Merge pull request #1698 from codeflash-ai/fix-runtime-calculation

fix runtime calculation for java

Author: misrasaurabh1

codeflash-java-runtime/src/main/java/com/codeflash/Comparator.java

@@ -144,8 +144,11 @@ private static Map<String, byte[]> readTestResults(String dbPath) throws Excepti
         try (Connection conn = DriverManager.getConnection(url);
              Statement stmt = conn.createStatement();
              ResultSet rs = stmt.executeQuery(
-                 "SELECT iteration_id, return_value FROM test_results WHERE loop_index = 1")) {
+                 "SELECT test_module_path, test_class_name, test_function_name, iteration_id, return_value FROM test_results WHERE loop_index = 1")) {
             while (rs.next()) {
+                String testModulePath = rs.getString("test_module_path");
+                String testClassName = rs.getString("test_class_name");
+                String testFunctionName = rs.getString("test_function_name");
                 String iterationId = rs.getString("iteration_id");
                 byte[] returnValue = rs.getBytes("return_value");
                 // Strip the CODEFLASH_TEST_ITERATION suffix (e.g. "7_0" -> "7")
@@ -155,7 +158,10 @@ private static Map<String, byte[]> readTestResults(String dbPath) throws Excepti
                 if (lastUnderscore > 0) {
                     iterationId = iterationId.substring(0, lastUnderscore);
                 }
-                results.put(iterationId, returnValue);
+                // Use module:class:function:iteration as key to uniquely identify
+                // each invocation across different test files, classes, and methods
+                String key = testModulePath + ":" + testClassName + ":" + testFunctionName + "::" + iterationId;
+                results.put(key, returnValue);
             }
         }
         return results;

codeflash/languages/java/instrumentation.py

@@ -1073,7 +1073,7 @@ def build_instrumented_body(
             timing_lines = [
                 f"{indent}for (int _cf_i{current_id} = 0; _cf_i{current_id} < _cf_innerIterations{current_id}; _cf_i{current_id}++) {{",
                 f"{inner_indent}int _cf_loopId{current_id} = _cf_outerLoop{current_id} * _cf_maxInnerIterations{current_id} + _cf_i{current_id};",
-                f'{inner_indent}System.out.println("!$######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":" + _cf_i{current_id} + "######$!");',
+                f'{inner_indent}System.out.println("!$######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":" + "{current_id}" + "######$!");',
                 f"{inner_indent}long _cf_end{current_id} = -1;",
                 f"{inner_indent}long _cf_start{current_id} = 0;",
                 f"{inner_indent}try {{",
@@ -1083,7 +1083,7 @@ def build_instrumented_body(
                 f"{inner_indent}}} finally {{",
                 f"{inner_body_indent}long _cf_end{current_id}_finally = System.nanoTime();",
                 f"{inner_body_indent}long _cf_dur{current_id} = (_cf_end{current_id} != -1 ? _cf_end{current_id} : _cf_end{current_id}_finally) - _cf_start{current_id};",
-                f'{inner_body_indent}System.out.println("!######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":" + _cf_i{current_id} + ":" + _cf_dur{current_id} + "######!");',
+                f'{inner_body_indent}System.out.println("!######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":" + "{current_id}" + ":" + _cf_dur{current_id} + "######!");',
                 f"{inner_indent}}}",
                 f"{indent}}}",
             ]
@@ -1140,7 +1140,7 @@ def build_instrumented_body(
             timing_lines = [
                 f"{indent}for (int _cf_i{current_id} = 0; _cf_i{current_id} < _cf_innerIterations{current_id}; _cf_i{current_id}++) {{",
                 f"{inner_indent}int _cf_loopId{current_id} = _cf_outerLoop{current_id} * _cf_maxInnerIterations{current_id} + _cf_i{current_id};",
-                f'{inner_indent}System.out.println("!$######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":{current_id}_" + _cf_i{current_id} + "######$!");',
+                f'{inner_indent}System.out.println("!$######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":{current_id}" + "######$!");',
                 f"{inner_indent}long _cf_end{current_id} = -1;",
                 f"{inner_indent}long _cf_start{current_id} = 0;",
                 f"{inner_indent}try {{",
@@ -1150,7 +1150,7 @@ def build_instrumented_body(
                 f"{inner_indent}}} finally {{",
                 f"{inner_body_indent}long _cf_end{current_id}_finally = System.nanoTime();",
                 f"{inner_body_indent}long _cf_dur{current_id} = (_cf_end{current_id} != -1 ? _cf_end{current_id} : _cf_end{current_id}_finally) - _cf_start{current_id};",
-                f'{inner_body_indent}System.out.println("!######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":{current_id}_" + _cf_i{current_id} + ":" + _cf_dur{current_id} + "######!");',
+                f'{inner_body_indent}System.out.println("!######" + _cf_mod{current_id} + ":" + _cf_cls{current_id} + "." + _cf_test{current_id} + ":" + _cf_fn{current_id} + ":" + _cf_loopId{current_id} + ":{current_id}" + ":" + _cf_dur{current_id} + "######!");',
                 f"{inner_indent}}}",
                 f"{indent}}}",
             ]

codeflash/languages/java/resources/codeflash-runtime-1.0.0.jar

@@ -748,7 +748,6 @@ def detect_unused_helper_functions(
     """
     # Skip this analysis for non-Python languages since we use Python's ast module
     if not is_python():
-        logger.debug("Skipping unused helper function detection for non-Python languages")
         return []
 
     if isinstance(optimized_code, CodeStringsMarkdown) and len(optimized_code.code_strings) > 0:

codeflash/languages/python/context/unused_definition_remover.py

@@ -605,23 +605,16 @@ def parse_sqlite_test_results(sqlite_file_path: Path, test_files: TestFiles, tes
             else:
                 # Try original_file_path first (for existing tests that were instrumented)
                 test_type = test_files.get_test_type_by_original_file_path(test_file_path)
-                logger.debug(f"[PARSE-DEBUG] test_module={test_module_path}, test_file_path={test_file_path}")
-                logger.debug(f"[PARSE-DEBUG]   by_original_file_path: {test_type}")
                 # If not found, try instrumented_behavior_file_path (for generated tests)
                 if test_type is None:
                     test_type = test_files.get_test_type_by_instrumented_file_path(test_file_path)
-                    logger.debug(f"[PARSE-DEBUG]   by_instrumented_file_path: {test_type}")
                 # Default to GENERATED_REGRESSION for Jest/Java tests when test type can't be determined
                 if test_type is None and (is_jest or is_java_test):
                     test_type = TestType.GENERATED_REGRESSION
-                    logger.debug(
-                        f"[PARSE-DEBUG]   defaulting to GENERATED_REGRESSION ({'Jest' if is_jest else 'Java'})"
-                    )
                 elif test_type is None:
                     # Skip results where test type cannot be determined
                     logger.debug(f"Skipping result for {test_function_name}: could not determine test type")
                     continue
-                logger.debug(f"[PARSE-DEBUG]   FINAL test_type={test_type}")
 
             # Deserialize return value
             # For Jest/Java: Store as serialized JSON - comparison happens via language-specific comparator
@@ -699,10 +692,6 @@ def parse_test_xml(
         return test_results
     # Always use tests_project_rootdir since pytest is now the test runner for all frameworks
     base_dir = test_config.tests_project_rootdir
-    logger.debug(f"[PARSE-XML] base_dir for resolution: {base_dir}")
-    logger.debug(
-        f"[PARSE-XML] Registered test files: {[str(tf.instrumented_behavior_file_path) for tf in test_files.test_files]}"
-    )
 
     # For Java: pre-parse fallback stdout once (not per testcase) to avoid O(n²) complexity
     java_fallback_stdout = None
@@ -763,13 +752,9 @@ def parse_test_xml(
             if test_file_name is None:
                 if test_class_path:
                     # TODO : This might not be true if the test is organized under a class
-                    logger.debug(f"[PARSE-XML] Resolving test_class_path={test_class_path} in base_dir={base_dir}")
                     test_file_path = resolve_test_file_from_class_path(test_class_path, base_dir)
 
                     if test_file_path is None:
-                        logger.error(
-                            f"[PARSE-XML] ERROR: Could not resolve test_class_path={test_class_path}, base_dir={base_dir}"
-                        )
                         logger.warning(f"Could not find the test for file name - {test_class_path} ")
                         continue
                 else:
@@ -782,15 +767,11 @@ def parse_test_xml(
                 logger.warning(f"Could not find the test for file name - {test_file_path} ")
                 continue
             # Try to match by instrumented file path first (for generated/instrumented tests)
-            logger.debug(f"[PARSE-XML] Looking up test_type by instrumented_file_path: {test_file_path}")
             test_type = test_files.get_test_type_by_instrumented_file_path(test_file_path)
-            logger.debug(f"[PARSE-XML] Lookup by instrumented path result: {test_type}")
             if test_type is None:
                 # Fallback: try to match by original file path (for existing unit tests that were instrumented)
                 # JUnit XML may reference the original class name, resolving to the original file path
-                logger.debug(f"[PARSE-XML] Looking up test_type by original_file_path: {test_file_path}")
                 test_type = test_files.get_test_type_by_original_file_path(test_file_path)
-                logger.debug(f"[PARSE-XML] Lookup by original path result: {test_type}")
             if test_type is None:
                 # Log registered paths for debugging
                 registered_paths = [str(tf.instrumented_behavior_file_path) for tf in test_files.test_files]