PEP446 (CLOEXEC by default) violation with fcntl.fcntl(..., fcntl.F_DUPFD)

[socketpair]

BPO	27150
Nosy	@vstinner, @socketpair, @vadmium

^{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 2016-05-29.07:34:25.758>
labels = ['type-bug', 'library', 'docs']
title = 'PEP446 (CLOEXEC by default) violation with fcntl.fcntl(..., fcntl.F_DUPFD)'
updated_at = <Date 2016-05-29.18:26:37.748>
user = 'https://github.com/socketpair'

bugs.python.org fields:

activity = <Date 2016-05-29.18:26:37.748>
actor = 'socketpair'
assignee = 'docs@python'
closed = False
closed_date = None
closer = None
components = ['Documentation', 'Library (Lib)']
creation = <Date 2016-05-29.07:34:25.758>
creator = 'socketpair'
dependencies = []
files = []
hgrepos = []
issue_num = 27150
keywords = []
message_count = 4.0
messages = ['266591', '266592', '266595', '266610']
nosy_count = 4.0
nosy_names = ['vstinner', 'docs@python', 'socketpair', 'martin.panter']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue27150'
versions = ['Python 3.4', 'Python 3.5', 'Python 3.6']

[socketpair]

1. Need fix the code
2. Fix documentation
3. Update PEP-446 description.

There is F_DUPFD_CLOEXEC constant, that must be used instead of F_DUPFD transparently when kernel >= 2.6.24.

[vadmium]

I don’t see any violation or anything needing fixing in the documentation or code. Can you elaborate?

The fcntl() documentation says “The values used for ‘cmd’ [F_DUPFD, etc] are operating system dependent, and are available as constants in the ‘fcntl’ module, using the same names as used in the relevant C header files.” The newest version of Posix specifies both the F_DUPFD and F_DUPFD_CLOEXEC. It would be misleading for Python to use one when the user specified the other.

If you want a best-effort version of F_DUPFD_CLOEXEC for a single-threaded program, why not use os.dup()? I presume this uses F_DUPFD_CLOEXEC where appropriate, and uses some non-atomic fallback if necessary.

[vstinner]

I don't think that it's possible nor worth it to wrap all fcntl() and all
ioctl() calls. I see these functions as very low-level APIs and so I
consider that the caller is responsible of what he/she does.

If you want to change something, explian in fcntl module doc that the PEP-466 is not implemented in this module znd it's a deliberate choice.

I agree with Martin, it's the responsability of the dev to use the right
command depending on the use case.

[socketpair]

OK, it will be consistent, if docs mention that F_DUPFD will not set O_CLOEXEC automatically, mentiont that F_DUPFD_CLOEXEC should be used for that, and also mention PEP-446 somewhere.

[serhiy-storchaka]

F_DUPFD is not explicitly documented, so I do not think that we should do anything here.