Implementation of the PEP 433: Easier suppression of file descriptor inheritance

[vstinner]

BPO	17036
Nosy	@vstinner
Superseder	bpo-18571: Implementation of the PEP 446: non-inheritable file descriptors
Files	bc88690df059.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:

assignee = None
closed_at = <Date 2013-08-28.19:56:42.371>
created_at = <Date 2013-01-25.22:01:25.597>
labels = ['interpreter-core', 'type-feature']
title = 'Implementation of the PEP 433: Easier suppression of file descriptor inheritance'
updated_at = <Date 2013-08-28.19:56:42.370>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2013-08-28.19:56:42.370>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2013-08-28.19:56:42.371>
closer = 'vstinner'
components = ['Interpreter Core']
creation = <Date 2013-01-25.22:01:25.597>
creator = 'vstinner'
dependencies = []
files = ['28887']
hgrepos = ['173']
issue_num = 17036
keywords = ['patch']
message_count = 10.0
messages = ['180630', '180636', '180637', '180638', '180903', '180904', '180907', '180908', '180957', '196411']
nosy_count = 2.0
nosy_names = ['vstinner', 'Arfrever']
pr_nums = []
priority = 'normal'
resolution = 'rejected'
stage = None
status = 'closed'
superseder = '18571'
type = 'enhancement'
url = 'https://bugs.python.org/issue17036'
versions = ['Python 3.4']

[vstinner]

I create this issue to review and track the implementation of the PEP-433: "Easier suppression of file descriptor inheritance"
http://www.python.org/dev/peps/pep-0433/

[vstinner]

Here is a patch of the latest revision.

configure and Python/importlib.h must be regenerated (are not part of the diff).

[vstinner]

I tested 9bdfa1a3ea8c.patch on Fedora 16 (Linux 3.6, glibc 2.14), Debian Etch (Linux 2.6.19), OpenBSD 5.2, OpenIndiana, FreeBSD 8.2, Windows 7.

I tested both cases (works/not supported) for O_CLOEXEC and SOCK_CLOEXEC. For the WSA_FLAG_NO_HANDLE_INHERIT flag, I only tested a Windows version not supporting it (Windows 7 SP1 is required).

[vstinner]

The select module may need more work.

[vstinner]

New patch:

• sys.setdefaultcloexec() takes again an argument, so sys.setdefaultcloexec(False) is allowed
• add cloexec parameter to select.devpoll(), select.kqueue() and select.epoll()
• when a function accepts a file name and a file descriptor: the cloexec parameter is ignored if the argument is a file descriptor (it was already done for open(), but not for socket.socket on Windows)
• revert enhancements using cloexec=True to simplify the patch: will be done in another issue
• fix various bugs in error handling (close newly created file descriptors on error)
• release the GIL when calling the os: os.urandom(), os.pipe(), os.dup(), etc.

[vstinner]

My TODO list is almost empty: the implementation is done. I just see possible enhancement on Windows: socket.socket() and os.dup() can use an atomic flag to set close-on-exec if native functions are used (WSASocket, DuplicateHandle) instead of the POSIX API. But it can be done later.

[vstinner]

"revert enhancements using cloexec=True to simplify the patch: will be done in another issue"

I just created the issue bpo-17070 to track this task.

[vstinner]

My TODO list is almost empty

Oh, I forgot one point: I stil don't know if the close-on-exec flag of
file descriptors of pass_fds argument of subprocess.Popen should be
set. If close-on-exec flag is set globally, it's not convinient to
have to clear the flag manually on each file descriptor.

[vstinner]

I tested bc88690df059.patch on Solaris 9: test_os failed on openpty(cloexec=True). It looks like setting close-on-exec on master_fd must be done after grantpt().

[vstinner]

The PEP-433 was rejected, replaced with the PEP-446.