gh-126366: Make native generators thread safe #126371
Conversation
|
cc @colesbury |
|
I will try to take a look at today :) |
ZeroIntensity
changed the title
list.__init__
|
Closing in favor of #120327 |
|
Wait I closed my PR in favour of yours... 😆 @ZeroIntensity |
|
Ah! I'll reopen this and list you as a co-author if yours has merge conflicts. |
| @@ -1104,8 +1104,10 @@ dummy_func( | |||
| (Py_TYPE(receiver_o) == &PyGen_Type || Py_TYPE(receiver_o) == &PyCoro_Type) && | |||
| ((PyGenObject *)receiver_o)->gi_frame_state < FRAME_EXECUTING) | |||
| { | |||
| _PyInterpreterFrame *gen_frame; | |||
There was a problem hiding this comment.
I have a feeling you'll need to fix the specializations too. But right now they aren't enabled, so no way to test them until Matt re-enables them :(.
There was a problem hiding this comment.
PTAL at GH-127426 and #127426 (comment) now. I think we're good for the specializations.
This reverts commit f8b06ae.
| int exc, int closing) | ||
| { | ||
| PySendResult result; | ||
| Py_BEGIN_CRITICAL_SECTION(gen); |
There was a problem hiding this comment.
If we start to put locks, maybe other functions would deserve that lock as well:
- _PyGen_yf()
- gen_close()
- _gen_throw()
- _gen_getframe()
What about these getters?
- gen_getrunning()
- gen_getsuspended()
In frameobject.c, see frame_is_suspended() (should be safe).
In ceval.c, see clear_gen_frame().
| PyGenObject *gen = (PyGenObject *)receiver_o; | ||
| _PyInterpreterFrame *gen_frame = &gen->gi_iframe; | ||
| Py_BEGIN_CRITICAL_SECTION(gen); |
There was a problem hiding this comment.
What about _SEND_GEN_FRAME, YIELD_VALUE, _FOR_ITER_GEN_FRAME and RETURN_GENERATOR instructions?
There was a problem hiding this comment.
I'm not sure we need to mess with things like YIELD_VALUE. A thread can't yield another thread's generator, can it?
Edit: Well, actually, I'll check. Sending to a generator that's yielding might break things.
Co-authored-by: Victor Stinner <vstinner@python.org>
…nsity/cpython into list-init-thread-safety
…it-thread-safety
Co-authored-by: Ken Jin kenjin@python.org
ChainMapof aCounterin threads #126366