gh-120198: Stop the world when setting __class__ on free-threaded build #120672
Conversation
Fidget-Spinner
commented
Jun 18, 2024
•
Fidget-Spinner
commented
- Issue: Read and write of __class__ in two threads causes crash #120198
Objects/typeobject.c
Outdated
| } | ||
| err: |
There was a problem hiding this comment.
To have a more regular try: ... finally: _PyEval_StartTheWorld() pattern, you can add an int res = -1; variable, replace goto err with goto done, and set res to 0 on success (3 lines above).
There was a problem hiding this comment.
I think @vstinner's suggestion is fine. Or you can refactor the parts that should be in a stop-the-world call into it's own function, like we often do for locks.
Another advantage of moving the body to a separate function is that it makes it more clear what data crosses the stop-the-world boundary -- some data loaded before the stop-the-world call may not be valid after it.
Include/internal/pycore_dict.h
Outdated
| @@ -323,6 +323,8 @@ _PyInlineValuesSize(PyTypeObject *tp) | |||
| int | |||
| _PyDict_DetachFromObject(PyDictObject *dict, PyObject *obj); | |||
|
|
|||
| PyDictObject *_PyObject_materialize_managed_dict_lock_held(PyObject *); | |||
There was a problem hiding this comment.
Let's name this like the other functions: _PyObject_MaterializeManagedDict_LockHeld and move the definition up next to _PyObject_MaterializeManagedDict.
Objects/typeobject.c
Outdated
| } | ||
| err: |
There was a problem hiding this comment.
I think @vstinner's suggestion is fine. Or you can refactor the parts that should be in a stop-the-world call into it's own function, like we often do for locks.
Another advantage of moving the body to a separate function is that it makes it more clear what data crosses the stop-the-world boundary -- some data loaded before the stop-the-world call may not be valid after it.
|
Thanks @Fidget-Spinner for the PR 🌮🎉.. I'm working now to backport this PR to: 3.13. |
|
Sorry, @Fidget-Spinner, I could not cleanly backport this to |
