pthread_kill() doesn't work on the main thread on FreeBSD6

[vstinner]

BPO	12392
Nosy	@vstinner
Files	thread_init_freebsd6.patch pthread_kill_main_thread.diff

^{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 = <Date 2011-06-26.20:29:28.401>
created_at = <Date 2011-06-23.22:56:01.208>
labels = []
title = "pthread_kill() doesn't work on the main thread on FreeBSD6"
updated_at = <Date 2011-11-08.17:31:01.153>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2011-11-08.17:31:01.153>
actor = 'flox'
assignee = 'none'
closed = True
closed_date = <Date 2011-06-26.20:29:28.401>
closer = 'vstinner'
components = []
creation = <Date 2011-06-23.22:56:01.208>
creator = 'vstinner'
dependencies = []
files = ['22431', '22445']
hgrepos = []
issue_num = 12392
keywords = ['patch']
message_count = 8.0
messages = ['138875', '138876', '138877', '138970', '138973', '138974', '139596', '139806']
nosy_count = 3.0
nosy_names = ['vstinner', 'neologix', 'python-dev']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue12392'
versions = ['Python 3.3']

[vstinner]

pthread_kill() doesn't work on the main thread on FreeBSD6: sending a signal to the main thread does nothing. It works on the main thread just after the creation of the first thread.

PyThread__init_thread() has 3 implementations in Python/thread_pthread.h:

• call pthread_init()
• create a dummy thread (no-op) and join it
• do nothing

pthread_init() doesn't exist on FreeBSD6. If the dummy thread implementation is used, pthread_kill() works directly on the main thread at startup.

[vstinner]

Attached patch implements the suggested fix.

[vstinner]

FreeBSD 7 is not affected by this issue.

To test this issue, call signal.pthread_kill(threading.get_ident(), signal.SIGINT) in an interpreter: it should raise a KeyboardInterrupt. On FreeBSD6, it does nothing. Or run ./python -m test -v test_signal: test_pendings hangs because the test reads 3 bytes from a pipe whereas the pipe is empty. The write end is non blocking, but the read end is blocking.

It would be possible to fix the test to fail instead of blocking: set the read end as non blocking (setUp() does always set the write end as non blocking).

[neologix]

Attached patch implements the suggested fix.

The patch looks good to me.

It would be possible to fix the test to fail instead of blocking

I think this issue deserves a specific test, since:

• test_pending tests something completely different
• at the time test_pending gets to run, there's a high chance that
  threads have already been created

I've attached a patch spawning a new interpreter to test that.

[python-dev]

New changeset 024827a9db64 by Victor Stinner in branch 'default':
Issue bpo-12392: fix thread initialization on FreeBSD 6
http://hg.python.org/cpython/rev/024827a9db64

[vstinner]

I've attached a patch spawning a new interpreter to test that.

Thanks, I commited your test at the same time of the fix.

[vstinner]

This issue introduced regressions in the faulthandler module: see issue bpo-12469.

[vstinner]

As discussed in issue bpo-12469, the commit 024827a9db64 makes things worse: it changes the behaviour of many functions related to signal handling (e.g. sigwait()) just to be able to use pthread_kill() on the main thread. I reverted the patch: commit 34061f0d35ba.