Uh oh!
There was an error while loading. Please reload this page.
bpo-36854: Move GC runtime state from _PyRuntimeState to PyInterpreterState. - #13219
bpo-36854: Move GC runtime state from _PyRuntimeState to PyInterpreterState.#13219ericsnowcurrently wants to merge 3 commits into
Conversation
4bdd07f to
1607996Compare
vstinner
left a comment
There was a problem hiding this comment.
I like what I see: -) But I have some comments on the actual implementation.
| // This will unblock any joining threads. | ||
| tstate->on_delete(tstate->on_delete_data); | ||
| tstate->on_delete = NULL; | ||
| tstate->on_delete_data = NULL; |
There was a problem hiding this comment.
This change looks to be unrelated. If it's related, I would prefer to see it as a separated PR to prepare that one.
Same comment for other thread changes:
// All threading module threads are marked as "done" later
// in PyThreadState_Clear().
and
// This will unblock any joining threads.
// We also do this in PyThreadState_Clear(), but do it here to be sure.
| filename, lineno, "_PyObject_GC_TRACK"); | ||
| PyGC_Head *last = (PyGC_Head*)(_PyRuntime.gc.generation0->_gc_prev); | ||
| PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE(); |
There was a problem hiding this comment.
Would it be possible to use a "struct _gc_runtime_state *state" variable instead?
struct _gc_runtime_state *state = &_PyInterpreterState_GET_UNSAFE()->gc;
| /*[clinic end generated code: output=45a427e9dce9155c input=81ac4940ca579707]*/ | ||
| { | ||
| _PyRuntime.gc.enabled = 1; | ||
| PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE(); |
There was a problem hiding this comment.
Would it be possible to use a "struct _gc_runtime_state *state" variable instead?
struct _gc_runtime_state *state = &_PyInterpreterState_GET_UNSAFE()->gc;
If the line is too lone, maybe add a GET_STATE() macro which returns &_PyInterpreterState_GET_UNSAFE()->gc.
Same comment for the whole file. I would prefer to avoid introducing "PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE(); " lines if it's only used to get the GC state.
| { | ||
| while (_PyRuntime.gc.trash_delete_later) { | ||
| PyObject *op = _PyRuntime.gc.trash_delete_later; | ||
| PyInterpreterState *interp = _PyInterpreterState_GET_UNSAFE(); |
There was a problem hiding this comment.
Again, a "state" variable would be more appropriate here.
| } | ||
| *interp_p = interp; | ||
| _PyGC_Initialize(&interp->gc); |
There was a problem hiding this comment.
Hum, maybe _PyGC_Initialize() should take interp argument instead, to be consistent with other _PyGC functions.
bedevere-bot
commented
May 10, 2019
When you're done making the requested changes, leave the comment: |
csabella
commented
May 31, 2019
@ericsnowcurrently, were you wanting to land this for 3.8? |
vstinner
commented
Nov 20, 2019
I wrote a different change, PR #17287, that I just merged. So I close this old PR. Thanks anyway Eric for working on this, my work is partially based on yours ;-) |
ericsnowcurrently
commented
Nov 22, 2019
Thanks for doing it!!! |
vstinner
commented
Nov 22, 2019
You're welcome. |
Note: as part of this PR, I've made sure that Python threads (created by the threading module) are marked as "deleted" earlier in finalization. That marker is how joining threads get unblocked. It was happening in
PyThreadState_Delete()but now it will happen inPyThreadState_Clear(). This is necessary to ensure that the callback gets called before much interpreter/runtime finalization happens. (It could impact daemon threads, but we already can't guarantee behavior for those once finalization starts.)https://bugs.python.org/issue36854