 @@ class TestDiff:
     pytest_error: str
     original_value: any
     candidate_value: any
+    original_pass: bool
+    candidate_pass: bool
     test_src_code: Optional[str] = None
 
 
@@ -86,6 +88,8 @@ def compare_test_results(original_results: TestResults, candidate_results: TestR
                     original_value=original_test_result.return_value,
                     candidate_value=cdd_test_result.return_value,
                     pytest_error=cdd_pytest_error,
+                    original_pass=original_test_result.did_pass,
+                    candidate_pass=cdd_test_result.did_pass,
                 )
             )
 
@@ -111,6 +115,8 @@ def compare_test_results(original_results: TestResults, candidate_results: TestR
                     original_value=original_test_result.stdout,
                     candidate_value=cdd_test_result.stdout,
                     pytest_error=cdd_pytest_error,
+                    original_pass=original_test_result.did_pass,
+                    candidate_pass=cdd_test_result.did_pass,
                 )
             )
             break
@@ -128,6 +134,8 @@ def compare_test_results(original_results: TestResults, candidate_results: TestR
                     original_value=original_test_result.did_pass,
                     candidate_value=cdd_test_result.did_pass,
                     pytest_error=cdd_pytest_error,
+                    original_pass=original_test_result.did_pass,
+                    candidate_pass=cdd_test_result.did_pass,
                 )
             )
             break