gh-157605: Distinguish C frames from cleared Python frames

Author: pablogsal

Lib/test/test_external_inspection.py

@@ -1719,6 +1719,42 @@ def test_empty_native_thread_stack(self):
             lock.release()
             _testcapi.join_temporary_c_thread()
 
+    @skip_if_not_supported
+    @unittest.skipIf(
+        sys.platform == "linux" and not PROCESS_VM_READV_SUPPORTED,
+        "Test only runs on Linux with process_vm_readv support",
+    )
+    def test_popping_python_frame_is_not_native(self):
+        script = """\
+def leaf(depth):
+    if depth:
+        leaf(depth - 1)
+
+while True:
+        leaf(300)
+"""
+        with _managed_subprocess([sys.executable, "-c", script]) as process:
+            for _ in busy_retry(SHORT_TIMEOUT):
+                try:
+                    unwinder = RemoteUnwinder(
+                        process.pid, native=True, gc=False, cache_frames=False,
+                    )
+                except RuntimeError:
+                    continue
+                break
+            samples = 0
+            for _ in range(10_000):
+                try:
+                    frames = unwinder.get_stack_trace()[0].threads[0].frame_info
+                except TRANSIENT_ERRORS:
+                    continue
+                names = [frame.funcname for frame in frames]
+                if "leaf" not in names:
+                    continue
+                samples += 1
+                self.assertNotIn(("leaf", "<native>"), zip(names, names[1:]))
+            self.assertGreater(samples, 1000)
+
     @skip_if_not_supported
     @unittest.skipIf(
         sys.platform == "linux" and not PROCESS_VM_READV_SUPPORTED,

Modules/_remote_debugging/frames.c

@@ -163,21 +163,23 @@ find_frame_in_chunks(StackChunkList *chunks, uintptr_t remote_ptr)
  * FRAME PARSING FUNCTIONS
  * ============================================================================ */
 
+enum { FRAME_PARSE_INTERPRETER = 2 };
+
 int
 is_frame_valid(
     RemoteUnwinderObject *unwinder,
     uintptr_t frame_addr,
     uintptr_t code_object_addr
 ) {
     if ((void*)code_object_addr == NULL) {
-        return 0;
+        return 0;  // Frame being cleared
     }
 
     void* frame = (void*)frame_addr;
 
     char owner = GET_MEMBER(char, frame, unwinder->debug_offsets.interpreter_frame.owner);
     if (owner == FRAME_OWNED_BY_INTERPRETER) {
-        return 0;  // C frame or sentinel base frame
+        return FRAME_PARSE_INTERPRETER;  // C frame or sentinel base frame
     }
 
     if (owner != FRAME_OWNED_BY_GENERATOR && owner != FRAME_OWNED_BY_THREAD) {
@@ -313,6 +315,7 @@ process_frame_chain(
     ctx->last_frame_visited = 0;
 
     while ((void*)frame_addr != NULL) {
+        int parse_result = 0;
         PyObject *frame = NULL;
         uintptr_t next_frame_addr = 0;
         uintptr_t stackpointer = 0;
@@ -326,14 +329,15 @@ process_frame_chain(
         assert(frame_count <= MAX_FRAMES);
 
         if (ctx->chunks && ctx->chunks->count > 0) {
-            if (parse_frame_from_chunks(unwinder, &frame, frame_addr, &next_frame_addr, &stackpointer, ctx->chunks) == 0) {
+            parse_result = parse_frame_from_chunks(
+                unwinder, &frame, frame_addr, &next_frame_addr, &stackpointer, ctx->chunks);
+            if (parse_result == 0) {
                 goto parsed_frame;
             }
             PyErr_Clear();
         }
         {
             uintptr_t address_of_code_object = 0;
-            int parse_result;
             if (ctx->prefetch.frame && ctx->prefetch.frame_addr == frame_addr) {
                 parse_result = parse_frame_buffer(
                     unwinder, &frame, ctx->prefetch.frame,
@@ -368,6 +372,7 @@ process_frame_chain(
         // returning into.
         else if (unwinder->native &&
                  frame == NULL &&
+                 parse_result == FRAME_PARSE_INTERPRETER &&
                  next_frame_addr &&
                  PyList_GET_SIZE(ctx->frame_info) > 0 &&
                  !(unwinder->gc && next_frame_addr == ctx->gc_frame))