[issue15748] Various symlink test failures in test_shutil on FreeBSD

Change by Mark Lawrence : -- nosy: -BreamoreBoy ___ Python tracker ___ ___ Python-bugs-list mailing list Unsubscribe:

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Mark Lawrence added the comment: Is any action needed here to take this forward, can it be closed as out of date or what? -- nosy: +BreamoreBoy versions: +Python 3.5 -Python 3.2, Python 3.3 ___ Python tracker rep...@bugs.python.org

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Link to discussion (initiated by Andriy) on z...@lists.illumos.org: http://thread.gmane.org/gmane.os.illumos.zfs/69 -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue15748

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Not really... the problem is that the chmod 0007 actually corrupts the ZFS metadata regarding the symlink by always setting the length to 24. Andriy did some debugging here: http://lists.freebsd.org/pipermail/freebsd-fs/2012-August/015005.html So, the link

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Turns out our unit test uncovered a very subtle corner-case bug in ZFS, requiring the following patch to FreeBSD: a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sa.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sa.c index 69374fb..7f61517 100644 ---

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Larry Hastings added the comment: Could we work around it? Ignore the length we get back from readlink() and just measure the buffer ourselves? Or, if we wanted to be *really* paranoid, write a \0 at the length they give us back and *then* strlen it. --

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Link to my freebsd-fs e-mail: http://lists.freebsd.org/pipermail/freebsd-fs/2012-August/014964.html -- ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue15748

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Reproduced behaviour with test_readlink.sh. Sending an e-mail to freebsd-fs to see what they think. -- Added file: http://bugs.python.org/file26979/test_readlink.sh ___ Python tracker rep...@bugs.python.org

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Hm... tested against other types of file systems and didn't see any issues. Tested manually against ZFS again: % ~/src/cpython/python /tmp/test_readlink.py temp dir: /mnt/t-zfs/trent/d_6i51 path 1: /mnt/t-zfs/trent/d_6i51/qux, length 1: 27, buf 1:

[issue15748] Various symlink test failures in test_shutil on FreeBSD

STINNER Victor added the comment: I don't understand if the issue comes from symlink() or readlink(). Can you reproduce the issue using commands on the command line? $ ln -s setup.py link `link' - `setup.py' $ readlink link setup.py (You may need to test different length for filenames.) Do

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Larry Hastings added the comment: haypo: If a series of shell commands duplicated the problem, then okay, but I suspect the problem has to do with persistent state in the process and will only be reproducible with a C program. -- ___ Python tracker

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Looks like os.readlink() is busted: /home/trent/src/cpython/Lib/shutil.py(107)copyfile() - os.symlink(os.readlink(src), dst) (Pdb) s TypeError: embedded NUL character /home/trent/src/cpython/Lib/shutil.py(107)copyfile() - os.symlink(os.readlink(src), dst)

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Changes by Georg Brandl ge...@python.org: -- nosy: +larry ___ Python tracker rep...@bugs.python.org http://bugs.python.org/issue15748 ___ ___ Python-bugs-list mailing

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Larry Hastings added the comment: I'll ask the obvious: was that an optimized build? Not that that explains the bug--but that might explain why length appeared to contain 24. It's hard to believe this is really an OS bug... -- ___ Python tracker

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Hi Larry, Funnily enough, I just tried to step through a heavily hacked version of posix_readlink again, only to get gdb telling me I couldn't 'print' the variables I had added in. Turns out, `./configure --with-pydebug` on FreeBSD ends up with '-O2' being

[issue15748] Various symlink test failures in test_shutil on FreeBSD

Trent Nelson added the comment: Well, bugger me, check this out: import os import stat import tempfile d = tempfile.mkdtemp() src = os.path.join(d, 'foo') dst = os.path.join(d, 'bar') src_link = os.path.join(d, 'baz') dst_link = os.path.join(d, 'qux') sf = open(dst, 'w') sf.write('foo')

[issue15748] Various symlink test failures in test_shutil on FreeBSD

New submission from Trent Nelson: FreeBSD slaves are running into this: == ERROR: test_copy2_symlinks (test.test_shutil.TestShutil) -- Traceback (most recent