Merge pull request #292 from Wolfvin/fix/issue-291-python-module-level-calls

fix(parser): extract Python module-level calls (closes #291)

Author: Wolfvin

scripts/parsers/python_parser.py

@@ -149,6 +149,15 @@ def _extract_references_impl(self, content: str, file_path: str) -> Dict[str, Li
 
         source = content.encode('utf-8')
 
+        # Issue #291: synthetic source id for module-top-level calls (calls not
+        # nested inside any function/class body — e.g. `setup_app()` written as
+        # a bare top-level statement). Mirrors the JS/TS convention from #219:
+        # a `<file>:0:<module>` id with no corresponding graph_nodes row, so
+        # `list`/`search` stay free of fake `<module>` entries while ref_count
+        # (computed from the target side) and `trace --direction up` still see
+        # the caller. `graph_model.is_module_level_source_id()` recognises it.
+        module_node_id = f"{file_path}:0:<module>"
+
         MAX_DEPTH = 200
 
         # Issue #116/#163: iterative DFS walk. The previous recursive form
@@ -266,8 +275,13 @@ def _extract_references_impl(self, content: str, file_path: str) -> Dict[str, Li
                 # Skip decorators - they reference functions but aren't calls
                 continue
 
-            elif node.type == 'call' and fn_id:
-                # Function call: name(args) or obj.method(args)
+            elif node.type == 'call':
+                # Function call: name(args) or obj.method(args).
+                # Issue #291: when fn_id is None the call sits at module
+                # top-level (not inside any def/class body) — attribute it to
+                # the synthetic <module> source id so module-level calls still
+                # produce a CALLS edge (fixes false dead-code + rc undercount).
+                source_id = fn_id if fn_id else module_node_id
                 func_node = node.child_by_field_name('function')
                 if func_node:
                     keep_alive.append(func_node)
@@ -285,14 +299,14 @@ def _extract_references_impl(self, content: str, file_path: str) -> Dict[str, Li
                             if method_name not in PYTHON_SKIP_NAMES:
                                 is_self = obj_name == 'self'
                                 edges.append({
-                                    "from": fn_id,
+                                    "from": source_id,
                                     "to_fn": method_name,
                                     "via_self": is_self
                                 })
                     elif func_node.type == 'identifier':
                         if call_name not in PYTHON_SKIP_NAMES:
                             edges.append({
-                                "from": fn_id,
+                                "from": source_id,
                                 "to_fn": call_name
                             })
 

scripts/trace_engine.py

@@ -377,6 +377,7 @@ def _bfs_trace_indexed(
         max_results: Max entries to return (prevents timeout)
     """
     from edge_resolver import get_callers, get_callees
+    from graph_model import is_module_level_source_id
 
     chain = []
     visited: Set[str] = set()
@@ -471,6 +472,24 @@ def _bfs_trace_indexed(
 
             visited.add(neighbor_id)
 
+            # Issue #291/#223: a synthetic module-level caller id
+            # (`<file>:0:<module>`) has no node_by_id row. Without this the
+            # flat path falls through to the "unknown"/resolved=False branch,
+            # disagreeing with the graph path (which emits fn="<module>",
+            # module_level=True). Mirror the graph path exactly so flat==graph.
+            # Module scope is the top of a file's call hierarchy — do NOT
+            # enqueue for further BFS.
+            if is_module_level_source_id(neighbor_id):
+                chain.append({
+                    "depth": depth,
+                    "direction": direction_label,
+                    "node_id": neighbor_id,
+                    "fn": "<module>",
+                    "module_level": True,
+                    "path": f"{path} → {neighbor_id}",
+                })
+                continue
+
             if neighbor_id in node_by_id:
                 n = node_by_id[neighbor_id]
                 chain_entry = {

tests/test_graph_accuracy_golden.py

@@ -153,9 +153,37 @@ def _callers_of(backend: dict, fn: str) -> set:
 """
 
 
+_MODULE_LEVEL_CALL_PY = """\
+# Guards #291: Python calls at module top level (not inside any def/class) must
+# count toward the callee's rc — analogous to #219 for TS/JS. Before #291 the
+# Python parser only emitted edges for calls inside a function body, so a
+# function called ONLY at module level got rc=0 / status=dead (false positive).
+def setup_app():
+    return 1
+
+
+def helper():
+    return 2
+
+
+def caller():
+    return helper()
+
+
+def py_never_called():
+    # Control: genuinely dead — never called anywhere. Must STAY dead.
+    return 99
+
+
+setup_app()   # module-level call -> synthetic <module> caller
+caller()      # module-level call -> synthetic <module> caller
+"""
+
+
 def _build_workspace(tmp_path) -> str:
     ws = tmp_path / "golden_ws"
     (ws / "src").mkdir(parents=True)
+    (ws / "src" / "mod_level.py").write_text(_MODULE_LEVEL_CALL_PY)
     (ws / "src" / "mod_level.ts").write_text(_MODULE_LEVEL_CALL_TS)
     (ws / "src" / "handler.ts").write_text(_ASYNC_HANDLER_TS)
     (ws / "src" / "svc.ts").write_text(_SVC_TS)
@@ -279,6 +307,63 @@ def test_same_file_rust_const_not_dead(self, scanned):
             f"#220 regression: same-file-used Rust const RED flagged dead by the engine: {red_hits[:1]}"
         )
 
+    def test_python_module_level_call_counts_toward_rc(self, backend):
+        """#291: Python `setup_app`/`caller` called only at module top level.
+
+        Both are called via a bare top-level statement (not inside any function
+        body). Before #291 the Python parser emitted no edge for module-level
+        calls, so both had rc=0 / status=dead (false positive). Each must now
+        have rc>=1 with a caller from mod_level.py.
+        """
+        for fn in ("setup_app", "caller"):
+            rc = _rc(backend, fn)
+            assert rc >= 1, f"#291 regression: Python {fn} rc={rc}, expected >=1 (module-level call)"
+            callers = _callers_of(backend, fn)
+            assert any("mod_level.py" in c for c in callers), (
+                f"#291: expected a caller from mod_level.py for {fn}, got {callers}"
+            )
+
+    def test_python_module_level_caller_visible_in_trace_up(self, scanned):
+        """#291/#223: `trace --direction up setup_app` surfaces the <module> caller.
+
+        The module-level caller uses the synthetic `<file>:0:<module>` id (same
+        format as TS/JS), so `graph_model.is_module_level_source_id()` recognises
+        it and trace-up emits a `module_level=True` / `fn="<module>"` entry.
+        """
+        ws, _ = scanned
+        from commands.trace import execute
+
+        class _Args:
+            name = "setup_app"
+            direction = "up"
+            depth = 10
+            domain = "auto"
+            limit = 20
+            offset = 0
+            max_results = 1000
+            use_graph = True
+            deep = False
+            format = "json"
+
+        result = execute(_Args(), ws)
+        up = result.get("chains", {}).get("up", [])
+        module_callers = [c for c in up if c.get("module_level") or c.get("fn") == "<module>"]
+        assert module_callers, (
+            f"#291 regression: module-level caller of setup_app dropped from trace-up. up callers: {up}"
+        )
+
+    def test_python_genuinely_dead_still_detected(self, backend):
+        """#291 control: an unreferenced Python function IS still dead (rc 0).
+
+        Guards against a fix that inflates rc to hide false positives — a
+        genuinely-never-called Python function must retain rc 0.
+        """
+        nodes = _nodes_named(backend, "py_never_called")
+        assert nodes, "control node py_never_called missing"
+        assert nodes[0].get("ref_count", 0) == 0, (
+            f"#291 control: py_never_called should have rc 0, got {nodes[0].get('ref_count')}"
+        )
+
     def test_genuinely_dead_still_detected(self, backend):
         """Control: an unreferenced internal function IS still dead (rc 0)."""
         nodes = _nodes_named(backend, "trulyUnusedInternal")