@markshannon I do not understand why this patch is causing the behavior that it is, I would have guessed that this is a fairly innocuous change, do you have any idea of what is happening?
With that said, I am not sure this is the correct approach. It seems to me like we should be propagating the i_lineno in propogate_line_numbers, but b_nofallthrough is set, which is preventing that. I don't have a great understanding of this code, so I thought it was probably easier, and perhaps even better anyway, to fill in i_lineno when adding the instruction. But it seems I am doing something wrong.
So, if you could give me some pointers, it would be great!

See #26781 (comment)

See #26781 (comment)

Indeed, I am right, this is an ownership issue with the locals dictionary:

#11 0x0000555555796cc9 in PyFrame_FastToLocalsWithError (f=f@entry=0x7ffff72a4a50) at Objects/frameobject.c:1009
1009                if (PyObject_SetItem(locals, name, value) != 0) {
(gdb) p locals
$1 = (PyObject *) 0x7ffff6a8a930
(gdb) p locals->ob_refcnt
$2 = 0
(gdb)

@markshannon, please take a look at this seems to be a regression either of PEP 626 or the latest changes in the frames/code object. I am not fully sure if you are accounting on this but I honestly don't see how we could have dropped the ownership of the locals dict when PyFrame_FastToLocalsWithError is called from the trace function.

I think the issue might be in the new 3.11 bytecode.

Before falling out to PyFrame_FastToLocalsWithError, we are triggering

After some digging, it seems like we are poping from an empty stack. What we end up popping in the test that triggers the assert happens to be the locals dict, which then gets decref'ed, and blows up afterwards.

--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -2457,6 +2457,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, PyFrameObject *f, int throwflag)
         }

         case TARGET(GEN_START): {
+            assert(!EMPTY());
             PyObject *none = POP();
             Py_DECREF(none);
             if (!Py_IsNone(none)) {

python: Python/ceval.c:2460: _PyEval_EvalFrameDefault: Assertion `!EMPTY()' failed.

I have also managed to trigger another issue.

--- a/Lib/test/test_sys_settrace.py
+++ b/Lib/test/test_sys_settrace.py
@@ -1447,6 +1447,7 @@ def test_jump_backwards_out_of_with_block(output):
     async def test_jump_backwards_out_of_async_with_block(output):
         output.append(1)
         async with asynctracecontext(output, 2):
+            global test
             output.append(3)

     @jump_test(2, 5, [5])

python: Python/ceval.c:2248: _PyEval_EvalFrameDefault: Assertion `EMPTY()' failed.

If you have any tips on how to use lltrace, that'd be very helpful, I haven't managed to figure that out 😅

If you have any tips on how to use lltrace, that'd be very helpful, I haven't managed to figure that out 😅

Yeah, wait for #26822 to land and then just set __ltrace__ = True in the scope of the function you want to check. You need to run in debug mode.