You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
assignee=Noneclosed_at=<Date2019-03-21.13:09:05.227>created_at=<Date2018-05-23.17:41:20.401>labels= ['interpreter-core', '3.8', 'type-bug', '3.7']
title='Fix and improve errors handling in the garbage collector'updated_at=<Date2019-03-21.13:09:05.227>user='https://github.com/serhiy-storchaka'
There are following bugs in the garbage collector.
If the garbage collector fails to add an object with __del__ or referenced by an object with __del__ to gc.garbage (in handle_legacy_finalizers()), it leaks it and other not added objects with __del__ and referenced by them. They become no longer accessible by the garbage collector.
PyGC_Collect() is not documented, but it is a public C API. And it can be called by user with an exception set. PyGC_Collect() then can either crash or just silent the exception. It is safer to safe possible exception and restore it after collecting.
A pointer to released member can be used (compared with NULL) in invoke_gc_callback(). This is an undefined behavior.
PR 7126 adds several assert(!PyErr_Occurred()) for earlier detection of leaked exceptions. And the failure in adding to gc.garbage is no longer fatal. It is better to not add an object to gc.garbage in this iteration than crash when we run garbage collection when have no memory.
See bpo-33712, bpo-33713 and bpo-33714 for three cases in the stdlib where an exception can be set in the tp_clear handler. This caused a crash in the garbage collector, or can be just silenced if failed at the shutdown stage. In the master branch it will cause writing a traceback to the stderr. It would be better to handle exceptions locally in the tp_clear handlers. But perhaps it may be worth to handle leaked exceptions in the garbage collector too. I'm just not sure about a solution. Writing a traceback to the stderr can cause a regression if it was just silenced before. But silencing it can hide bugs.