Merge pull request #1974 from codeflash-ai/fix/false-positive-test-discovery

Fix false positive test discovery from substring matching

Author: Saga4

codeflash/languages/javascript/support.py

@@ -7,6 +7,7 @@
 from __future__ import annotations
 
 import logging
+import re
 import subprocess
 import xml.etree.ElementTree as ET
 from pathlib import Path
@@ -230,11 +231,14 @@ def discover_tests(
         """
         result: dict[str, list[TestInfo]] = {}
 
-        # Build index: function_name → qualified_name for O(1) lookup
-        # This avoids iterating all functions for every test file (was O(NxM), now O(N+M))
+        # Build indices for O(1) lookup per imported name (avoids O(NxM) loop)
         function_name_to_qualified: dict[str, str] = {}
+        class_name_to_qualified_names: dict[str, list[str]] = {}
         for func in source_functions:
             function_name_to_qualified[func.function_name] = func.qualified_name
+            for parent in func.parents:
+                if parent.type == "ClassDef":
+                    class_name_to_qualified_names.setdefault(parent.name, []).append(func.qualified_name)
 
         # Find all test files using language-specific patterns
         test_patterns = self._get_test_patterns()
@@ -249,28 +253,41 @@ def discover_tests(
                 analyzer = get_analyzer_for_file(test_file)
                 imports = analyzer.find_imports(source)
 
-                # Build a set of imported function names
+                # Build a set of imported names, resolving aliases and namespace member access
                 imported_names: set[str] = set()
                 for imp in imports:
                     if imp.default_import:
                         imported_names.add(imp.default_import)
+                        # Extract member access patterns: e.g. `math.calculate(...)` → "calculate"
+                        for m in re.finditer(rf"\b{re.escape(imp.default_import)}\.(\w+)", source):
+                            imported_names.add(m.group(1))
+                    if imp.namespace_import:
+                        imported_names.add(imp.namespace_import)
+                        for m in re.finditer(rf"\b{re.escape(imp.namespace_import)}\.(\w+)", source):
+                            imported_names.add(m.group(1))
                     for name, alias in imp.named_imports:
-                        imported_names.add(alias or name)
+                        imported_names.add(name)
+                        if alias:
+                            imported_names.add(alias)
 
                 # Find test functions (describe/it/test blocks)
                 test_functions = self._find_jest_tests(source, analyzer)
 
-                # Match source functions to tests using the index
-                # Only check functions that are actually imported in this test file
+                # Match via indices: function names and class names → qualified names
+                matched_qualified_names: set[str] = set()
                 for imported_name in imported_names:
                     if imported_name in function_name_to_qualified:
-                        qualified_name = function_name_to_qualified[imported_name]
-                        if qualified_name not in result:
-                            result[qualified_name] = []
-                        for test_name in test_functions:
-                            result[qualified_name].append(
-                                TestInfo(test_name=test_name, test_file=test_file, test_class=None)
-                            )
+                        matched_qualified_names.add(function_name_to_qualified[imported_name])
+                    if imported_name in class_name_to_qualified_names:
+                        matched_qualified_names.update(class_name_to_qualified_names[imported_name])
+
+                for qualified_name in matched_qualified_names:
+                    if qualified_name not in result:
+                        result[qualified_name] = []
+                    for test_name in test_functions:
+                        result[qualified_name].append(
+                            TestInfo(test_name=test_name, test_file=test_file, test_class=None)
+                        )
             except Exception as e:
                 logger.debug("Failed to analyze test file %s: %s", test_file, e)
 

tests/languages/javascript/test_false_positive_discovery.py

@@ -0,0 +1,109 @@
+"""Test for false positive test discovery bug (Bug #4)."""
+
+from pathlib import Path
+from tempfile import TemporaryDirectory
+
+import pytest
+
+from codeflash.discovery.functions_to_optimize import FunctionToOptimize
+from codeflash.languages.javascript.support import TypeScriptSupport
+from codeflash.models.models import CodePosition
+
+
+def test_discover_tests_should_not_match_mocked_functions():
+    """Test that functions mentioned only in mocks are not matched as test targets.
+
+    Regression test for Bug #4: False positive test discovery due to substring matching.
+
+    When a test file mocks a function (e.g., vi.mock("./restart-request.js", () => ({...}))),
+    that function should NOT be considered as tested by that file, since it's only mocked,
+    not actually called or tested.
+    """
+    support = TypeScriptSupport()
+
+    with TemporaryDirectory() as tmpdir:
+        test_root = Path(tmpdir)
+
+        # Create a test file that MOCKS parseRestartRequestParams but doesn't test it
+        test_file = test_root / "update.test.ts"
+        test_file.write_text(
+            '''
+import { updateSomething } from "./update.js";
+
+vi.mock("./restart-request.js", () => ({
+  parseRestartRequestParams: (params: any) => ({ sessionKey: undefined }),
+}));
+
+describe("updateSomething", () => {
+  it("should update successfully", () => {
+    const result = updateSomething();
+    expect(result).toBe(true);
+  });
+});
+'''
+        )
+
+        # Source function that is only mocked, not tested
+        source_function = FunctionToOptimize(
+            qualified_name="parseRestartRequestParams",
+            function_name="parseRestartRequestParams",
+            file_path=test_root / "restart-request.ts",
+            starting_line=1,
+            ending_line=10,
+            function_signature="",
+            code_position=CodePosition(line_no=1, col_no=0),
+            file_path_relative_to_project_root="restart-request.ts",
+        )
+
+        # Discover tests
+        result = support.discover_tests(test_root, [source_function])
+
+        # The bug: discovers update.test.ts as a test for parseRestartRequestParams
+        # because "parseRestartRequestParams" appears as a substring in the mock
+        # Expected: should NOT match (empty result)
+        assert (
+            source_function.qualified_name not in result or len(result[source_function.qualified_name]) == 0
+        ), f"Should not match mocked function, but found: {result.get(source_function.qualified_name, [])}"
+
+
+def test_discover_tests_should_match_actually_imported_functions():
+    """Test that functions actually imported and tested ARE correctly matched.
+
+    This is the positive case to ensure we don't break legitimate test discovery.
+    """
+    support = TypeScriptSupport()
+
+    with TemporaryDirectory() as tmpdir:
+        test_root = Path(tmpdir)
+
+        # Create a test file that ACTUALLY imports and tests the function
+        test_file = test_root / "restart-request.test.ts"
+        test_file.write_text(
+            '''
+import { parseRestartRequestParams } from "./restart-request.js";
+
+describe("parseRestartRequestParams", () => {
+  it("should parse valid params", () => {
+    const result = parseRestartRequestParams({ sessionKey: "abc" });
+    expect(result.sessionKey).toBe("abc");
+  });
+});
+'''
+        )
+
+        source_function = FunctionToOptimize(
+            qualified_name="parseRestartRequestParams",
+            function_name="parseRestartRequestParams",
+            file_path=test_root / "restart-request.ts",
+            starting_line=1,
+            ending_line=10,
+            function_signature="",
+            code_position=CodePosition(line_no=1, col_no=0),
+            file_path_relative_to_project_root="restart-request.ts",
+        )
+
+        result = support.discover_tests(test_root, [source_function])
+
+        # Should match: function is imported and tested
+        assert source_function.qualified_name in result, f"Should match imported function, but got: {result}"
+        assert len(result[source_function.qualified_name]) > 0, "Should find at least one test"