Py_NewInterpreter causes bogus fatal error along the failure path

[hackhaslam]

BPO	45970
Nosy	@@hackhaslam
PRs	bpo-45970: Fix fatal error along the failure path of Py_NewIntperter #29900

^{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

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2021-12-03.05:35:05.839>
labels = ['expert-C-API', 'type-crash', '3.11']
title = 'Py_NewInterpreter causes bogus fatal error along the failure path'
updated_at = <Date 2021-12-03.05:44:54.364>
user = 'https://github.com/hackhaslam'

bugs.python.org fields:

activity = <Date 2021-12-03.05:44:54.364>
actor = 'python-dev'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['C API']
creation = <Date 2021-12-03.05:35:05.839>
creator = 'jason.haslam'
dependencies = []
files = []
hgrepos = []
issue_num = 45970
keywords = ['patch']
message_count = 1.0
messages = ['407566']
nosy_count = 2.0
nosy_names = ['python-dev', 'jason.haslam']
pr_nums = ['29900']
priority = 'normal'
resolution = None
stage = 'patch review'
status = 'open'
superseder = None
type = 'crash'
url = 'https://bugs.python.org/issue45970'
versions = ['Python 3.11']

[hackhaslam]

The failure cleanup of Py_NewInterpreter causes a bogus fatal error on the call to PyThreadState_Delete. The error is about deleting the current thread state. The saved thread state should be swapped back in before deleting the new thread state.

[iritkatriel]

@hackhaslam can you provide a test to reproduce what you are reporting?

[hackhaslam]

I can't remember any more what I did to cause Py_NewInterpreter to fail. It's okay. I gave up on this after I tried several times to sign the CLA without success.

[iritkatriel]

We've moved to within since then and the CLA process has changed. Do you want to try again?

Also, the PR would need a unit test so we need to know how to reproduce the bug.

[iritkatriel]

CC @ericsnowcurrently

[hackhaslam]

I don't remember what I was doing to get into the failure path. I can see a few different cases (lines with goto error in new_interpreter) but I don't know how to trigger any of them. The only obvious thing is not holding the GIL when calling Py_NewInterpreter, but that causes its own more immediate fatal error. As far as I'm concerned, this isn't worth pursuing any farther.

[iritkatriel]

Ok, thanks. I'll close it then.