Created on 2013-08-01 18:20 by pitrou, last changed 2013-08-01 18:57 by pitrou. This issue is now closed.
| Files | ||||
|---|---|---|---|---|
| File name | Uploaded | Description | Edit | |
| atexit_leaks.patch | pitrou, 2013-08-01 18:47 | review | ||
| Messages (4) | |||
|---|---|---|---|
| msg194098 - (view) | Author: Antoine Pitrou (pitrou) *  |
Date: 2013-08-01 18:20 | |
atexit uses PEP 3121 module initialization, and by this scheme each subinterpreter gets a separate module state. However, atexit's m_free doesn't actually decref the callbacks: it only frees the callbacks array, thinking the callbacks were already decref'ed when run. This is ok at main interpreter shutdown, but not at subinterpreter shutdown (see issue 18618). |
|||
| msg194100 - (view) | Author: Antoine Pitrou (pitrou) * |
Date: 2013-08-01 18:47 | |
Here is a patch. It also enables GC on the atexit module. |
|||
| msg194101 - (view) | Author: Roundup Robot (python-dev) | Date: 2013-08-01 18:56 | |
New changeset 71b63a32b1e3 by Antoine Pitrou in branch 'default': Issue #18619: Fix atexit leaking callbacks registered from sub-interpreters, and make it GC-aware. http://hg.python.org/cpython/rev/71b63a32b1e3 |
|||
| msg194102 - (view) | Author: Antoine Pitrou (pitrou) * |
Date: 2013-08-01 18:57 | |
The patch was simple enough that I decided to push it as-is :) Don't hesitate to make comments if you want, though. |
|||
| History | |||
|---|---|---|---|
| Date | User | Action | Args |
| 2013-08-01 18:57:05 | pitrou | set | status: open -> closed resolution: fixed messages: + msg194102 stage: needs patch -> resolved |
| 2013-08-01 18:56:20 | python-dev | set | nosy:
+ python-dev messages: + msg194101 |
| 2013-08-01 18:47:23 | pitrou | set | files:
+ atexit_leaks.patch keywords: + patch messages: + msg194100 |
| 2013-08-01 18:20:15 | pitrou | create | |