kqueue not reporting EOF under certain circumstances

[VolodymyrKostyrko]

BPO	9591
Files	test.py: select.kqueue test, reads data from stdin, reports kevents, exits on file end (?)

^{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 2012-07-17.10:24:07.560>
created_at = <Date 2010-08-13.18:47:22.897>
labels = ['invalid', 'type-bug', 'library']
title = 'kqueue not reporting EOF under certain circumstances'
updated_at = <Date 2012-07-17.10:24:07.558>
user = 'https://bugs.python.org/VolodymyrKostyrko'

bugs.python.org fields:

activity = <Date 2012-07-17.10:24:07.558>
actor = 'Volodymyr.Kostyrko'
assignee = 'none'
closed = True
closed_date = <Date 2012-07-17.10:24:07.560>
closer = 'Volodymyr.Kostyrko'
components = ['Library (Lib)']
creation = <Date 2010-08-13.18:47:22.897>
creator = 'Volodymyr.Kostyrko'
dependencies = []
files = ['18506']
hgrepos = []
issue_num = 9591
keywords = []
message_count = 4.0
messages = ['113811', '114432', '114479', '165695']
nosy_count = 1.0
nosy_names = ['Volodymyr.Kostyrko']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = None
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue9591'
versions = ['Python 3.1']

[VolodymyrKostyrko]

This one is BSD related. FreeBSD 8.1.

This works:
# cat test.py | ./test.py
-1 684 32768 0 0

This hangs:
# ./test.py < file
-1 684 0 0 0

The difference is that in second case popped kevent lacks any data on EOF.

[terryjreedy]

I presume the first example should have been
# cat file | ./test.py
or seceond should have been
# ./test.py < test.py
so that test.py gets same input on stdin in either case.

For other readers: kqueue and kevent are bsd-specific functions and classes in the select module. In the first output, 32786 (== 1 << 15) is the EOF flag missing in the second output.

I wonder is this is really a python issue. I would expect that kqueue and kevent are thin wrappers for the os calls and objects and that python should be indifferent to the source of stdin data. You could check the C source. I do not know if there are any developers currently familiar with the innards of bsd.

[VolodymyrKostyrko]

Ok, I'll try to ask FreeBSD developers, too bad I'm not that familiar with C to write example other way than test if it's a system problem.

[VolodymyrKostyrko]

I was wrong. The current behavior is correct. The difference between execution is a difference between pipe file object and vnode file object. vnode backed file objects doesn't send any signals and care should be taken to detect file position.