os.set_inheritable() looks to be broken on OpenIndiana, regression of Python 3.6

[vstinner]

BPO	27847
Nosy	@jcea, @vstinner, @ned-deily

^{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 2017-06-28.01:05:54.445>
created_at = <Date 2016-08-24.09:57:23.259>
labels = ['deferred-blocker', '3.7']
title = 'os.set_inheritable() looks to be broken on OpenIndiana, regression of Python 3.6'
updated_at = <Date 2017-06-28.01:05:54.443>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2017-06-28.01:05:54.443>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2017-06-28.01:05:54.445>
closer = 'vstinner'
components = []
creation = <Date 2016-08-24.09:57:23.259>
creator = 'vstinner'
dependencies = []
files = []
hgrepos = []
issue_num = 27847
keywords = []
message_count = 4.0
messages = ['273544', '273545', '282489', '297099']
nosy_count = 3.0
nosy_names = ['jcea', 'vstinner', 'ned.deily']
pr_nums = []
priority = 'deferred blocker'
resolution = 'out of date'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue27847'
versions = ['Python 3.6', 'Python 3.7']

[vstinner]

On Python 3.5, it seems ok:
---
test_get_inheritable_cloexec (test.test_socket.InheritanceTest) ... ok
test_set_inheritable (test.test_socket.InheritanceTest) ... ok
test_set_inheritable_cloexec (test.test_socket.InheritanceTest) ... ok
---
http://buildbot.python.org/all/builders/x86%20OpenIndiana%203.5/builds/1159/steps/test/logs/stdio

But not on on Python 3.6:
======================================================================
FAIL: test_get_set_inheritable (test.test_os.FDInheritanceTests)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/export/home/buildbot/32bits/3.x.cea-indiana-x86/build/Lib/test/test_os.py", line 2723, in test_get_set_inheritable
    self.assertEqual(os.get_inheritable(fd), True)
AssertionError: False != True

======================================================================
FAIL: test_openpty (test.test_os.FDInheritanceTests)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/export/home/buildbot/32bits/3.x.cea-indiana-x86/build/Lib/test/test_os.py", line 2796, in test_openpty
    self.assertEqual(os.get_inheritable(master_fd), False)
AssertionError: True != False

======================================================================
FAIL: test_set_inheritable_cloexec (test.test_os.FDInheritanceTests)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/export/home/buildbot/32bits/3.x.cea-indiana-x86/build/Lib/test/test_os.py", line 2747, in test_set_inheritable_cloexec
    0)
AssertionError: 1 != 0

---

http://buildbot.python.org/all/builders/x86%20OpenIndiana%203.x/builds/11800/steps/test/logs/stdio

[vstinner]

Hum, the main difference between Python 3.5 and 3.6 is this optimization:

changeset: 101039:d268f108ba80
user: Victor Stinner <victor.stinner@gmail.com>
date: Sun Apr 17 16:51:52 2016 +0200
files: Python/fileutils.c
description:
Avoid fcntl() if possible in set_inheritable()

Issue bpo-26770: set_inheritable() avoids calling fcntl() twice if the FD_CLOEXEC
is already set/cleared. This change only impacts platforms using the fcntl()
implementation of set_inheritable() (not Linux nor Windows).

[ned-deily]

Although Victor initiated some discussions on python-dev about support for OpenIndiana and other Solaris-related derivatives, I don't we ever reached a consensus, no one has proposed a solution, and currently the OpenIndiana 3.6 buildbot is not functioning well (I've pinged the owner). I am not willing to hold 3.6.0 for a resolution. Let's come back to this later.

https://mail.python.org/pipermail/python-dev/2016-September/146538.html

[vstinner]

I'm not interested to work on OpenIndiana, and I consider that we should just remove the OpenIndiana buildbot, so I abandon this issue.