Segfault caused by weird combination of imports and yield from

[3921e517-b8ff-4a95-ba81-e08735eacad3]

BPO	17669
Nosy	@brettcannon, @vsajip, @ncoghlan, @pitrou, @benjaminp, @bitdancer, @ericsnowcurrently, @phmc
Files	segfault.zip: minimum case to reproduce gen_fix.patch

^{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: ```python assignee = 'https://github.com/benjaminp' closed_at = created_at = labels = ['interpreter-core', 'OS-windows', 'type-crash'] title = 'Segfault caused by weird combination of imports and yield from' updated_at = user = 'https://bugs.python.org/fhamand' ``` bugs.python.org fields: ```python activity = actor = 'pconnell' assignee = 'benjamin.peterson' closed = True closed_date = closer = 'python-dev' components = ['Interpreter Core', 'Windows'] creation = creator = 'fhamand' dependencies = [] files = ['29744', '29748'] hgrepos = [] issue_num = 17669 keywords = ['patch'] message_count = 6.0 messages = ['186354', '186355', '186365', '186370', '186530', '193797'] nosy_count = 11.0 nosy_names = ['brett.cannon', 'vinay.sajip', 'ncoghlan', 'pitrou', 'benjamin.peterson', 'Arfrever', 'r.david.murray', 'python-dev', 'eric.snow', 'pconnell', 'fhamand'] pr_nums = [] priority = 'normal' resolution = 'fixed' stage = 'resolved' status = 'closed' superseder = None type = 'crash' url = 'https://bugs.python.org/issue17669' versions = ['Python 3.3', 'Python 3.4'] ```

[3921e517-b8ff-4a95-ba81-e08735eacad3]

I've found a very strange bug in python 3.3

It's taken me around an hour just to narrow it down to a small case where it happens.

I cannot for the life of me figure out the exact cause. It seems to have something to do with "yield from".

I've attached a case which reproduces this. Run python3.3 test.py, it should segfault on linux, "stopped responding" on windows.

Tested with OS: Linux Debian-60-squeeze-32-minimal 2.6.32-5-686-bigmem PY: Python 3.3.1rc1 (default, Mar 30 2013, 21:44:39)

OS: Windows 7 64 bit PY: Python 3.3.0 (v3.3.0:bd8afb90ebf2, Sep 29 2012, 10:55:48) [MSC v.1600 32 bit (Intel)] on win32

NOTES: If you get rid of "import logging" in generators.py, it only crashes if there's no __pycache__

[3921e517-b8ff-4a95-ba81-e08735eacad3]

The file contents so people dont have to download the zip:

generators.py: ---------------------------

def subgen():
    yield

def other_gen(self):
    move = yield from subgen()

game.py: ---------------------------

class Game(object):
    def __init__(self):
        self.gen = self.first_gen()
        next(self.gen)
    def first_gen(self):
        while True:
            from generators import \
                other_gen
            yield from other_gen(self)

test.py: ---------------------------

from game import Game
Game()

[bitdancer]

The crashing version has 'import logging' at the top of generators.py. I did not experience a crash when it was absent even if there was no __pycache__.

It also doesn't crash if the import is moved out of the body of first_gen.

FAULTHANDLER doesn't fair too well on this one, no real surprise :)

Fatal Python error: Segmentation fault

Current thread 0xb75856c0: zsh: segmentation fault PYTHONFAULTHANDLER=true ../python test.py

[benjaminp]

This is the patch. I'll have to think about whether there's a self-contained way to test this.

[1762cc99-3127-4a62-9baf-30c3d0f51ef7]

New changeset 35cb75b9d653 by Benjamin Peterson in branch '3.3': don't run frame if it has no stack (closes bpo-17669) http://hg.python.org/cpython/rev/35cb75b9d653

New changeset 0b2d4089180c by Benjamin Peterson in branch 'default': merge 3.3 (bpo-17669) http://hg.python.org/cpython/rev/0b2d4089180c

[1762cc99-3127-4a62-9baf-30c3d0f51ef7]

New changeset 516303f32bad by Benjamin Peterson in branch '3.3': add a test for issue bpo-17669 (closes bpo-18565) http://hg.python.org/cpython/rev/516303f32bad