gh-158364: Don't report other interpreters' threads in sys._current_f…

[Himesh-rupchandani]

…rames()

sys._current_frames() materialized a PyFrameObject for every thread of every interpreter. A frame object created for a thread of another interpreter belongs to that interpreter: it is stored in its _PyInterpreterFrame.frame_obj and deallocated when that interpreter pops the frame, while the calling interpreter holds the reference from the returned dict and drops it at some arbitrary later time. Since PEP 684 gives each interpreter its own obmalloc arenas, the block ends up being freed by a different interpreter than the one that allocated it, which corrupts the heap and typically aborts the process inside free().

sys._current_exceptions() has the same problem: it hands out references to exception objects owned by other interpreters.

Both functions now only report the threads of the calling interpreter. This matches PyUnstable_DumpTracebackThreads() (used by faulthandler), which already only dumps the threads of one interpreter. As a consequence, only the current interpreter's world needs to be stopped.

• Issue: sys._current_frames() allocates other interpreters' frame objects with the calling interpreter's allocator, corrupting the owner's arena #158364

[read-the-docs-community]

Documentation build overview

📚 cpython-previews | 🛠️ Build #34818209 | 📁 Comparing e05c195 against main (3330712)

  🔍 Preview build  

2 files changed

± library/sys.html
± whatsnew/changelog.html

[Himesh-rupchandani]

Thanks for the review! Force-pushed with all of it addressed:

Both versionchanged blocks are gone — Doc/library/sys.rst is untouched now.
NEWS entry shortened to your wording.
Dropped the _Py_EnsureTstateNotNULL() call and both long comments from
_PyThread_CurrentFrames() / _PyThread_CurrentExceptions(). The loop
comments now just say "the current interpreter's thread states" to match the
new loop.
Deleted test_current_frames_subinterpreter_thread.
Both remaining tests now use threading_helper.start_threads(..., unlock=left.set), wait without timeouts, use entered/left events, and
if True: instead of textwrap.dedent.
The PR is now 3 files, +114/-52.

[Himesh-rupchandani]

A thread only shows up in sys._current_exceptions() with something other
than None while it is actively handling an exception, so the thread has to
be parked inside an except block — otherwise the subinterpreter's call has
nothing of ours to hand out and the test doesn't exercise the broken path.
entered/left just keep it there while the subinterpreter samples. I've
added a one-line comment saying so.

[ZeroIntensity]

LGTM, congrats on your first contribution!

[miss-islington-app]

Thanks @Himesh-rupchandani for the PR, and @ZeroIntensity for merging it 🌮🎉.. I'm working now to backport this PR to: 3.13, 3.14, 3.15.
🐍🍒⛏🤖

[bedevere-app]

GH-158471 is a backport of this pull request to the 3.15 branch.

[miss-islington-app]

Sorry, @Himesh-rupchandani and @ZeroIntensity, I could not cleanly backport this to 3.13 due to a conflict.
Please backport using cherry_picker on command line.

cherry_picker c3cb46870bdb18d104557d8b69d451a692c64c8a 3.13

[bedevere-app]

GH-158472 is a backport of this pull request to the 3.14 branch.

[bedevere-app]

GH-158473 is a backport of this pull request to the 3.13 branch.