Make regrtest with --huntrleaks check for fd leaks

[sbt]

BPO	18174
Nosy	@pitrou, @vstinner, @ezio-melotti
PRs	[2.7] bpo-18174: Fix file descriptor leaks in tests #7408 [2.7] bpo-18174: regrtest -R 3:3 now also detects FD leak #7409 [2.7] bpo-18174: Fix fd_count() on FreeBSD #7420 bpo-18174: regrtest -R 3:3 checks for handle leak #7827 bpo-33966, multiprocessing: Pool wait for worker start #7966
Files	fdleak.patch fdleak.patch fdleak-2.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 2018-09-19.22:48:53.733>
created_at = <Date 2013-06-09.19:47:53.846>
labels = ['type-feature', 'tests']
title = 'Make regrtest with --huntrleaks check for fd leaks'
updated_at = <Date 2018-09-19.22:48:53.732>
user = 'https://bugs.python.org/sbt'

bugs.python.org fields:

activity = <Date 2018-09-19.22:48:53.732>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2018-09-19.22:48:53.733>
closer = 'vstinner'
components = ['Tests']
creation = <Date 2013-06-09.19:47:53.846>
creator = 'sbt'
dependencies = []
files = ['30518', '30561', '36147']
hgrepos = []
issue_num = 18174
keywords = ['patch']
message_count = 25.0
messages = ['190871', '190915', '190918', '191031', '223790', '223874', '223881', '223882', '223885', '223887', '224192', '224200', '224201', '224203', '224236', '224239', '224241', '251849', '252169', '252172', '252183', '318720', '318732', '318735', '325815']
nosy_count = 6.0
nosy_names = ['pitrou', 'vstinner', 'ezio.melotti', 'neologix', 'python-dev', 'sbt']
pr_nums = ['7408', '7409', '7420', '7827', '7966']
priority = 'normal'
resolution = 'wont fix'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue18174'
versions = ['Python 2.7', 'Python 3.4', 'Python 3.5']

[sbt]

regrtest already tests for refcount leaks and memory allocation leaks. It can also be made to check for file descriptor leaks (and perhaps also handles on Windows).

Running with the attached patch makes it look like test_openpty, test_shutil, test_subprocess, test_uuid all leak fds on Linux, but I have not investigated:

$ ./python -m test.regrtest -R 3:3 test_openpty test_shutil test_subprocess test_uuid
[1/4] test_openpty
123456
......
test_openpty leaked [2, 2, 2] fds, sum=6
[2/4/1] test_shutil
beginning 6 repetitions
123456
......
test_shutil leaked [4, 4, 4] fds, sum=12
[3/4/2] test_subprocess
beginning 6 repetitions
123456
......
test_subprocess leaked [5, 5, 5] fds, sum=15
[4/4/3] test_uuid
beginning 6 repetitions
123456
......
test_uuid leaked [1, 1, 1] fds, sum=3
4 tests failed:
    test_openpty test_shutil test_subprocess test_uuid

[python-dev]

New changeset a7381fe515e8 by Richard Oudkerk in branch '2.7':
Issue bpo-18174: Fix fd leaks in tests.
http://hg.python.org/cpython/rev/a7381fe515e8

New changeset 46fe1bb0723c by Richard Oudkerk in branch '3.3':
Issue bpo-18174: Fix fd leaks in tests.
http://hg.python.org/cpython/rev/46fe1bb0723c

[sbt]

The test_shutil leak is caused by bpo-17899. The others are fixed by a7381fe515e8 and 46fe1bb0723c.

[sbt]

Updated version which adds checks for handle leaks on Windows.

[BreamoreBoy]

Just flagging this up as testing is rather important.

[neologix]

The patch looks good.
I just think it would be nice to expose _fdcount() in test.support.

[vstinner]

Why not using os.fstat() instead of os.dup() to check if a file descriptor is open or not?

It's strange to create a new file descriptor with os.dup() to count the file descriptors.

[neologix]

Why not using os.fstat() instead of os.dup() to check if a file descriptor is open or not?

I asked myself the same question, but IIRC, fstat() doesn't always
work on Windows (of course).

[vstinner]

I asked myself the same question, but IIRC, fstat() doesn't always
work on Windows (of course).

Can you please elaborate?

[neologix]

STINNER Victor added the comment:

> I asked myself the same question, but IIRC, fstat() doesn't always
work on Windows (of course).

Can you please elaborate?

Not really, since I don't know much about Windows, but that's
something I think I heard.
Richard will be able to give more details.

[sbt]

I can't remember why I did not use fstat() -- probably it did not occur to me.

[vstinner]

Modified patch to use os.fstat(), and try to use /proc/self/fd/ on Linux.

[python-dev]

New changeset 379aad232000 by Victor Stinner in branch '3.4':
Issue bpo-11453, bpo-18174: Fix leak of file descriptor in test_asyncore
http://hg.python.org/cpython/rev/379aad232000

New changeset 0ced2d2325fb by Victor Stinner in branch 'default':
(Merge 3.4) Issue bpo-11453, bpo-18174: Fix leak of file descriptor in test_asyncore
http://hg.python.org/cpython/rev/0ced2d2325fb

[python-dev]

New changeset 746339776f19 by Victor Stinner in branch '3.4':
Issue bpo-18174: Fix leak of file descriptor in test_tempfile
http://hg.python.org/cpython/rev/746339776f19

New changeset 017d701116d5 by Victor Stinner in branch 'default':
(Merge 3.4) Issue bpo-18174: Fix leak of file descriptor in test_tempfile
http://hg.python.org/cpython/rev/017d701116d5