gh-154194: Degrade frames in Tachyon instead of failing the sample (#154195)

* degrade gracefully

* news

* better NEWS wording

* do not raise on MAX_REMOTE_STR_READ

* bye MAX_REMOTE_STR_READ

* fix -m asyncio ps|pstree

* test truncation and linetable sentinel

* simpler

* simpler

* redundant now

* respect #157790 in the news

---------

Co-authored-by: Pablo Galindo Salgado <Pablogsal@gmail.com>
(cherry picked from commit 7d25916)

Author: maurycy

Include/internal/pycore_global_objects_fini_generated.h

@@ -54,6 +54,9 @@ struct _Py_global_strings {
         STRUCT_FOR_STR(native, "<native>")
         STRUCT_FOR_STR(str_replace_inf, "1e309")
         STRUCT_FOR_STR(type_params, ".type_params")
+        STRUCT_FOR_STR(unknown_file, "<unknown file>")
+        STRUCT_FOR_STR(unknown_function, "<unknown function>")
+        STRUCT_FOR_STR(unreadable_frame, "<unreadable frame>")
         STRUCT_FOR_STR(utf_8, "utf-8")
     } literals;
 

Include/internal/pycore_global_strings.h

@@ -27,6 +27,10 @@ def __init__(
 # ─── indexing helpers ───────────────────────────────────────────
 def _format_stack_entry(elem: str|FrameInfo) -> str:
     if not isinstance(elem, str):
+        if elem.location is None:
+            if elem.filename in ("", "~"):
+                return f"{elem.funcname}"
+            return f"{elem.funcname} {elem.filename}"
         if elem.location.lineno == 0 and elem.filename == "":
             return f"{elem.funcname}"
         else:
@@ -190,8 +194,7 @@ def build_task_table(result):
             # Build coroutine stack string
             frames = [frame for coro in task_info.coroutine_stack
                      for frame in coro.call_stack]
-            coro_stack = " -> ".join(_format_stack_entry(x).split(" ")[0]
-                                   for x in frames)
+            coro_stack = " -> ".join(x.funcname for x in frames)
 
             # Handle tasks with no awaiters
             if not task_info.awaited_by:
@@ -202,8 +205,7 @@ def build_task_table(result):
             # Handle tasks with awaiters
             for coro_info in task_info.awaited_by:
                 parent_id = coro_info.task_name
-                awaiter_frames = [_format_stack_entry(x).split(" ")[0]
-                                for x in coro_info.call_stack]
+                awaiter_frames = [x.funcname for x in coro_info.call_stack]
                 awaiter_chain = " -> ".join(awaiter_frames)
                 awaiter_name = id2name.get(parent_id, "Unknown")
                 parent_id_str = (hex(parent_id) if isinstance(parent_id, int)

Include/internal/pycore_runtime_init_generated.h

@@ -1558,6 +1558,82 @@ def test_table_output_format(self):
 
 class TestAsyncioToolsEdgeCases(unittest.TestCase):
 
+    def test_frames_without_location_tree(self):
+        """Frames the unwinder could not fully read - should not crash."""
+        input_ = [
+            AwaitedInfo(
+                thread_id=1,
+                awaited_by=[
+                    TaskInfo(
+                        task_id=1,
+                        task_name="Task-A",
+                        coroutine_stack=[
+                            CoroInfo(
+                                call_stack=[
+                                    FrameInfo("<unreadable frame>", "~", None),
+                                    FrameInfo("<unknown function>", "app.py", None),
+                                    FrameInfo("big", "big.py", None),
+                                ],
+                                task_name=1
+                            )
+                        ],
+                        awaited_by=[]
+                    )
+                ]
+            )
+        ]
+        self.assertEqual(
+            tools.build_async_tree(input_),
+            [[
+                "└── (T) Task-A",
+                "    └──  big big.py",
+                "        └──  <unknown function> app.py",
+                "            └──  <unreadable frame>",
+            ]],
+        )
+
+    def test_frames_without_location_table(self):
+        """Frame names are not truncated at the first space."""
+        input_ = [
+            AwaitedInfo(
+                thread_id=1,
+                awaited_by=[
+                    TaskInfo(
+                        task_id=1,
+                        task_name="Task-A",
+                        coroutine_stack=[
+                            CoroInfo(
+                                call_stack=[
+                                    FrameInfo("<unreadable frame>", "~", None)
+                                ],
+                                task_name=1
+                            )
+                        ],
+                        awaited_by=[
+                            CoroInfo(
+                                call_stack=[
+                                    FrameInfo("<unknown function>", "app.py", None)
+                                ],
+                                task_name=2
+                            )
+                        ]
+                    )
+                ]
+            )
+        ]
+        self.assertEqual(
+            tools.build_task_table(input_),
+            [[
+                1,
+                "0x1",
+                "Task-A",
+                "<unreadable frame>",
+                "<unknown function>",
+                "Unknown",
+                "0x2",
+            ]],
+        )
+
     def test_task_awaits_self(self):
         """A task directly awaits itself - should raise a cycle."""
         input_ = [

Include/internal/pycore_unicodeobject_generated.h

@@ -4070,6 +4070,93 @@ def test_get_stats_disabled_raises(self):
             client_socket.sendall(b"done")
 
 
+@requires_remote_subprocess_debugging()
+@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",
+)
+class TestMetadataDegradation(RemoteInspectionTestBase):
+    """Tests for graceful degradation of oversized code-object metadata."""
+
+    def test_long_qualname_truncated_not_dropped(self):
+        """A qualname longer than 1024 chars is truncated instead of
+        failing the whole sample."""
+        name = "f" * 1100
+        src = f"def {name}(sample):\n    return sample()\n"
+        ns = {}
+        exec(src, ns)
+
+        trace = ns[name](RemoteUnwinder(os.getpid()).get_stack_trace)
+        frame = self._find_frame_in_trace(
+            trace, lambda f: f.funcname.startswith("fff")
+        )
+        self.assertIsNotNone(frame)
+        self.assertEqual(frame.funcname, "f" * 1024)
+
+    def test_long_filename_truncated(self):
+        """A filename longer than 1024 chars is truncated instead of
+        failing the whole sample."""
+        src = "def g(sample):\n    return sample()\n"
+        ns = {}
+        exec(compile(src, "x" * 1500 + ".py", "exec"), ns)
+
+        trace = ns["g"](RemoteUnwinder(os.getpid()).get_stack_trace)
+        frame = self._find_frame_in_trace(trace, lambda f: f.funcname == "g")
+        self.assertIsNotNone(frame)
+        self.assertEqual(frame.filename, "x" * 1024)
+
+    def test_oversized_linetable_degrades_to_no_location(self):
+        """A linetable over MAX_LINETABLE_SIZE degrades to a frame without
+        location instead of failing the whole sample."""
+        src = (
+            "def big(sample):\n"
+            + "    x = 1\n" * 20_000
+            + "    return sample()\n"
+        )
+        ns = {}
+        exec(compile(src, "big_linetable.py", "exec"), ns)
+        big = ns["big"]
+        self.assertGreater(len(big.__code__.co_linetable), 64 * 1024)
+
+        trace = big(RemoteUnwinder(os.getpid()).get_stack_trace)
+        frame = self._find_frame_in_trace(
+            trace, lambda f: f.funcname == "big"
+        )
+        self.assertIsNone(frame.location)
+        self.assertEqual(frame.filename, "big_linetable.py")
+
+    @unittest.skipIf(
+        sys.platform == "win32",
+        "Process death maps to ProcessLookupError only on POSIX platforms",
+    )
+    def test_dead_process_raises_not_degrades(self):
+        """Death of the target raises ProcessLookupError instead of
+        degrading to synthetic frames."""
+        script_body = """\
+            import time
+            sock.sendall(b"ready")
+            time.sleep(10_000)
+            """
+        with self._target_process(script_body) as (p, client_socket, make_unwinder):
+            _wait_for_signal(client_socket, b"ready")
+            unwinder = make_unwinder()
+            _get_stack_trace_with_retry(unwinder)
+
+            p.kill()
+            p.wait()
+
+            for _ in busy_retry(SHORT_TIMEOUT, error=False):
+                try:
+                    unwinder.get_stack_trace()
+                except ProcessLookupError:
+                    break
+                except RuntimeError:
+                    continue
+            else:
+                self.fail("ProcessLookupError never raised for dead process")
+
+
 @requires_remote_subprocess_debugging()
 class TestFrameChainLimits(RemoteInspectionTestBase):
     """Frame chain walks abort instead of looping/overflowing on deep chains."""

Lib/asyncio/tools.py

@@ -665,6 +665,26 @@ def test_same_line_different_columns(self):
         collector, count = self.roundtrip(samples)
         self.assertEqual(count, 3)
 
+    def test_synthetic_frames_roundtrip(self):
+        """Degraded/sentinel frames (location=None) survive the binary format."""
+        frames = [
+            FrameInfo(("~", None, name, None))
+            for name in (
+                "<GC>",
+                "<native>",
+                "<unknown function>",
+                "<unknown file>",
+                "<unreadable frame>",
+            )
+        ]
+        frames.append(FrameInfo(("app.py", None, "<unknown function>", None)))
+        frames.append(FrameInfo(("<unknown file>", None, "real_func", None)))
+        samples = [[make_interpreter(0, [make_thread(1, frames)])]]
+
+        collector, count = self.roundtrip(samples)
+        self.assertEqual(count, 1)
+        self.assert_samples_equal(samples, collector)
+
 
 class TestBinaryEdgeCases(BinaryFormatTestBase):
     """Tests for edge cases in binary format."""

Lib/test/test_asyncio/test_tools.py

@@ -0,0 +1,3 @@
+Fix the sampling profiler dropping entire samples when a non-fatal read fails;
+frames now keep any readable metadata, and long funcnames and filenames are
+truncated instead. Patch by Maurycy Pawłowski-Wieroński.

Lib/test/test_external_inspection.py

@@ -182,7 +182,7 @@ typedef enum _WIN32_THREADSTATE {
 #define set_exception_cause(unwinder, exc_type, message)                              \
     do {                                                                              \
         assert(PyErr_Occurred() && "function returned -1 without setting exception"); \
-        if (unwinder->debug && !_Py_RemoteDebug_HasPermissionError()) {               \
+        if (unwinder->debug && !_Py_RemoteDebug_IsFatalReadError()) {                 \
             _set_debug_exception_cause(exc_type, message);                            \
         }                                                                             \
     } while (0)